Provider setup index
Find pages by the provider you are connecting. A page may involve several providers, in which case it appears under each of them.
361 topic pages, distilled from 1728 real issuesHow this is counted
DeepSeek82
CC Switch local proxy failed while handlingUpgrade CC Switch to the latest version and start a new Codex conversation; invalid schema or image blocks left in old session history will keep triggering 400.502 Bad Gateway Codex proxy takeover overwrites config.tomlDisable global proxy and switch to rule mode with direct connection to 127.0.0.1, or upgrade cc-switch to avoid fully overwriting config.toml.auth.json refilled with an existing account's auth.json - CodexUpgrade to v3.17.0+ and turn off 非接管切换时保留官方登录 (keep official login on non-takeover switch). Back up auth.json before upgrading.General config has no effect and gets overwritten by earlier config - Codex CC SwitchUpgrade to CC Switch v3.17.0+ to fix Codex general-config merging. If an old snapshot still overwrites it, switch to another provider and back to self-heal.cc-switch DeepSeek usage statistics inaccurate / balance query failsUpgrade to v3.19.2 and rebuild Codex usage. The balance query needs a correct {{apiKey}}, or a proxy.Codex DeepSeek HTTP 400 errorsUpgrade to v3.18.0 to fix the tool schema 400. Multi-turn tool_call_id and Responses native mode need case-by-case handling.Config lost after switching: Claude Code / Codex / WSL path checkSwitching provider rewrites settings.json. Put hooks/statusLine/enabledPlugins into the shared/general config, not just the local file.config.toml and auth.json still hold the previous third-party APIUpgrade CC Switch to v3.16.1+, restart, and switch Codex provider again. If official OAuth broke, run codex logout then codex login --device-auth.content[].thinking must be passed backUpgrade cc-switch to v3.16.0 or newer. As a stopgap, roll Claude Code back to 2.1.150 or turn thinking off.Codex model catalog template `gpt-5.5`Regenerate ~/.codex/models_cache.json and keep the Codex ChatGPT login, so CC Switch can write the route and show DeepSeek models.Codex model catalog template `gpt-5.5`Run `codex debug models --bundled > ~/.codex/models_cache.json` to generate the missing model cache, then switch the Codex route again.<!-- Failed to upload "image.png" -->Screenshot upload failures do not affect the issue itself; please describe the error message in text.settings.json plugins and hooks lost after switching providerThis is cc-switch's full-snapshot overwrite design. Put the fields you need to keep into the shared / general config and tick write-general-config.Codex desktop shows "Custom" or 401 UnauthorizedEnable "Keep official login when switching to third party" in cc-switch settings and ensure Codex is logged into the official ChatGPT account, or start a new…api failed deserialize json body intoDepending on the error: roll Claude Code back, upgrade cc-switch, turn the local proxy off, or stop sending unsupported fields.Codex provider API keys overwrite each other after a restart or switchEnable CC Switch local proxy takeover, or use experimental_bearer_token in config.toml, to bypass the auth.json sync defect.unexpected status 502 Bad Gateway CodexSet 127.0.0.1 to direct in your proxy software, or configure NO_PROXY in the Codex environment, so the local proxy is not intercepted.Codex desktop not showing third-party modelsLog in with an official ChatGPT account in Codex desktop and enable "preserve official login" in cc-switch to restore display.Codex shows no model list and will not switch modelsTurn on keeping the official login when switching to a third party in cc-switch, and make sure you signed in to official ChatGPT in Codex once.No DEEPSEEK settings in the json config fileUpgrade cc-switch to v3.20.0+, tick the reasoning level in the model mapping and save again, to fix config not being written or overwritten by a template.Connection failedMost related issues remain unresolved; upgrade to v3.16.3+ to fix the SSL renegotiation bug, or restart the application to refresh proxy status.Is codex no longer supported now that it became ChatGPT? - cc-switchUpgrade to cc-switch v3.17.0, which supports provider switching after Codex merged into ChatGPT Desktop, so old config is not overwritten.codex deepseek cannot select a model - cc-switchCodex desktop hides third-party models with no official ChatGPT login. Sign in once, enable keeping the official login in cc-switch, and restart Codex.Invalid schema for function 'codex_app__automation_update'Upgrade cc-switch to v3.18.0+ to fix upstream HTTP 400 errors caused by missing type: object in Codex tool parameters.Only a 200K context window shows in actual use - Claude CodeChange the default model in `~/.claude/settings.json` to the `[1m]` variant, or run `/model claude-opus-4-8[1m]` in the session, to activate the 1M context.404 Not FoundDeepSeek does not support the Responses API — use the built-in preset to route through Chat Completions. For image errors, drop the image modality.model does not support thinkingTurn the Proxy Rectifier off in cc-switch advanced settings, or upgrade to a fixed version. Qwen tool call loops and empty thinking blocks remain unresolved.[SRV-006] connection errorUpgrade cc-switch to v3.16.4, which fixes proxy forwarding disconnects with a custom DeepSeek provider and billing falling back to the request model.temporarily unavailable, so auto mode cannotUpgrade cc-switch to v3.15.0+ for the model mapping and thinking conversion fixes, or override the default model env var in general config.ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYOpen `~/.claude/settings.json`, delete the duplicate auth variable so only the one your provider needs remains, save, and restart Claude Code.The Codex Desktop UI project list is completely empty - CodexCC Switch changed the model_provider key name. Set it back to custom, or upgrade. Lost official login and fewer reasoning tiers remain unresolved.Missing supports_reasoning_summaries field - CodexUpgrade cc-switch to v3.18.0 or newer, which fixes required fields missing from the generated model catalogue and stopping Codex from starting.Error running remote compact taskTurn Remote Compaction off in Codex settings and use local compaction, working around this protocol conversion defect.Ticking 1M for DeepSeek does nothing? cc-switch local proxy configWith the cc-switch local proxy on, do not enter [1m] and leave the main model empty. The [1m] suffix only works with the proxy off.deepseek check failed: Not found (404), the endpoint does not exist - CodexCodex needs the Responses API while DeepSeek only supports the Chat API — use a third-party proxy to convert between the protocols.DeepSeek API key overwritten, causing a 401Upgrade cc-switch to v3.16.2+ and re-enter the DeepSeek API key, so it is not overwritten by another key when switching provider.400 content[].thinking must be passed backUpgrade cc-switch to v3.15.0 or newer to fix lost thinking blocks, or downgrade Claude Code and use the native DeepSeek Anthropic endpoint.DeepSeek usage statistics show the wrong cost / cache hit rateUpgrade to v3.16.3+, fixing input double-counting on the chat→message conversion path, streaming without usage, and wrong model billing attribution.502 Bad GatewayCodex/Claude Desktop reports 502 due to CLI proxy conflicts or app-server bypassing proxy. Remove proxy config or downgrade versions to fix.Invalid schema for function - Codex errorUpgrade cc-switch for the tool schema conversion fix. If upstream rejects oneOf strictly, use a historical task or switch model for now.Request body format invalid error - Claude CodeUpgrade cc-switch and enable the Rectifier, or add MAX_THINKING_TOKENS=0, for compatibility with third-party API formats.unexpected status 404 Not FoundUpgrade cc-switch to v3.16.4+ and set the upstream format explicitly to Chat Completions on the Codex provider form."is not a model this version of Claude Code recognizes" - CC SwitchSet CLAUDE_CODE_MAX_CONTEXT_TOKENS or append [1m] to the model name to define the correct context window size.401 authentication failed: Claude Desktop gateway token invalidClear the official Claude Desktop OAuth login state, or edit .claude.json, to resolve the 401 invalid gateway token.Nothing happens, cannot sign in to the official channel - CodexUpgrade cc-switch to v3.16.2 or newer and delete the .codex and .cc switch folders in your home directory, fixing auth.json being overwritten.codex never shows the ds model - CodexCodex desktop needs an official login state to show third-party models. Enable keep-official-login-when-switching in cc-switch and restart.Console Go: Upstream request failedThis 400 comes from the OpenCode Go gateway rejecting an incompatible payload. Restart the proxy to restore session mapping, or avoid WebSearch and images.It reports 401 in claude - Claude CodeTurn the cc-switch local proxy on to convert the API format, and make sure `ANTHROPIC_BASE_URL` is injected into the Claude Code process.The 'deepseek-v4-flash' modelEnable the keep-official-login option in cc-switch, confirm local route takeover is on, then restart Codex desktop.<tool_use_error>InputValidationErrorAn SSE parsing bug in the cc-switch conversion layer loses tool calls. Connect to the API directly, or switch to an official node, for now.Not logged in · Please run /login ClaudeEdit `~/.claude.json` to move `PROXY_MANAGED` into the `approved` list, or answer yes at the prompt.The proxy takeover claude switch cannot be ticked in Claude Desktop modeIn claude-desktop mode, the takeover section hard-codes the Claude Code (claude) switch instead of the Claude Desktop one, so clicking does nothing.unexpected status 401 Unauthorized CodexUse the built-in DeepSeek template instead of a hand-built config, or add http_headers to the config file by hand to pass the API Key.Configuring DeepSeek in Codex errors about gpt-5.5Install the Codex CLI and run codex debug models --bundled to regenerate models_cache.json.Invalid schema: codex_app__automation_updateUpgrade CC Switch to v3.18.0 or newer to fix the 400 caused by type: null when Codex converts the DeepSeek protocol.The apply_patch tool is unavailable with the Codex proxy onUpgrade to cc-switch v3.16.2 or newer. On a relay, make sure the backend model matches what Codex expects or tool calls fail with routing on.The model cannot discover any tools - CodexUpgrade cc-switch past 3.16.1 and edit cc-switch-model-catalog.json to set supports_search_tool to false for the DeepSeek models.This model is not supported when usingComment out model_provider = custom in config.yaml to go back to the official OpenAI provider.No such file or directory (os error 2)Change the model_catalog_json absolute path in config.toml to the relative cc-switch-model-catalog.json, or upgrade cc-switch.unexpected status 401 Unauthorized - CodexEdit the Codex config.toml by hand to add the model_provider declaration and the matching provider section, or upgrade cc-switch to the latest version.DeepSeek prefix cache hit rate drop - CC Switch fixUpgrade to cc-switch v3.16.3 to fix cache hit rate calculation errors caused by format conversion.Codex conversation history disappears after switching providerAfter upgrading to v3.16.0 or above, CC Switch automatically migrates third-party provider history into the custom bucket; original files are backed up.Test failed - Claude CodeChange the cc-switch outbound proxy port to match your local proxy software (7897, for instance) to restore the connection.Gateway rejected model "claude-haiku-4-5"Turn isFullUrl=true on in the configuration and enter the complete API path, stopping cc-switch from appending /v1/messages automatically.The Hermes Skill sync path does not check the HERMES_HOME environment variableCC Switch hard-codes the Skills path; await a fix supporting a custom HERMES_HOME.HTTP 400 maximum context length CodexUpgrade cc-switch to fix the missing auto_compact_token_limit, or manually set the context window limit in the model catalog json.AbsolutePathBuf deserialized without a base pathOn Codex for macOS, relative paths written by cc-switch cause this error. Change `model_catalog_json` in `~/.codex/config.toml` to an absolute path.Invalid pages parameter: "" Claude CodeSwitch the API format to anthropic, or upgrade cc-switch to fix the empty pages field from Responses conversion.Invalid schema: got 'type: null' - Claude CodeUpgrade cc-switch to v3.17.0 or later to fix the missing type field error during OpenAI format conversion.API Error: 400 Param Incorrect Claude CodeUpgrade CC Switch to v3.16.2+ to fix 400 errors caused by missing reasoning_content in MiMo thinking mode.Proxy billing model not proxied, then error 405 - Claude CodeModify env vars like ANTHROPIC_DEFAULT_SONNET_MODEL to the target model name, as cc-switch proxy only reads this field for mapping.Claude Code Tool Call Failure via NVIDIA Proxy - CC Switch FixNVIDIA free-tier models are unstable; switch to stable providers like deepseek-ai/deepseek-v4-flash.thinking options type cannot be disabledUpgrade cc-switch to v3.16.4 or later; the agent automatically strips the reasoning_effort parameter conflicting with DeepSeek.unexpected status 404 Not FoundDeepSeek does not support the Responses API. Add /v1 to base_url and remove wire_api = "responses".Unexpected status 503 Service Unavailable CodexConfigure v2ray or other proxy tools to bypass LAN traffic, resolving the 503 error when Codex accesses cc-switch local routing.Upstream request failedCheck if the upstream provider supports the current model or has regional restrictions; switch to an available model.Volcengine Agent Plan Auto-Identify Model List Failed 404Manually modify config.toml to match official docs and add models in modelCatalog to bypass 404.Add CoderPlan provider preset for Claude CodeFeature request: add a CoderPlan provider preset to cc-switch. A Claude Code preset PR exists, but merge status is unconfirmed.Low cache hit rate around 40% and very slow responses in Claude CodeSet CLAUDE_CODE_ATTRIBUTION_HEADER to false in .claude/settings.json to stop dynamic headers from breaking third-party model prompt caching.Third-party model still claims based on GPT-5 - CodexMove the `baseInstructions` override out of the `if profile != ProxyChat` block so it applies to ProxyChat. Currently it is not injected even when set.unsupported call: - CodexUpgrade cc-switch to v3.16.4+ to fix tool call name/id overwritten by empty deltas during streaming conversion.Body text appears in thinking blocks, visible text nearly emptyUpgrade cc-switch to a build with PR #4210. Reactive thinking-placeholder handling fixes long-context freezes and body-text collapse.
GLM / Zhipu43
CC Switch local proxy failed while handlingUpgrade CC Switch to the latest version and start a new Codex conversation; invalid schema or image blocks left in old session history will keep triggering 400.502 Bad Gateway Codex proxy takeover overwrites config.tomlDisable global proxy and switch to rule mode with direct connection to 127.0.0.1, or upgrade cc-switch to avoid fully overwriting config.toml.cc-switch DeepSeek usage statistics inaccurate / balance query failsUpgrade to v3.19.2 and rebuild Codex usage. The balance query needs a correct {{apiKey}}, or a proxy.content[].thinking must be passed backUpgrade cc-switch to v3.16.0 or newer. As a stopgap, roll Claude Code back to 2.1.150 or turn thinking off.Codex model catalog template `gpt-5.5`Regenerate ~/.codex/models_cache.json and keep the Codex ChatGPT login, so CC Switch can write the route and show DeepSeek models.settings.json Full Overwrite Causes Plugin FailureUpgrade to v3.20.1+ to fix the bug that clears enabledPlugins when switching configs.api failed deserialize json body intoDepending on the error: roll Claude Code back, upgrade cc-switch, turn the local proxy off, or stop sending unsupported fields.No plan shown (parsed as empty)Upgrade CC Switch to the latest version, or change the filter in your custom script from TOKENS_LIMIT to CREDIT_LIMIT for the new Zhipu credit-based API.Auth conflictLog out of the official Claude login, clear conflicting ANTHROPIC_* variables, set hasCompletedOnboarding to true in ~/.claude.json, and restart the terminal.Codex desktop not showing third-party modelsLog in with an official ChatGPT account in Codex desktop and enable "preserve official login" in cc-switch to restore display.model does not support thinkingTurn the Proxy Rectifier off in cc-switch advanced settings, or upgrade to a fixed version. Qwen tool call loops and empty thinking blocks remain unresolved.Your access token could not be refreshedUpgrade cc-switch to v3.17.0+, which fixes the PROXY_MANAGED placeholder polluting OAuth credentials or causing a 401 during takeover.ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYOpen `~/.claude/settings.json`, delete the duplicate auth variable so only the one your provider needs remains, save, and restart Claude Code.Ticking 1M for DeepSeek does nothing? cc-switch local proxy configWith the cc-switch local proxy on, do not enter [1m] and leave the main model empty. The [1m] suffix only works with the proxy off.Failed to fetch models - cc-switchUpgrade cc-switch to v3.16.4+ for the 404 probe fix. On a Coding Plan, change the API address by hand and enter the model names yourself.The interface shows "query failed"Upgrade past v3.16.2 for the query routing fix, or use a custom balance query template. Put hooks in general config, and raise the failover timeout.invalid type: sequenceBack up and delete the old config.toml so Codex regenerates a default config, fixing the incompatible hooks format.MCP config files edited by hand are overwritten by cc-switchcc-switch writes wholesale. Sync plugins through general config, and upgrade to v3.13.0 for the MCP cross-app leak fix.Request body format invalid error - Claude CodeUpgrade cc-switch and enable the Rectifier, or add MAX_THINKING_TOKENS=0, for compatibility with third-party API formats."is not a model this version of Claude Code recognizes" - CC SwitchSet CLAUDE_CODE_MAX_CONTEXT_TOKENS or append [1m] to the model name to define the correct context window size.built-in provider IDs: openai - CodexUpgrade CC Switch to v3.16.4+, delete the old config and save again. Enable unified Codex session history and use openai_base_url.codex never shows the ds model - CodexCodex desktop needs an official login state to show third-party models. Enable keep-official-login-when-switching in cc-switch and restart.Console Go: Upstream request failedThis 400 comes from the OpenCode Go gateway rejecting an incompatible payload. Restart the proxy to restore session mapping, or avoid WebSearch and images.Calling GLM-5 reports error 1311 - cc-switchA provider permissions problem: enable GLM-5 / MiniMax / Kimi access with your provider, or ask them to confirm your subscription entitlements.ending with a leaked `</t` close-tagCaused by the cc-switch streaming converter not deduplicating and not filtering think tags. Update to a version containing the PR #6319 fix.Not found (404)Upgrade cc-switch to v3.16.3 for the 404 from /v1/ appended wrongly to a Base URL, and enable the media fallback so images cannot trip the breaker.exceeded retry limit, last status: 429Limit client concurrency and retry frequency, and fix the cc-switch bug where hot switching disables automatic failover, with SQL.Range of input length should be [1, 1048576]Upgrade CC Switch to v3.19.0 for the fix where images were forwarded as base64 text and inflated tokens, or avoid view_image under the proxy.401 Unauthorized: Incorrect API key providedAfter switching to OpenAI Official or turning the proxy off, clear the leftover third-party OPENAI_API_KEY from ~/.codex/auth.json and restart Codex.ModelScope reports 401 token invalid - Claude CodeAdd ANTHROPIC_MODEL to the provider environment variables in cc-switch to resolve the 401. It is not synced to settings.json by design.[proxy::response_processor] [Claude] stream errorA stream decode error or timeout proxying Claude Code loses logs or records 0 tokens. Fix the model extraction fallback to show the right model.The model cannot discover any tools - CodexUpgrade cc-switch past 3.16.1 and edit cc-switch-model-catalog.json to set supports_search_tool to false for the DeepSeek models.DeepSeek prefix cache hit rate drop - CC Switch fixUpgrade to cc-switch v3.16.3 to fix cache hit rate calculation errors caused by format conversion.API Error: 422 format conversion errorThe upstream NVIDIA NIM GLM-5.2 model intermittently returns an empty choices array. Turn on failover in cc-switch or retry the request.Model always thinks; disable not supported; use low, high, or maxSet glm-5.3 to forced thinking in cc-switch config; disable thinking.type: disabled.HTTP 400 maximum context length CodexUpgrade cc-switch to fix the missing auto_compact_token_limit, or manually set the context window limit in the model catalog json.Invalid schema: got 'type: null' - Claude CodeUpgrade cc-switch to v3.17.0 or later to fix the missing type field error during OpenAI format conversion.Claude Code Tool Call Failure via NVIDIA Proxy - CC Switch FixNVIDIA free-tier models are unstable; switch to stable providers like deepseek-ai/deepseek-v4-flash.Reasoning Level Not AdaptedRe-add the OpenCode Go preset after upgrading to get the correct reasoning_effort config.Volcengine Agent Plan Auto-Identify Model List Failed 404Manually modify config.toml to match official docs and add models in modelCatalog to bypass 404.Test returns 504: Claude Code unusableCheck whether cc-switch loaded the `glm-4.7` model mapping into the proxy; otherwise Claude Code requests glm directly and returns 504.CC Switch online update freezes and becomes unresponsiveCC Switch online update can freeze under FlClash TUN mode. Switch to system proxy mode or manually download and reinstall.Claude Code cannot be applied in VS CodeRemove old Claude Code proxy settings in VS Code, then toggle Apply to Claude Code plugin in cc-switch again.
OpenAI (official)43
auth.json refilled with an existing account's auth.json - CodexUpgrade to v3.17.0+ and turn off 非接管切换时保留官方登录 (keep official login on non-takeover switch). Back up auth.json before upgrading.General config has no effect and gets overwritten by earlier config - Codex CC SwitchUpgrade to CC Switch v3.17.0+ to fix Codex general-config merging. If an old snapshot still overwrites it, switch to another provider and back to self-heal.settings.json overwritten by proxy takeover causing plugin config loss - Claude CodePlace custom fields such as hooks and statusLine into the "General Configuration" panel, or upgrade to v3.12.2+ to fix the legacy backup restore logic.config.toml and auth.json still hold the previous third-party APIUpgrade CC Switch to v3.16.1+, restart, and switch Codex provider again. If official OAuth broke, run codex logout then codex login --device-auth.Codex desktop shows "Custom" or 401 UnauthorizedEnable "Keep official login when switching to third party" in cc-switch settings and ensure Codex is logged into the official ChatGPT account, or start a new…Cannot query usage for an official subscription in the Codex UI - CodexEnable local route mapping so specific models are counted. Third-party APIs need their own query script; Claude official and Free accounts are unsupported.Codex provider API keys overwrite each other after a restart or switchEnable CC Switch local proxy takeover, or use experimental_bearer_token in config.toml, to bypass the auth.json sync defect.Cannot sign in to a Codex account that uses a Google email - CodexUpgrade cc-switch to v3.16.1+, or edit ~/.codex/auth.json by hand to restore auth_mode to chatgpt and clear OPENAI_API_KEY.Codex desktop not showing third-party modelsLog in with an official ChatGPT account in Codex desktop and enable "preserve official login" in cc-switch to restore display.Codex shows no model list and will not switch modelsTurn on keeping the official login when switching to a third party in cc-switch, and make sure you signed in to official ChatGPT in Codex once.auth.json under .claude is forcibly replaced after a restart - CodexUpgrade cc-switch to the latest version, or turn proxy takeover off before signing in officially, so a stale snapshot cannot overwrite official credentials.ChatGPT will not authenticate even with the global proxy on - cc-switchUpgrade cc-switch to v3.16.4+ and set the proxy address under Settings > Routing > Global outbound proxy, fixing OAuth not going through the proxy.codex session expired - CodexBack up the config, then sync the auth.json token into the cc-switch database with a script, or turn usage refresh off so old data cannot overwrite it."Not logged in · Please run /login" - Claude CodeUpgrade cc-switch to v3.19.2+, or change ANTHROPIC_API_KEY to ANTHROPIC_AUTH_TOKEN in settings.json by hand to skip the approval prompt.404 Not FoundDeepSeek does not support the Responses API — use the built-in preset to route through Chat Completions. For image errors, drop the image modality.Your access token could not be refreshedUpgrade cc-switch to v3.17.0+, which fixes the PROXY_MANAGED placeholder polluting OAuth credentials or causing a 401 during takeover.DeepSeek API key overwritten, causing a 401Upgrade cc-switch to v3.16.2+ and re-enter the DeepSeek API key, so it is not overwritten by another key when switching provider.invalid type: sequenceBack up and delete the old config.toml so Codex regenerates a default config, fixing the incompatible hooks format.Unknown parameterTurn the cc-switch unified session history switch off and move sessions tagged custom back to openai, fixing the namespace parameter error.Unsupported content type: redacted_thinkingRun a local proxy such as openrouter-proxy to filter out redacted_thinking blocks, and point the Claude Code config at that proxy.Nothing happens, cannot sign in to the official channel - CodexUpgrade cc-switch to v3.16.2 or newer and delete the .codex and .cc switch folders in your home directory, fixing auth.json being overwritten.built-in provider IDs: openai - CodexUpgrade CC Switch to v3.16.4+, delete the old config and save again. Enable unified Codex session history and use openai_base_url.codex never shows the ds model - CodexCodex desktop needs an official login state to show third-party models. Enable keep-official-login-when-switching in cc-switch and restart.The 'deepseek-v4-flash' modelEnable the keep-official-login option in cc-switch, confirm local route takeover is on, then restart Codex desktop.Error running remote compact task - CodexUpgrade cc-switch to v3.12.0 or newer, which added forwarding for the Codex `/responses/compact` route in the local proxy, fixing the 404 or 502.Gtk-CRITICAL: gtk_widget_get_scale_factor failedOn a GTK assertion failure running cc-switch under Linux or WSL2, edit settings.json to turn silent startup off, or clean the leftovers and reinstall.<tool_use_error>InputValidationErrorAn SSE parsing bug in the cc-switch conversion layer loses tool calls. Connect to the API directly, or switch to an official node, for now.401 Unauthorized: Incorrect API key providedAfter switching to OpenAI Official or turning the proxy off, clear the leftover third-party OPENAI_API_KEY from ~/.codex/auth.json and restart Codex.HTTP 401 Unauthorized - CC Switch AnthropicUpgrade cc-switch to a fixed build so it sends the x-api-key and anthropic-version headers Anthropic requires.unexpected status 401 Unauthorized CodexUse the built-in DeepSeek template instead of a hand-built config, or add http_headers to the config file by hand to pass the API Key.[mcp_servers.node_repl] written back automatically by CC SwitchTurn off or delete node_repl in MCP management at the top right of CC Switch to stop it being written back into the Codex config file.Codex still asks for a login when openedDelete the Codex default configuration or the whole .codex folder to skip the login check.HTTP 400 Invalid JSON /responsesChange meta.apiFormat of the Codex provider to openai_responses so requests are forwarded straight through.stop: cancel task, id_task = 2 Codex timeout overwriteIncrease startup_timeout_sec, and save the MCP configuration once more after it is overwritten a second time.Codex subagent token statistics are missing and too lowUpgrade to the latest code, where the parser derives a unique request ID from payload.id instead of session_id, so subagent usage lands in the database.unexpected status 401 Unauthorized - CodexEdit the Codex config.toml by hand to add the model_provider declaration and the matching provider section, or upgrade cc-switch to the latest version.expected value at line 1 column 1Upgrade to cc-switch v3.14.1 or newer to fix the JSON parse failure caused by Codex OAuth forcing an SSE stream.AbsolutePathBuf deserialized without a base pathOn Codex for macOS, relative paths written by cc-switch cause this error. Change `model_catalog_json` in `~/.codex/config.toml` to an absolute path.model is not supported when using CodexDisable cc-switch proxy takeover to avoid writing unsupported model IDs to Codex, preventing official subscription bans or request rejections."Not logged in ? Please run /login" ClaudeEnable "API Proxy Takeover" in cc-switch when using hosted accounts like Copilot or Codex; direct upstream connections cause auth failures.Atomic replace failed os error 50Change the Claude Code configuration directory from a WSL path back to a Windows local path, or roll back to v3.19.0.Codex cannot select models after configurationEnable keep-official-login in Codex app enhancements, then log in again and switch providers to restore the model list.Codex skill toggle not working: still loads legacy directory ~/.agents/skillsRemove leftover copies in ~/.agents/skills or disable them in config; CC-Switch cannot manage that shared directory.
Gemini / Google41
Codex config.toml configuration lost after switching providerUpgrade cc-switch to v3.17.0+, which changes to incremental merge instead of full overwrite of config.toml.settings.json plugins and hooks lost after switching providerThis is cc-switch's full-snapshot overwrite design. Put the fields you need to keep into the shared / general config and tick write-general-config.Cannot sign in to a Codex account that uses a Google email - CodexUpgrade cc-switch to v3.16.1+, or edit ~/.codex/auth.json by hand to restore auth_mode to chatgpt and clear OPENAI_API_KEY.EGL_BAD_PARAMETER white screen error - cc-switchInstall the native .deb or AUR package instead, or run the AppImage with LD_PRELOAD pointing at the system library, to resolve the ABI conflict.Failed to parse Hermes configUpgrade cc-switch to v3.16.3+ to fix the CRLF-driven config append bug, or delete the duplicate keys in config.yaml by hand.Unknown name "prompt_cache_key" - cc-switchUpgrade cc-switch to v3.16.2+ to fix prompt_cache_key injection, or use an OpenRouter endpoint that natively speaks the Claude Code format.Function call is missing a thought_signatureUpgrade or patch the cc-switch forwarder/session logic so Codex tool call requests carry a thought_signature.Test script: error, no result returnedMiniMax changed the balance query API response shape, breaking the old parser. Upgrade cc-switch to v3.16.2+, or use a custom template for now.Warmup request downgraded to modelTurn the Copilot warmupDowngrade option off in CC Switch so requests are not silently replaced with gpt-5-mini.ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYOpen `~/.claude/settings.json`, delete the duplicate auth variable so only the one your provider needs remains, save, and restart Claude Code.Not installed / not installed or not executable - CC SwitchUpgrade CC Switch to v3.20.0+ for custom npm prefix and macOS detection failures; the false "several installations" on Windows needs v3.16.5+.not installed or not executableUpgrade CC Switch to v3.16.2+, or set the CLI install directory by hand in settings, and clear noise printed by your shell at startup.Gemini CLI session history will not display - Gemini CLIGemini CLI 0.39.1 changed the history path. With the consumer service being retired, the old path will not be fixed — wait for Antigravity CLI support.A default Gemini config appears and cannot be deleted - CC Switchdefault is the expected provider CC Switch creates by importing your existing live config. Switch the current provider in settings first, then delete it.Gemini CLI usage data is not counted - CC SwitchThe consumer Gemini CLI is retired in favour of Antigravity CLI, so CC Switch no longer supports its usage statistics. For template or Grok problems, upgrade.JSON parse error: expected value at line 1Upgrade cc-switch to fix JSON corruption from concurrent writes, and add custom fields to general config so a switch cannot overwrite them.MCP config files edited by hand are overwritten by cc-switchcc-switch writes wholesale. Sync plugins through general config, and upgrade to v3.13.0 for the MCP cross-app leak fix.selected model (kimi-k2.5) may not exist errorDrop the trailing /v1 from the API address and use the Anthropic-compatible endpoint. For an old session with images, /compact or switch model.skills installed to the wrong path - cc-switchOlder cc-switch installs skills into ~/.claude/skills/skills, which Claude Code ignores. Upgrade, or move the contents up to ~/.claude/skills/.Claude code does not support gemini - CC Switchcc-switch cannot currently use Gemini as a Claude Code provider — wait for the local proxy feature, or use a third-party proxy project.Restarting the terminal and running gemini asks for an api-key - Gemini CLIUpdate cc-switch for the save-clears-config bug. If it still fails, that is a Gemini CLI read bug — put the environment variables in bash_profile.Function call is missingUse the community cc-switch-gemini-fixer patch, or inject skip_thought_signature_validator into history messages, to bypass the signature validation.Gtk-CRITICAL: gtk_widget_get_scale_factor failedOn a GTK assertion failure running cc-switch under Linux or WSL2, edit settings.json to turn silent startup off, or clean the leftovers and reinstall.claude cli exists locally but the about settings say not installed - CC SwitchUpgrade CC Switch to v3.16.5 or newer, fixing the false conflict from multiple npm global package entry points on Windows and the garbled paths.HTTP 401 Unauthorized - CC Switch AnthropicUpgrade cc-switch to a fixed build so it sends the x-api-key and anthropic-version headers Anthropic requires.codex cli failed - CC SwitchUpgrading cc-switch to v3.16.3 or newer fixes the local environment detection and the blocked upgrade path.The proxy takeover claude switch cannot be ticked in Claude Desktop modeIn claude-desktop mode, the takeover section hard-codes the Claude Code (claude) switch instead of the Claude Desktop one, so clicking does nothing.The model cannot discover any tools - CodexUpgrade cc-switch past 3.16.1 and edit cc-switch-model-catalog.json to set supports_search_tool to false for the DeepSeek models.Function call is missing a thought_signatureTriggering a tool call with Gemini Native in Claude Desktop proxy mode loses the thought_signature to a session ID mismatch. Await the official fix.The Gemini CLI reading the project .env breaks cc-switchWrap the gemini command in PowerShell to load the ~/.gemini/.env environment variables by hand.cc-switch keeps saying there is an environment variable conflict and then deletes the variable Gemini CLIIgnore the environment variable conflict prompt to keep your system variables, or upgrade cc-switch to v3.12.3 to fix saving in Official mode."Not logged in ? Please run /login" ClaudeEnable "API Proxy Takeover" in cc-switch when using hosted accounts like Copilot or Codex; direct upstream connections cause auth failures.http://127.0.0.1:15721 proxy configuration stuckUpgrade to CC Switch v3.16.2 to fix proxy hangs from corrupted backups, or use "Replace Config" mode to manually set env vars.Upstream request failedCheck if the upstream provider supports the current model or has regional restrictions; switch to an available model."event handler is re-entrant" error or startup crashUninstall Windows update KB5081263 or wait for cc-switch upgrade to avoid crashes and UI rendering failures caused by TAO EventLoop re-entry.cc-switch crashes immediately after openingNo confirmed fix. Reinstalling does not help. Collect debug logs and wait for an official fix.CC Switch online update freezes and becomes unresponsiveCC Switch online update can freeze under FlClash TUN mode. Switch to system proxy mode or manually download and reinstall.Atomic replace failed os error 50Change the Claude Code configuration directory from a WSL path back to a Windows local path, or roll back to v3.19.0.Gemini local proxy says invalid token cc-switchMove the Gemini provider channel from codex back to gemini. With the local proxy enabled, this avoids wrong routing and invalid token errors.show 0 skills from cc-switchUpgrade cc-switch to v3.17.0 and re-toggle the skill state to fix Windows Hermes Desktop reading the LOCALAPPDATA path.Installed but cannot run in current environment, please check CC SwitchUpgrade to CC Switch v3.16.1+ to fix Windows `cmd /C` quote handling that misreports installed CLI as broken.
Kimi / Moonshot40
CC Switch local proxy failed while handlingUpgrade CC Switch to the latest version and start a new Codex conversation; invalid schema or image blocks left in old session history will keep triggering 400.502 Bad Gateway Codex proxy takeover overwrites config.tomlDisable global proxy and switch to rule mode with direct connection to 127.0.0.1, or upgrade cc-switch to avoid fully overwriting config.toml.General config has no effect and gets overwritten by earlier config - Codex CC SwitchUpgrade to CC Switch v3.17.0+ to fix Codex general-config merging. If an old snapshot still overwrites it, switch to another provider and back to self-heal.Config lost after switching: Claude Code / Codex / WSL path checkSwitching provider rewrites settings.json. Put hooks/statusLine/enabledPlugins into the shared/general config, not just the local file.content[].thinking must be passed backUpgrade cc-switch to v3.16.0 or newer. As a stopgap, roll Claude Code back to 2.1.150 or turn thinking off.<!-- Failed to upload "image.png" -->Screenshot upload failures do not affect the issue itself; please describe the error message in text.settings.json plugins and hooks lost after switching providerThis is cc-switch's full-snapshot overwrite design. Put the fields you need to keep into the shared / general config and tick write-general-config.API Error: 401 authentication_errorCheck the API key field, Base URL, key type, and cc-switch version, so Claude Code actually reads a valid credential.Codex connecting to Kimi reports HTTP 400 tool schema validation failureUpgrade cc-switch to v3.18.0+; if the error persists, wait for the MFJS schema cleaning feature to go live.No DEEPSEEK settings in the json config fileUpgrade cc-switch to v3.20.0+, tick the reasoning level in the model mapping and save again, to fix config not being written or overwritten by a template.Is codex no longer supported now that it became ChatGPT? - cc-switchUpgrade to cc-switch v3.17.0, which supports provider switching after Codex merged into ChatGPT Desktop, so old config is not overwritten.Invalid schema for function 'codex_app__automation_update'Upgrade cc-switch to v3.18.0+ to fix upstream HTTP 400 errors caused by missing type: object in Codex tool parameters.404 Not FoundDeepSeek does not support the Responses API — use the built-in preset to route through Chat Completions. For image errors, drop the image modality.The Codex Desktop UI project list is completely empty - CodexCC Switch changed the model_provider key name. Set it back to custom, or upgrade. Lost official login and fewer reasoning tiers remain unresolved.Error running remote compact taskTurn Remote Compaction off in Codex settings and use local compaction, working around this protocol conversion defect.400 content[].thinking must be passed backUpgrade cc-switch to v3.15.0 or newer to fix lost thinking blocks, or downgrade Claude Code and use the native DeepSeek Anthropic endpoint.kimi k2 reports 404 - CC SwitchA 404 from the connection test does not mean Kimi is unavailable. Set ANTHROPIC_BASE_URL to https://api.kimi.com/coding/ and verify by running Claude Code.selected model (kimi-k2.5) may not exist errorDrop the trailing /v1 from the API address and use the Anthropic-compatible endpoint. For an old session with images, /compact or switch model."is not a model this version of Claude Code recognizes" - CC SwitchSet CLAUDE_CODE_MAX_CONTEXT_TOKENS or append [1m] to the model name to define the correct context window size.401 authentication failed: Claude Desktop gateway token invalidClear the official Claude Desktop OAuth login state, or edit .claude.json, to resolve the 401 invalid gateway token.Calling GLM-5 reports error 1311 - cc-switchA provider permissions problem: enable GLM-5 / MiniMax / Kimi access with your provider, or ask them to confirm your subscription entitlements.ending with a leaked `</t` close-tagCaused by the cc-switch streaming converter not deduplicating and not filtering think tags. Update to a version containing the PR #6319 fix.Unsupported parameter(s): prompt_cache_keyUpgrade cc-switch to the latest version, fixing prompt_cache_key not being stripped automatically when forwarding to the Nvidia API.HTTP 401 exceeding the 256K context limit - CC Switch & CodexLower the AUTO COMPACT trigger threshold in the Codex config.toml so the real token count of the request body stays under the Kimi 256K limit.Codex kimi keeps sending requests when using the proxyThis is Codex App client behaviour, not a cc-switch fault. Check the client retry or keepalive mechanism.Upstream HTTP 404 cc-switchUpgrade cc-switch to remove ?beta=true from NVIDIA requests. Kimi does not support Codex's Responses API; use Claude Code or a local adapter.invalid_authentication_error test failureUpgrade cc-switch to fix failed {{apiKey}} parsing in custom usage query scripts, or temporarily hard-code the real API Key.Invalid schema: got 'type: null' - Claude CodeUpgrade cc-switch to v3.17.0 or later to fix the missing type field error during OpenAI format conversion.The user sent an empty message.Set Kimi API format in cc-switch to OpenAI chat completions and enable proxy, or export history and rebuild the session.invalid_request_error Kimi For CodingSet ENABLE_TOOL_SEARCH=false in the Kimi configuration in cc-switch.Kimi balance query unavailableUpdate cc-switch to the latest version. The backend has automatically fixed the Kimi balance recognition logic.API Error: 400 Param Incorrect Claude CodeUpgrade CC Switch to v3.16.2+ to fix 400 errors caused by missing reasoning_content in MiMo thinking mode.Nvidia check failed: Not found (404)Upgrade cc-switch to v3.16.3+; new version uses lightweight HTTP probing, fixing NVIDIA interface false 404 errors.thinking enabled but reasoning_content missingDisable Thinking mode in Claude Code, or upgrade cc-switch to fix Kimi/Moonshot tool call errors.Volcengine Agent Plan Auto-Identify Model List Failed 404Manually modify config.toml to match official docs and add models in modelCatalog to bypass 404.Add CoderPlan provider preset for Claude CodeFeature request: add a CoderPlan provider preset to cc-switch. A Claude Code preset PR exists, but merge status is unconfirmed.Bash(...) Waiting... Claude CodeDowngrade Claude Code to v2.1.104 or v2.0.64 (admin required), or switch to a compatible provider such as Alibaba Cloud Bailian.Third-party model still claims based on GPT-5 - CodexMove the `baseInstructions` override out of the `if profile != ProxyChat` block so it applies to ProxyChat. Currently it is not injected even when set.unsupported call: - CodexUpgrade cc-switch to v3.16.4+ to fix tool call name/id overwritten by empty deltas during streaming conversion.Body text appears in thinking blocks, visible text nearly emptyUpgrade cc-switch to a build with PR #4210. Reactive thinking-placeholder handling fixes long-context freezes and body-text collapse.
MiniMax36
Codex config.toml configuration lost after switching providerUpgrade cc-switch to v3.17.0+, which changes to incremental merge instead of full overwrite of config.toml.cc-switch DeepSeek usage statistics inaccurate / balance query failsUpgrade to v3.19.2 and rebuild Codex usage. The balance query needs a correct {{apiKey}}, or a proxy.Config lost after switching: Claude Code / Codex / WSL path checkSwitching provider rewrites settings.json. Put hooks/statusLine/enabledPlugins into the shared/general config, not just the local file.settings.json Full Overwrite Causes Plugin FailureUpgrade to v3.20.1+ to fix the bug that clears enabledPlugins when switching configs.Codex desktop shows "Custom" or 401 UnauthorizedEnable "Keep official login when switching to third party" in cc-switch settings and ensure Codex is logged into the official ChatGPT account, or start a new…API Error: 401 authentication_errorCheck the API key field, Base URL, key type, and cc-switch version, so Claude Code actually reads a valid credential.No plan shown (parsed as empty)Upgrade CC Switch to the latest version, or change the filter in your custom script from TOKENS_LIMIT to CREDIT_LIMIT for the new Zhipu credit-based API.Connection failedMost related issues remain unresolved; upgrade to v3.16.3+ to fix the SSL renegotiation bug, or restart the application to refresh proxy status.Qwen model errors or truncation - CC SwitchUpgrade to CC Switch v3.16.4 and use the matching new Qwen / OpenCode Go presets.Is codex no longer supported now that it became ChatGPT? - cc-switchUpgrade to cc-switch v3.17.0, which supports provider switching after Codex merged into ChatGPT Desktop, so old config is not overwritten.codex deepseek cannot select a model - cc-switchCodex desktop hides third-party models with no official ChatGPT login. Sign in once, enable keeping the official login in cc-switch, and restart Codex.Invalid schema for function 'codex_app__automation_update'Upgrade cc-switch to v3.18.0+ to fix upstream HTTP 400 errors caused by missing type: object in Codex tool parameters.Test script: error, no result returnedMiniMax changed the balance query API response shape, breaking the old parser. Upgrade cc-switch to v3.16.2+, or use a custom template for now.temporarily unavailable, so auto mode cannotUpgrade cc-switch to v3.15.0+ for the model mapping and thinking conversion fixes, or override the default model env var in general config.Error running remote compact taskTurn Remote Compaction off in Codex settings and use local compaction, working around this protocol conversion defect.DeepSeek usage statistics show the wrong cost / cache hit rateUpgrade to v3.16.3+, fixing input double-counting on the chat→message conversion path, streaming without usage, and wrong model billing attribution.Failed to fetch models - cc-switchUpgrade cc-switch to v3.16.4+ for the 404 probe fix. On a Coding Plan, change the API address by hand and enter the model names yourself.Invalid type for 'messages'Upgrade cc-switch to the latest version for the conversion fix, or downgrade the Codex CLI to 0.80.0 and use wire_api="chat" to work around it.unexpected status 404 Not FoundUpgrade cc-switch to v3.16.4+ and set the upstream format explicitly to Chat Completions on the Codex provider form.AuthError Missing API key - cc-switch auth header fixThe upstream Anthropic-compatible endpoint only reads x-api-key, so set the auth field to ANTHROPIC_API_KEY rather than ANTHROPIC_AUTH_TOKEN.401 authentication failed: Claude Desktop gateway token invalidClear the official Claude Desktop OAuth login state, or edit .claude.json, to resolve the 401 invalid gateway token.Only 258k context in codex - CodexOnce cc-switch maps a minimax model onto gpt-5.5, the hard-coded Codex and Claude Code context limits apply and the 1M context is lost.Calling GLM-5 reports error 1311 - cc-switchA provider permissions problem: enable GLM-5 / MiniMax / Kimi access with your provider, or ask them to confirm your subscription entitlements.ending with a leaked `</t` close-tagCaused by the cc-switch streaming converter not deduplicating and not filtering think tags. Update to a version containing the PR #6319 fix.HTTP 400 unknown model 'ark-code-latest'Change the MiniMax provider model ID in CC Switch from the Volcengine-only ark-code-latest to MiniMax-M3 to resolve this 400.exceeded retry limit, last status: 429Limit client concurrency and retry frequency, and fix the cc-switch bug where hot switching disables automatic failover, with SQL.responses_feature_not_supported: web_searchUpgrade cc-switch to v3.16.5, which disables web_search in the config automatically to work around the Xiaomi MiMo gateway limitation.status=429 - Claude DesktopStreaming requests in CC Switch 3.15.0 skip the model rewrite and produce a 429. There is no official fix yet.Why is the model the claude command uses in a plain terminal different from the one in cc-switch?Set the MiniMax you want as the primary model and run claude from a cc-switch terminal. Without takeover on, the mapped name is not a substitution.Codex conversation history disappears after switching providerAfter upgrading to v3.16.0 or above, CC Switch automatically migrates third-party provider history into the custom bucket; original files are backed up.The user sent an empty message.Set Kimi API format in cc-switch to OpenAI chat completions and enable proxy, or export history and rebuild the session.Minimax configuration overwrittenUpdate cc-switch to the fixed version to resolve the issue where Minimax configuration is incorrectly overwritten during switching.Add CoderPlan provider preset for Claude CodeFeature request: add a CoderPlan provider preset to cc-switch. A Claude Code preset PR exists, but merge status is unconfirmed.Low cache hit rate around 40% and very slow responses in Claude CodeSet CLAUDE_CODE_ATTRIBUTION_HEADER to false in .claude/settings.json to stop dynamic headers from breaking third-party model prompt caching.unsupported call: - CodexUpgrade cc-switch to v3.16.4+ to fix tool call name/id overwritten by empty deltas during streaming conversion.There's an issue with the selected modelEnable local proxy and rectifier, and correctly set ANTHROPIC_BASE_URL to resolve NVIDIA or MiniMax model call failures.
Third-party relay29
Codex config.toml configuration lost after switching providerUpgrade cc-switch to v3.17.0+, which changes to incremental merge instead of full overwrite of config.toml.settings.json overwritten by proxy takeover causing plugin config loss - Claude CodePlace custom fields such as hooks and statusLine into the "General Configuration" panel, or upgrade to v3.12.2+ to fix the legacy backup restore logic.Codex model catalog template `gpt-5.5`Run `codex debug models --bundled > ~/.codex/models_cache.json` to generate the missing model cache, then switch the Codex route again.<!-- Failed to upload "image.png" -->Screenshot upload failures do not affect the issue itself; please describe the error message in text.Cannot sign in to a Codex account that uses a Google email - CodexUpgrade cc-switch to v3.16.1+, or edit ~/.codex/auth.json by hand to restore auth_mode to chatgpt and clear OPENAI_API_KEY.No plan shown (parsed as empty)Upgrade CC Switch to the latest version, or change the filter in your custom script from TOKENS_LIMIT to CREDIT_LIMIT for the new Zhipu credit-based API.missing information - cc-switchA roundup of documentation gaps. Each case needs its own clarification, or waiting for the official documentation to fill in.Codex shows no model list and will not switch modelsTurn on keeping the official login when switching to a third party in cc-switch, and make sure you signed in to official ChatGPT in Codex once.No DEEPSEEK settings in the json config fileUpgrade cc-switch to v3.20.0+, tick the reasoning level in the model mapping and save again, to fix config not being written or overwritten by a template.Connection failedMost related issues remain unresolved; upgrade to v3.16.3+ to fix the SSL renegotiation bug, or restart the application to refresh proxy status.Only a 200K context window shows in actual use - Claude CodeChange the default model in `~/.claude/settings.json` to the `[1m]` variant, or run `/model claude-opus-4-8[1m]` in the session, to activate the 1M context.Function call is missing a thought_signatureUpgrade or patch the cc-switch forwarder/session logic so Codex tool call requests carry a thought_signature.model does not support thinkingTurn the Proxy Rectifier off in cc-switch advanced settings, or upgrade to a fixed version. Qwen tool call loops and empty thinking blocks remain unresolved.I selected proxy forwarding to a remote host but it still forwards locally - CC SwitchDelete the .cc-switch config directory to clear the cache, and upgrade to v3.17.0 for the relay restriction and statistics fixes.Claude Official check failed: could not extract base_url - CodexCC Switch proxying and load balancing do not support an official Claude subscription — use a third-party API key. Forcing it errors and risks a ban.The Codex Desktop UI project list is completely empty - CodexCC Switch changed the model_provider key name. Set it back to custom, or upgrade. Lost official login and fewer reasoning tiers remain unresolved.Switching with cc-switch has no effect - CodexCodex does not hot-switch. Restart the Codex app completely after switching relay in cc-switch; a migration tool can recover lost sessions.Ticking 1M for DeepSeek does nothing? cc-switch local proxy configWith the cc-switch local proxy on, do not enter [1m] and leave the main model empty. The [1m] suffix only works with the proxy off.Usage query must be a same-origin request - cc-switchSwitch to the generic or custom template in cc-switch and put the usage query URL straight into the request address, bypassing the same-origin restriction.Error running remote compact task - CodexUpgrade cc-switch to v3.12.0 or newer, which added forwarding for the Codex `/responses/compact` route in the local proxy, fixing the 404 or 502.The apply_patch tool is unavailable with the Codex proxy onUpgrade to cc-switch v3.16.2 or newer. On a relay, make sure the backend model matches what Codex expects or tool calls fail with routing on.stop: cancel task, id_task = 2 Codex timeout overwriteIncrease startup_timeout_sec, and save the MCP configuration once more after it is overwritten a second time.Your input exceeds the context window errorThe session context is too long for the model limit. Compact the context by hand or set an auto compact window.ERR_BAD_REQUEST api.anthropic.com ClaudeTurn on "skip the first-install confirmation" in cc-switch and clear the conflicting environment variables to resolve the Claude Code ERR_BAD_REQUEST.Minimax configuration overwrittenUpdate cc-switch to the fixed version to resolve the issue where Minimax configuration is incorrectly overwritten during switching.Configuration UI FlickeringUpgrade cc-switch to fix flickering from async form loading; read SQLite DB to fix env vars not updating with providers.unexpected status 403 ForbiddenDisable 'Keep Official Login' in CC Switch and restart to stop Codex from sending image gen requests causing 403 errors.auth.json does not update when switching channels; editing overwrites third-party Codex keyNo official fix. Editing a saved third-party Codex channel makes the official KEY in auth.json overwrite the third-party key. Delete and recreate the channel.Image generation is not enabled for this group - cc-switch 403 ErrorFilter image-generation requests manually or use a local bridge to read response headers. Official support is pending to resolve the 403 block.
GitHub Copilot24
settings.json overwritten by proxy takeover causing plugin config loss - Claude CodePlace custom fields such as hooks and statusLine into the "General Configuration" panel, or upgrade to v3.12.2+ to fix the legacy backup restore logic.settings.json Full Overwrite Causes Plugin FailureUpgrade to v3.20.1+ to fix the bug that clears enabledPlugins when switching configs.EGL_BAD_PARAMETER white screen error - cc-switchInstall the native .deb or AUR package instead, or run the AppImage with LD_PRELOAD pointing at the system library, to resolve the ABI conflict."Not logged in · Please run /login" - Claude CodeUpgrade cc-switch to v3.19.2+, or change ANTHROPIC_API_KEY to ANTHROPIC_AUTH_TOKEN in settings.json by hand to skip the approval prompt.Test script: error, no result returnedMiniMax changed the balance query API response shape, breaking the old parser. Upgrade cc-switch to v3.16.2+, or use a custom template for now.Your access token could not be refreshedUpgrade cc-switch to v3.17.0+, which fixes the PROXY_MANAGED placeholder polluting OAuth credentials or causing a 401 during takeover.temporarily unavailable, so auto mode cannotUpgrade cc-switch to v3.15.0+ for the model mapping and thinking conversion fixes, or override the default model env var in general config.Warmup request downgraded to modelTurn the Copilot warmupDowngrade option off in CC Switch so requests are not silently replaced with gpt-5-mini.Failed to fetch models - cc-switchUpgrade cc-switch to v3.16.4+ for the 404 probe fix. On a Coding Plan, change the API address by hand and enter the model names yourself.The interface shows "query failed"Upgrade past v3.16.2 for the query routing fix, or use a custom balance query template. Put hooks in general config, and raise the failover timeout.System message must be at the beginningChange the API format for NVIDIA Qwen, or upgrade cc-switch to v3.13.0+ to fix the 400 from providers such as Copilot.Gtk-CRITICAL: gtk_widget_get_scale_factor failedOn a GTK assertion failure running cc-switch under Linux or WSL2, edit settings.json to turn silent startup off, or clean the leftovers and reinstall.InputValidationError: Write failedTurn Tool Search, maximum-effort thinking, and the 1M context declaration off in cc-switch to reduce long-text truncation, or roll back to 3.13.0.Authorization header is badly formattedTurn "local routing" on in cc-switch and set the API format to native, resolving the 400 auth header error connecting GitHub Copilot to Claude Code.Not logged in · Please run /login ClaudeEdit `~/.claude.json` to move `PROXY_MANAGED` into the `approved` list, or answer yes at the prompt.Model mapping does not apply when connecting a Copilot subscriptionUpgrade to v3.15.0 or newer to fix the Copilot subscription model mapping not applying.API Error: 400 model_not_supportedRestart CC Switch to repair the proxy routing, or confirm whether the model ID requested became invalid after a Copilot policy change.The API returns a 400 error - Claude CodeUpgrade CC Switch to v3.16.1 for the 400 caused by MiMo missing reasoning_content; on Github Copilot, check the API format and proxy config.API Error: 400 Param Incorrect Claude CodeUpgrade CC Switch to v3.16.2+ to fix 400 errors caused by missing reasoning_content in MiMo thinking mode."Not logged in ? Please run /login" ClaudeEnable "API Proxy Takeover" in cc-switch when using hosted accounts like Copilot or Codex; direct upstream connections cause auth failures."event handler is re-entrant" error or startup crashUninstall Windows update KB5081263 or wait for cc-switch upgrade to avoid crashes and UI rendering failures caused by TAO EventLoop re-entry.Atomic replace failed os error 50Change the Claude Code configuration directory from a WSL path back to a Windows local path, or roll back to v3.19.0.Codex skill toggle not working: still loads legacy directory ~/.agents/skillsRemove leftover copies in ~/.agents/skills or disable them in config; CC-Switch cannot manage that shared directory.Body text appears in thinking blocks, visible text nearly emptyUpgrade cc-switch to a build with PR #4210. Reactive thinking-placeholder handling fixes long-context freezes and body-text collapse.
Qwen / 通义千问23
Codex model catalog template `gpt-5.5`Regenerate ~/.codex/models_cache.json and keep the Codex ChatGPT login, so CC Switch can write the route and show DeepSeek models.api failed deserialize json body intoDepending on the error: roll Claude Code back, upgrade cc-switch, turn the local proxy off, or stop sending unsupported fields.Codex provider API keys overwrite each other after a restart or switchEnable CC Switch local proxy takeover, or use experimental_bearer_token in config.toml, to bypass the auth.json sync defect.Codex connecting to Kimi reports HTTP 400 tool schema validation failureUpgrade cc-switch to v3.18.0+; if the error persists, wait for the MFJS schema cleaning feature to go live.EGL_BAD_PARAMETER white screen error - cc-switchInstall the native .deb or AUR package instead, or run the AppImage with LD_PRELOAD pointing at the system library, to resolve the ABI conflict.Qwen model errors or truncation - CC SwitchUpgrade to CC Switch v3.16.4 and use the matching new Qwen / OpenCode Go presets.kimi k2 reports 404 - CC SwitchA 404 from the connection test does not mean Kimi is unavailable. Set ANTHROPIC_BASE_URL to https://api.kimi.com/coding/ and verify by running Claude Code.Request body format invalid error - Claude CodeUpgrade cc-switch and enable the Rectifier, or add MAX_THINKING_TOKENS=0, for compatibility with third-party API formats.selected model (kimi-k2.5) may not exist errorDrop the trailing /v1 from the API address and use the Anthropic-compatible endpoint. For an old session with images, /compact or switch model.System message must be at the beginningChange the API format for NVIDIA Qwen, or upgrade cc-switch to v3.13.0+ to fix the 400 from providers such as Copilot.Console Go: Upstream request failedThis 400 comes from the OpenCode Go gateway rejecting an incompatible payload. Restart the proxy to restore session mapping, or avoid WebSearch and images.Range of input length should be [1, 1048576]Upgrade CC Switch to v3.19.0 for the fix where images were forwarded as base64 text and inflated tokens, or avoid view_image under the proxy.The thinking process is not shown after using the Claude Code proxy feature - cc-switchSwitch back to the Anthropic protocol to restore the thinking display; on `openai_chat`, change the block conversion in `streaming.rs`.usage null - Claude Code Qwen DashScopeUpgrade to cc-switch v3.16.3+, which fixes null streaming usage on the OpenAI-compatible interface automatically.This model is not supported when usingComment out model_provider = custom in config.yaml to go back to the official OpenAI provider.DeepSeek prefix cache hit rate drop - CC Switch fixUpgrade to cc-switch v3.16.3 to fix cache hit rate calculation errors caused by format conversion.Codex conversation history disappears after switching providerAfter upgrading to v3.16.0 or above, CC Switch automatically migrates third-party provider history into the custom bucket; original files are backed up.Model always thinks; disable not supported; use low, high, or maxSet glm-5.3 to forced thinking in cc-switch config; disable thinking.type: disabled.Invalid pages parameter: "" Claude CodeSwitch the API format to anthropic, or upgrade cc-switch to fix the empty pages field from Responses conversion.No endpoints available that support Anthropic - Claude CodeUpgrade Claude Code to v2.1.92+ and update cc-switch to fix OpenRouter parameter errors, or disable the local proxy.Unexpected item type in content.Enable content block filtering in config to remove Anthropic internal types unsupported by third-party backends.Bash(...) Waiting... Claude CodeDowngrade Claude Code to v2.1.104 or v2.0.64 (admin required), or switch to a compatible provider such as Alibaba Cloud Bailian.There's an issue with the selected modelRemove the trailing /v1 from ANTHROPIC_BASE_URL, launch from the CLI instead of the cc-switch button to use local routing, and check model support.
Anthropic (official)22
auth.json refilled with an existing account's auth.json - CodexUpgrade to v3.17.0+ and turn off 非接管切换时保留官方登录 (keep official login on non-takeover switch). Back up auth.json before upgrading.Cannot query usage for an official subscription in the Codex UI - CodexEnable local route mapping so specific models are counted. Third-party APIs need their own query script; Claude official and Free accounts are unsupported.API Error: 401 authentication_errorCheck the API key field, Base URL, key type, and cc-switch version, so Claude Code actually reads a valid credential.auth.json under .claude is forcibly replaced after a restart - CodexUpgrade cc-switch to the latest version, or turn proxy takeover off before signing in officially, so a stale snapshot cannot overwrite official credentials.codex deepseek cannot select a model - cc-switchCodex desktop hides third-party models with no official ChatGPT login. Sign in once, enable keeping the official login in cc-switch, and restart Codex.codex session expired - CodexBack up the config, then sync the auth.json token into the cc-switch database with a script, or turn usage refresh off so old data cannot overwrite it.Only a 200K context window shows in actual use - Claude CodeChange the default model in `~/.claude/settings.json` to the `[1m]` variant, or run `/model claude-opus-4-8[1m]` in the session, to activate the 1M context.Claude Official check failed: could not extract base_url - CodexCC Switch proxying and load balancing do not support an official Claude subscription — use a third-party API key. Forcing it errors and risks a ban.The cc-switch Claude config is overwritten - Claude CodeSign out of the official account in cc-switch and clear the login state to restore your developer mode proxy config.Unknown parameterTurn the cc-switch unified session history switch off and move sessions tagged custom back to openai, fixing the namespace parameter error.Config error: Claude Provider is missing base_url configurationTurn cc-switch local proxy takeover off and run /logout then /login in the terminal, or upgrade past v3.14.0.built-in provider IDs: openai - CodexUpgrade CC Switch to v3.16.4+, delete the old config and save again. Enable unified Codex session history and use openai_base_url.HTTP 401 Unauthorized - CC Switch AnthropicUpgrade cc-switch to a fixed build so it sends the x-api-key and anthropic-version headers Anthropic requires.Not logged in, please run /loginAdd hasCompletedOnboarding: true to ~/.claude.json and turn on "use custom Api key" in /config.usage null - Claude Code Qwen DashScopeUpgrade to cc-switch v3.16.3+, which fixes null streaming usage on the OpenAI-compatible interface automatically.Chat history under the original codex project disappears after connecting a third party - CodexUpgrade to cc-switch v3.16.4 or newer, turn on "unified Codex session history" in settings, and migrate the official sessions.ERR_BAD_REQUEST api.anthropic.com ClaudeTurn on "skip the first-install confirmation" in cc-switch and clear the conflicting environment variables to resolve the Claude Code ERR_BAD_REQUEST.Model always thinks; disable not supported; use low, high, or maxSet glm-5.3 to forced thinking in cc-switch config; disable thinking.type: disabled.model is not supported when using CodexDisable cc-switch proxy takeover to avoid writing unsupported model IDs to Codex, preventing official subscription bans or request rejections.Unexpected status 503 Service Unavailable CodexConfigure v2ray or other proxy tools to bypass LAN traffic, resolving the 503 error when Codex accesses cc-switch local routing.Codex cannot select models after configurationEnable keep-official-login in Codex app enhancements, then log in again and switch providers to restore the model list.rate_limit_error: Rate limited - Claude Codecc-switch does not support multi-account switching for Claude Code. Avoid third-party tools to prevent rate_limit_error.
NVIDIA17
Qwen model errors or truncation - CC SwitchUpgrade to CC Switch v3.16.4 and use the matching new Qwen / OpenCode Go presets.DeepSeek usage statistics show the wrong cost / cache hit rateUpgrade to v3.16.3+, fixing input double-counting on the chat→message conversion path, streaming without usage, and wrong model billing attribution.System message must be at the beginningChange the API format for NVIDIA Qwen, or upgrade cc-switch to v3.13.0+ to fix the 400 from providers such as Copilot.The proxy shows no requests at all - Claude CodeA known cc-switch bug: when an upstream such as Nvidia returns no usage data, the openai_chat path drops the log outright. No official fix yet.Unsupported parameter(s): prompt_cache_keyUpgrade cc-switch to the latest version, fixing prompt_cache_key not being stripped automatically when forwarding to the Nvidia API.Request forwarding failed: error sending request for urlCheck your local network connection and proxy configuration so cc-switch can reach the Nvidia API endpoint, resolving the 502 timeout.API Error: 422 format conversion errorThe upstream NVIDIA NIM GLM-5.2 model intermittently returns an empty choices array. Turn on failover in cc-switch or retry the request.Upstream HTTP 404 cc-switchUpgrade cc-switch to remove ?beta=true from NVIDIA requests. Kimi does not support Codex's Responses API; use Claude Code or a local adapter.Invalid pages parameter: "" Claude CodeSwitch the API format to anthropic, or upgrade cc-switch to fix the empty pages field from Responses conversion.KMS: DRM_IOCTL_MODE_CREATE_DUMB failed: Permission deniedResolve cc-switch white screen DRM permission errors on Linux by disabling WebKit hardware acceleration, specifying NVIDIA rendering, or using the CLI version.Nvidia check failed: Not found (404)Upgrade cc-switch to v3.16.3+; new version uses lightweight HTTP probing, fixing NVIDIA interface false 404 errors.not found(404) adding Claude Code desktop model APIUpgrade to v3.16.3+ to fix connectivity detection false positives, or ignore the 404 and test actual calls directly.Claude Code Tool Call Failure via NVIDIA Proxy - CC Switch FixNVIDIA free-tier models are unstable; switch to stable providers like deepseek-ai/deepseek-v4-flash.Unexpected status 503 Service Unavailable CodexConfigure v2ray or other proxy tools to bypass LAN traffic, resolving the 503 error when Codex accesses cc-switch local routing.Main window completely blank (pure white)Set WEBKIT_DISABLE_DMABUF_RENDERER=0 to restore normal rendering under WebKitGTK 2.52 + Mesa.Bash(...) Waiting... Claude CodeDowngrade Claude Code to v2.1.104 or v2.0.64 (admin required), or switch to a compatible provider such as Alibaba Cloud Bailian.There's an issue with the selected modelEnable local proxy and rectifier, and correctly set ANTHROPIC_BASE_URL to resolve NVIDIA or MiniMax model call failures.
OpenRouter16
Unknown name "prompt_cache_key" - cc-switchUpgrade cc-switch to v3.16.2+ to fix prompt_cache_key injection, or use an OpenRouter endpoint that natively speaks the Claude Code format.DeepSeek API key overwritten, causing a 401Upgrade cc-switch to v3.16.2+ and re-enter the DeepSeek API key, so it is not overwritten by another key when switching provider.502 Bad GatewayCodex/Claude Desktop reports 502 due to CLI proxy conflicts or app-server bypassing proxy. Remove proxy config or downgrade versions to fix.MCP config files edited by hand are overwritten by cc-switchcc-switch writes wholesale. Sync plugins through general config, and upgrade to v3.13.0 for the MCP cross-app leak fix.unexpected status 404 Not FoundUpgrade cc-switch to v3.16.4+ and set the upstream format explicitly to Chat Completions on the Codex provider form.Unsupported content type: redacted_thinkingRun a local proxy such as openrouter-proxy to filter out redacted_thinking blocks, and point the Claude Code config at that proxy.status=429 - Claude DesktopStreaming requests in CC Switch 3.15.0 skip the model rewrite and produce a 429. There is no official fix yet.ccstatusline shows the wrong model name - Claude CodeConfigure a Custom Command script in ccstatusline that reads the environment variables from the cc-switch database to show the real model.unexpected status 401 Unauthorized CodexUse the built-in DeepSeek template instead of a hand-built config, or add http_headers to the config file by hand to pass the API Key.The user sent an empty message.Set Kimi API format in cc-switch to OpenAI chat completions and enable proxy, or export history and rebuild the session.No endpoints available that support Anthropic - Claude CodeUpgrade Claude Code to v2.1.92+ and update cc-switch to fix OpenRouter parameter errors, or disable the local proxy.Reasoning Level Not AdaptedRe-add the OpenCode Go preset after upgrading to get the correct reasoning_effort config.thinking enabled but reasoning_content missingDisable Thinking mode in Claude Code, or upgrade cc-switch to fix Kimi/Moonshot tool call errors.Test returns 504: Claude Code unusableCheck whether cc-switch loaded the `glm-4.7` model mapping into the proxy; otherwise Claude Code requests glm directly and returns 504.Not logged in · Please run /loginWith the Codex provider, do not start Claude Code from the cc-switch built-in terminal. Open a global terminal so local routing handles OAuth.Third-party model still claims based on GPT-5 - CodexMove the `baseInstructions` override out of the `if profile != ProxyChat` block so it applies to ProxyChat. Currently it is not injected even when set.
Volcengine Ark / Doubao9
Codex connecting to Kimi reports HTTP 400 tool schema validation failureUpgrade cc-switch to v3.18.0+; if the error persists, wait for the MFJS schema cleaning feature to go live.Auth conflictLog out of the official Claude login, clear conflicting ANTHROPIC_* variables, set hasCompletedOnboarding to true in ~/.claude.json, and restart the terminal.Unkonw coding plan provider - cc-switch VolcengineUpgrade cc-switch to v3.20.0 or newer, then recreate the Volcengine Agent Plan provider or fix its Base URL.HTTP 400 unknown model 'ark-code-latest'Change the MiniMax provider model ID in CC Switch from the Volcengine-only ark-code-latest to MiniMax-M3 to resolve this 400.Range of input length should be [1, 1048576]Upgrade CC Switch to v3.19.0 for the fix where images were forwarded as base64 text and inflated tokens, or avoid view_image under the proxy.ModelScope reports 401 token invalid - Claude CodeAdd ANTHROPIC_MODEL to the provider environment variables in cc-switch to resolve the 401. It is not synced to settings.json by design.usage null - Claude Code Qwen DashScopeUpgrade to cc-switch v3.16.3+, which fixes null streaming usage on the OpenAI-compatible interface automatically.Gateway rejected model "claude-haiku-4-5"Turn isFullUrl=true on in the configuration and enter the complete API path, stopping cc-switch from appending /v1/messages automatically.Volcengine Agent Plan config.toml mismatch with official docsModify config.toml: set model_provider and section to volcengine-agent-plan; use env_key auth.
SiliconFlow / 硅基流动6
Codex model catalog template `gpt-5.5`Run `codex debug models --bundled > ~/.codex/models_cache.json` to generate the missing model cache, then switch the Codex route again.Auth conflictLog out of the official Claude login, clear conflicting ANTHROPIC_* variables, set hasCompletedOnboarding to true in ~/.claude.json, and restart the terminal.kimi k2 reports 404 - CC SwitchA 404 from the connection test does not mean Kimi is unavailable. Set ANTHROPIC_BASE_URL to https://api.kimi.com/coding/ and verify by running Claude Code.thinking enabled but reasoning_content missingDisable Thinking mode in Claude Code, or upgrade cc-switch to fix Kimi/Moonshot tool call errors.thinking type should be enabled or disabledThird-party APIs require thinking param as enabled/disabled. Fix via env var CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1 or alt+t toggle.There's an issue with the selected modelRemove the trailing /v1 from ANTHROPIC_BASE_URL, launch from the CLI instead of the cc-switch button to use local routing, and check model support.
ModelScope / 魔搭1
Not provider-specific143
cc-switch config not applied / not picked up after switching - Claude CodeConfirm the app actually reads the config path cc-switch writes to, then check routing and plugin toggles, duplicate keys, and version compatibility.cc-switch local proxy: 502 Bad Gateway, timeouts, cannot delete a providerUsually caused by unsupported official providers, config not applying, or UI state drift. Check proxy mode, provider type, and takeover state.cc-switch import / cloud sync failures: foreign keys, read-only database, overwritesUpgrade cc-switch and avoid syncing provider_health or a read-only db; after picking a file, click 导入 (Import) rather than 保存 (Save).Codex API key automatically modified leading to model xxx not foundUpgrade to v3.19.2+ to fix incorrect API Key convergence caused by reading auth.json when multiple suppliers share the same address.CC Switch config import / sync not taking effectUpgrade CC Switch to v3.16.1+, and check that WebDAV / cloud sync and the config write path are syncing correctly.There's an issue with the selected modelUpgrade CC Switch, restart the target tool, and start a new session so local routing strips the [1m] suffix and rematches the mapping.Claude Code errors or model mapping fails with the CC Switch proxy onUpgrade to v3.15.0 or newer, then set the global outbound proxy explicitly in cc-switch so a stale system proxy cannot be cached by an old process.Codex usage statistics doubled or inflatedUpgrade to v3.19.2 and run "Rebuild Codex Usage" on the usage page to correct historical inflated data.Failed to fetch the model listUpgrade cc-switch and add a model row before fetching the list. If it still fails, verify /v1/models permissions with curl.Usage statistics missing or the query failsTurn the usage statistics toggle on and check the provider script and protocol limits. Official statistics need Tool Search enabled.cc-switch crashes on launch / white screen / black screen / background crashUpgrade to the latest. Windows white screen: install WebView2. Proxy crash: change port, upgrade to v3.9.1+. Old macOS/Linux: use the CLI or distro package.'claude' is not an internal or external command, nor a runnable programUpgrade to v3.16.2+ to fix WSL/Windows path detection, or manually configure install directory override.cc-switch Skills management problems (duplicates, not detected, deleted repos returning, inconsistent state)Most Skills problems are fixed in later versions: v3.18.0 stops deleted default repos returning; v3.13.0 improves installed-Skill detection and dedup.cc-switch Skills import / sync problems (broken, not detected, directory out of sync)Usually a changed Skills directory layout, or the DB drifting from the file system. Manage through the cc-switch UI and check nested directory support.cc-switch usage query fails / keeps showing failed / undefinedUpgrade to v3.16.2+. If it still fails, fill in base_url and the access token in NewAPI format, or turn off auto-refresh and the usage query.cc-switch update check fails / update install failsUsually insufficient permissions or GitHub network trouble. Install as administrator or use the portable build, or download manually from GitHub Releases.cc-switch white screen / black screen / crash / no window on launchWork out whether it is corrupt window state, a WebView2/OS compatibility problem, or security software. Handle per symptom; use the CLI build if needed.cc-switch loses Claude Code / Codex MCP configurationUpgrade cc-switch, disable or use unified MCP takeover carefully, find the one broken MCP entry, and fix rewritten command/type/path fields.ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYRun `claude daemon stop --any` to kill leftover Claude background processes, then keep only one auth variable in `env` in `~/.claude/settings.json`.All installed claude code skills disappear - cc-switchCheck whether ~/.claude/skills was rewritten to ~/.agents/skills; if so move it back and restart Claude Code. Still missing: upgrade cc-switch and reinstall.Config not applied: MCP not imported and environment variables overwrittencc-switch overwrites wholesale. Extract hand-edited config into 通用配置 (general config), import MCP by hand, and tick the write-general-config option.ANTHROPIC_BASE_URL rewritten to 127.0.0.1With local routing on, ANTHROPIC_BASE_URL pointing at http://127.0.0.1:15721 is expected. Move shared fields into 通用配置 and restart Claude Code.All Codex settings and plugins reset after restarting CC SwitchPut cross-provider Codex/Claude Code config and plugins into 通用配置 (general config) so a switch cannot wipe them. Recover lost content from the default provider.cc-switch provider management: cannot delete, or adding failsMostly UI limits or version bugs. The last provider cannot be deleted; for input and delete failures, upgrade or turn the feature off.auth.json OPENAI_API_KEY not synced - CodexUpgrade cc-switch to v3.20.1. Switching Codex providers no longer overwrites auth.json, fully resolving the key retention issue.[CODEX-SYNC] deferred: parent rollout has not reached the child fork momentA cc-switch 3.18.0 Codex sync defect: a finished parent rollout older than the child fork timestamp defers forever. No fix yet — avoid creating forks.claude mcp list shows ✘ Failed to connectOn Windows cc-switch wraps the MCP npx command in cmd /c, breaking the stdio pipe. Edit ~/.claude.json back by hand, or rewrite with claude mcp add.cc-switch usage statistics anomalies: 0/200K, double billing, pricing that will not deleteUpgrade to v3.16.3 for zero tokens, v3.17.0 for better Codex derived-session dedup, and v3.19.0 for deleted built-in pricing returning.Codex errors after enabling the local proxyAdd `/v1` to the end of the Codex provider request address in CC Switch, save, and retry.Local environment check reports codex is not installedRun `codex --version` first. If the terminal works but cc-switch cannot detect it, the install path is outside its probe range — upgrade to v3.16.4+.There's an issue with the selected modelCheck the model name and provider config in cc-switch. The model test is unreliable — go by an actual request.Not logged in · Run /login - Claude CodeTurn CC Switch local proxy takeover on, check the API address includes /v1, or launch the terminal from CC Switch so env vars are injected.Atomic replace failed — request not supported, os error 50 - cc-switchUpgrade cc-switch to v3.20.0 or newer, which fixes the regression where atomic config file replacement failed on WSL paths.Opening cc on Windows 11 gives a process but no window - cc-switchOn Windows 11, try Windows 7 compatibility in the exe properties. If that fails, use cc-switch-cli to bypass the Tauri 2 GUI white screen.Auto-upgrade fails; reinstalling the new version still leaves the old oneUninstall the old version completely, clear the leftovers, and reinstall — that resolves a failed auto-upgrade or a version that will not update.config.toml quietly reverts to the earlier configurationThe cc-switch database is the single source of truth. Hand edits to local files are overwritten — change things in the MCP manager or general config instead.Codex OAuth switch to API Key ineffectiveUpgrade to v3.17.0+ and disable the 'Keep Official Plugin' option in settings.x-anthropic-billing-header breaks caching - CC SwitchUpgrade CC Switch to v3.15.0+, which fixes the dynamic billing header breaking prefix caching in openai_responses mode.Auth conflictDelete primaryApiKey from claude.json, or change the environment variable to ANTHROPIC_API_KEY, to resolve the auth conflict.bash: cannot set terminal process group - cc-switchA known 3.18.0 bug when checking the local proxy environment on Linux. Downgrade below 3.18.0 and configure the proxy in the terminal by hand.Cowork WebFetch cannot reach the external network - Claude DesktopAdd "coworkEgressAllowedHosts": ["*"] to the 3P gateway profile CC Switch generates, so the Cowork sandbox can reach external hosts.Codex usage statistics inflated: fork/resume sessions double-count parent history tokensUpgrade cc-switch to v3.18.0+ and rebuild the usage database to fix fork sessions double-counting parent history tokens.It says 1.6 billion used — usage statistics spiking - CodexUpgrade cc-switch to v3.18.0 to fix the sub-agent history replay double-count bug; it rebuilds the polluted Codex usage data automatically.Context window errors and /compact will not run - Claude Code / CodexSet the correct model_context_window for the model by hand in cc-switch, so a wrong default does not break compaction or cause a 422.env: node: No such file or directoryUpgrade cc-switch to v3.18.0+, which fixes node not being found when updating Codex because a macOS GUI launch has no PATH.Install reports libwebkit2gtk-4.1.so.0 is missing - cc-switchInstalling the rpm on Rocky 9 fails on a missing webkit dependency. Upgrading the system to Rocky 10.1 resolves it.No providers have been added yet - CC SwitchAfter an upgrade or cloud sync the UI state and the database fall out of sync. Restart the app fully, or re-import a backup, to get the provider list back.After upgrading to 3.15 the remaining quota is not visible at all - cc-switchUpgrade cc-switch to 3.16.0 to fix usage not displaying. For high disk/CPU/GPU use, turn the local proxy or the routing master switch off.On the add provider page, search results cannot be clicked to select - cc-switchUpdate to a version with the UI fix. As a stopgap, Tab to the preset after searching and press Enter to select it."There's an issue with the selected model"Upgrade cc-switch to v3.13.0+ for the environment variable sync fix, and use the alias mapping variables to override the default model.cc-switch login and auth config cannot be entered or deletedUpgrade to v3.17.0 or newer: the Google OAuth misdetection and the legacy-provider gating that hid the API key field are both fixed.'claude' is not recognized as an internal or external commandManually specify the full path to the claude executable in CC Switch settings, or ensure the path is added to the system PATH.cc-switch window display anomalies - UI rendering problemsUpgrade to a version with the UI fixes. Address bar overflow and rendering problems are fixed; the title bar background colour awaits a later fix.cc-switch Installation: Main Interface Not Opening or No ResponseUse the CLI version or adjust Windows OS versions to resolve Tauri compatibility issues causing UI failure.General config ignored - cc-switchUpgrade to cc-switch v3.12.2+, where general config fragments persist across a restart. As a stopgap, turn general config off and write the JSON directly.Adding a provider fails: crypto.randomUUID is not a functionUpgrade to cc-switch v3.9 or newer, which adds a crypto.randomUUID fallback for older macOS; or upgrade macOS past 12.3 Monterey / 11.6.5 Big Sur.Why is there no automatic model fetch option in the latest ubuntu build? - cc-switchUpgrade cc-switch to v3.12.0 or newer, which restored automatic model detection on every platform.After enabling app enhancements my original api cannot connect and returns 503 - CodexTurn app enhancements, or keep-official-login, off in cc-switch to resolve the 503 and the timeouts.Cannot import providers, cannot manage sessions - cc-switchUpgrade cc-switch for jsonc config support. Under WSL2, clone the database to a local Windows path on a schedule to work around SQLite locking.not installed or not executable - OpenCodeUpgrade CC Switch to v3.11.0+ for the path detection fix, or symlink opencode into ~/.local/bin by hand.Config changes in the GUI have no effect at allOMO v4.19.3 moved config to ~/.omo/omo.jsonc, so cc-switch writes to a dead path. Edit the new config file by hand, or upgrade cc-switch.I clicked both the msi and the exe and nothing happensLaunch the installer as administrator from PowerShell. If it still fails, uninstall the old version first, or use the portable build.Several installations detected, or shown as not installed - Codex/OpenCodeUpgrade CC Switch to v3.17.0 or newer, fixing npm path resolution on Windows and the wrapper script misdetection.assertion failedUpgrade cc-switch to the latest version for the underlying GUI framework crash fix, or try closing proxy software and resetting the network.skills will not fetch / plugins rolled back - cc-switchUpgrade cc-switch to v3.9.0 for the skills fetch fix. If plugin config is overwritten, back enablePlugins up into a general config fragment.Every click first pops up the system folder picker - cc-switchEdit src/App.tsx to remove the pickDirectory() call in handleOpenTerminal, so opening a terminal stops forcing a folder picker.Claude Code cannot call the model after configuration - CC SwitchConfirm cc-switch only rewrites config, then check the skip-first-install option, proxy and region restrictions, and upstream API compatibility.Claude Code cannot see a newly enabled MCP - cc-switchUpgrade cc-switch to the latest version, or copy the contents of the sibling .claude.json into the one inside CLAUDE_CONFIG_DIR by hand.zsh:1: command not found: claudeUpgrade cc-switch to v3.16.4 or newer, fixing PATH not loading when the terminal launches, or switch to the system default terminal for now.API Error: 400 invalid request: Claude Desktop model route is not configuredUpgrade cc-switch to v3.16.0+ and configure routing rules explicitly for sub-task models such as Haiku and Opus on the provider form.Your access token could not be refreshedQuit and restart Codex completely (or kill the codex process) to clear the stale authorisation, then sign in again.auth.json not updated after switching channel - CodexCodex auth.json is not updated after switching channels. Manually modify auth.json or wait for cc-switch to fix it.A disabled MCP becomes active again after restarting the machine - CodexNo official fix yet: clean the conflicting MCP nodes out of the config file by hand, and use deeplink import carefully so custom settings survive.Missing environment variableDelete the env_key entry from config.toml, or set OPENAI_API_KEY in Windows environment variables, and confirm the provider supports the native response format.Codex sub-agents in copy mode have their full usage counted wrongly - CodexUpgrade to v3.18.0, fixing usage double-counting from sub-agent copy mode and duplicate proxy streaming records inflating the totals.cc-switch keeps deleting my environment variablesClosing the conflict dialog does not delete anything. To clear it for good, remove the entry from your shell config or the Hermes config.yaml.Format conversion error: Failed to parse upstream response - Claude CodeUpgrade cc-switch for the streaming response parse fix, or enable full request path and complete the URL for non-standard APIs such as iFlytek.API Error: 400 invalid_parameter_errorCheck the base url and model config, enable the rectifier, or ignore the false 400 from Stream Check.Frozen after clicking updateForce quit and reopen the app, or update by hand to v3.16.3 or newer for the update deadlock fix.command turns into cmd automatically and the mcp service fails to connect - Claude CodeEdit MCP config inside the cc-switch UI so it is not overwritten, and correct npx commands rewritten wrongly as cmd by hand.The "+" button disappeared and providers cannot be addedMaximise the cc-switch window or drag it wider to reveal the hidden "+" add provider button.openclaw.json is not valid JSON - OpenClawAdd double quotes around the models key in ~/.openclaw/openclaw.json by hand to fix the strict JSON parse failure.Typing limit as the key clears the input automatically - OpenCodeThe cc-switch UI treats limit as a reserved word, so edit opencode.json directly to set the model context and output limits by hand.skills synced to apps I did not select - OpenCodeUpgrade cc-switch to v3.12.3 for the implicit inference bug fix, and turn off reading skills from the .claude directory in OpenCode settings.protocol handler not registeredOn Linux, change Exec to cc-switch %U in the CC Switch.desktop file so Chromium browsers recognise the import protocol. For WSL sync, check admin rights.The Skill source directory has no SKILL.md; refusing to sync to avoid overwriting the target directoryCC Switch fails to install when the skill repository layout does not match the skillId or is nested — install by hand with `npx skills add`.CC Switch does not show the remaining quotaCodex free accounts cannot query quota because of an official API limit, with no fix. Missing Grok Build usage needs a source change to streaming parsing.The interface does not refresh by itself; only dragging the window triggers a UI redrawUpgrade to v3.13.0+ for the Linux interface freeze fix; the Windows built-in Administrator needs a registry change to open the main window.os error 183, cannot create a file when that file already exists - Claude CodeOn Windows, cc-switch reports os error 183 when the config directory is a Junction. Remove the directory link for now, or wait for an official fix.Error 2503 - cc-switch MSI install permission deniedRun msiexec from an administrator command prompt, or download the Portable build, to work around the Windows 11 MSI permission restriction.WSL path configured but skills are not read - cc-switchWindows cc-switch cannot read or install WSL ~/.agents/skills. Install and run the Linux build of cc-switch inside WSL instead.No skill found in the ZIP file (it must contain a SKILL.md)Upgrade cc-switch to v3.16.2 or newer, and delete the old directory by hand before importing a Skill of the same name.API Error: 400 Invalid request ErrorRoll the Claude Code CLI or VSCode plugin back to an older version (2.1.150 or 145) and disable auto-update to resolve the 400.API Error: 404 - Claude Code path assembly or unregistered routeCheck whether the API address was assembled wrongly, or upgrade cc-switch for the 404 from OpenAI-compatible mode and the missing local route.Symlink target not found: .DirIcon broken symlink causes - CC SwitchAppImage install fails due to broken .DirIcon symlink. Fix by converting it to a relative path.macOS cc-switch version detection error / freezeModify the fastfetch call in .zshrc to add a [[ -t 1 ]] check for non-interactive output.cc-switch token count is 0An older database missing pricing for a new model makes token statistics read 0. Upgrade to a version containing the fix.The interface is completely blank - cc-switchUpgrade cc-switch to the latest version, which disables DMABUF rendering by default, or use the CLI build to resolve the Tauri 2 white screen.Input length exceeds model's maximumAdd the CLAUDE_CODE_AUTO_COMPACT_WINDOW environment variable to the provider config to cap the context window.missing field content HTTP 400 - CodexReplace `content:null` with an empty string in CC Switch to fix strict upstream rejections.Upstream HTTP 502 - CC Switch (Claude CodeCheck your local network proxy (Clash/ZeroTier) config and make sure it does not interfere with direct requests.effortLevel: max is not supportedSet the CLAUDE_CODE_EFFORT_LEVEL environment variable to high, or to max on v3.14.1+, avoiding the invalid max value mapping.Claude Code ignores the model switch in an old windowA Claude Code session already open does not pick up a model switch in cc-switch. Close and reopen the window, or use the new model in a new window.Opening a terminal does not use the complete configurationExtract through the "shared configuration" panel and tick write, so a full overwrite does not lose plugins and other shared config.The mcp I configured in .claude.json is reset as soon as the proxy startsCC Switch syncs by full overwrite. Import your MCP servers and manage them all inside CC Switch rather than editing .claude.json.Provider Auth cannot be restored when auth.json does not exist - CodexUpgrading to v3.13.0 resolves Provider Auth failing to restore during multi-device sync when auth.json is missing.auth.json automatically reuses the previous provider - CodexTurn off the "keep the official login when switching to a third party" option in settings.Codex history disappears and it will not work - CodexUpgrade to cc-switch v3.16.4 to fix zstd decode failures and CODEX_SQLITE_HOME state database probing that lose or desync sessions.Cannot install or update Codex CC SwitchUpgrade to v3.16.3+ and reinstall/update Codex from 设置 → 关于 (Settings > About); if v3.16.5 still fails, report it with logs.Connecting gpt5.5 in the codex desktop client, image input and reasoning strength switching do not work - CodexDelete the cc-switch config file under Codex and let it rebuild, or edit cc-switch-model-catalog.json to add the missing reasoning tiers and image modality."Expecting property name enclosed in doubleUpgrade CC Switch to v3.16.4 or newer to fix the JSON parse error caused by a large request body not being decompressed.Failed to initialize gtk backend cc-switchRunning cc-switch in a headless environment (a server or container) errors for lack of GTK. Use the cc-switch-cli command-line build instead.version `GLIBC_2.35' not foundThe official AppImage needs a newer GLIBC. On RHEL/Rocky Linux, use the community Flatpak build or one built on an older base image.env: node: No such file or directoryUpgrading cc-switch to v3.18.0 or newer fixes the environment variable resolution failure caused by a stripped PATH in the GUI.This software does not run on macOSRun `brew update` and retry the install, or download the DMG from GitHub Releases and install it by hand.UI freezes after clicking any input field on LinuxSet WEBKIT_DISABLE_COMPOSITING_MODE=1 or upgrade to v3.14.1 and enable the native title bar to fix Linux UI freezes and unresponsive buttons.Missing API key · Run /login - Claude CodeClear the conflicting ANTHROPIC_BASE_URL environment variable, or upgrade CC Switch to use its built-in environment variable conflict detection.error code is 2503 or icon turns white after upgradeRemove Icon="ProductIcon" from per-user-main.wxs to fix install errors and missing icons. Existing users must unpin and repin.S3 HEAD bucket failed: 404 Not FoundIf the endpoint contains the bucket name, remove the bucket field from the config to prevent path duplication.cc-switch routing mode overwrites Claude Code settings.jsonAdd custom env vars to CC-Switch Provider JSON, not directly to Claude Code settings.json.Installation failed: Skill directory does not exist cc-switchUpgrade to cc-switch v3.14.0 or use local ZIP installation to fix URL install failures for repos with SKILL.md in the root.Skill not loaded correctly - cc-switchcc-switch only recognizes directories containing SKILL.md; manually extract the skills subdirectory from marketplace structures for loading.stream disconnected before completion CodexUpgrade cc-switch to v3.14.0 or later to fix connection drops caused by proxy forwarding hop-by-hop response headers.'/d' is not recognized as an internal or external command CC SwitchBypass '/d' errors by using English directories or mklink /J junctions; fix permanently by ensuring .bat files use CRLF and chcp 65001.AnyRouter 429 error in Claude CodeAdd "model": "opus[1m]" to the provider JSON and restart Claude Code. If 429 persists, upstream rate limiting is likely; switch networks or use haiku.Save keeps failing and asks for the API endpointUpgrade to cc-switch v3.11.0 or later. It natively supports AWS Bedrock and no longer requires an API endpoint or API Key.cc-switch Codex session history missingDisable Codex routing or unified history in cc-switch, make model_provider match old sessions, then restart Codex to restore history.Error in .codex cc-switch.db causes cc-switch startup failureClose cc-switch, clean records with empty base_url in cc-switch.db, or delete the database file and restart.切换路由状态失败:{{detail}}Upgrade to v3.16.2 or later. The route panel then shows the real backend error detail.Local proxy toggle UI component is truncated and cannot be fully displayedUpgrade cc-switch to the latest version to fix UI truncation caused by window scaling and tray menu overflow.claude.ai connectors are disabledUnset `ANTHROPIC_API_KEY` to restore claude.ai login and organization connector loading.Model automatically switched to haiku cc-switchBuilt-in Claude Code behavior: haiku reads files. cc-switch only manages environment variables and does not change model selection.CLAUDE_CODE_ATTRIBUTION_HEADER causes cache invalidationSet CLAUDE_CODE_ATTRIBUTION_HEADER=0 to disable the dynamic request header and restore KV cache hits.Prompts /login and cannot use Codex API - Claude CodeDo not encode reasoning level in the model ID. Keep gpt-5.4; CC Switch maps it automatically.Cannot find authorization login entry in cc-switchUpgrade cc-switch or migrate providers from Claude Code to restore the Claude Desktop authorization login entry.unexpected status 403 ForbiddenDisable the cc-switch local proxy or check upstream API access to the codex-auto-review model.ANTHROPIC_CUSTOM_HEADERS lost after Deep Link importUpgrade to v3.12.0 or later to fix a bug where custom environment variables are dropped during Deep Link import.ENOENT: no such file or directory, scandir Claude CodeIn VSCode Remote SSH, configure Claude Code authentication on the remote Linux server and create missing directories; local settings do not sync.settings.json env not taking effect - Claude CodeClaude Code reads settings.json only at startup. After switching providers, restart Claude Code or set env vars manually.cc-switch: two of the five default MCPs fail to addInstall missing `uvx` and `mcp-server-time` dependencies (`pipx install uv`, `pipx install mcp-server-time`), then re-add the MCP.total_cost_usd not recalculated by cc-switchUpgrade cc-switch to v3.15.0+ to automatically recalculate historical costs after modifying model pricing.Unsupported parameter: 'max_tokens'The route forwards max_tokens to a model that only accepts max_completion_tokens. Verify parameter support or wait for a cc-switch fix.Please check WebDAV username/passwordcc-switch masks and clears WebDAV password on write-back. Upgrade to fix; temporarily re-enter password.
This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.