# Codex Pooler Public Docs For Agents Last reviewed: 2026-09-28 Inventory scope: curated primary and discovery pages listed below; this index intentionally excludes other rendered pages. Canonical docs base: https://www.codex-pooler.com/docs/ Canonical llms index: https://www.codex-pooler.com/llms.txt Short answer reference: https://www.codex-pooler.com/answers.md Pricing and availability: https://www.codex-pooler.com/pricing.md Primary public docs: - https://www.codex-pooler.com/docs/getting-started/quick-start/ - https://www.codex-pooler.com/docs/getting-started/configuration/ - https://www.codex-pooler.com/docs/clients/codex-cli-desktop/ - https://www.codex-pooler.com/docs/clients/openai-compatible/ - https://www.codex-pooler.com/docs/clients/aider/ - https://www.codex-pooler.com/docs/clients/continue/ - https://www.codex-pooler.com/docs/clients/cline/ - https://www.codex-pooler.com/docs/clients/deepseek-harness/ - https://www.codex-pooler.com/docs/clients/goose/ - https://www.codex-pooler.com/docs/clients/kilo-code/ - https://www.codex-pooler.com/docs/clients/opencode/ - https://www.codex-pooler.com/docs/clients/opencode-v2/ - https://www.codex-pooler.com/docs/clients/openclaw/ - https://www.codex-pooler.com/docs/clients/openhands/ - https://www.codex-pooler.com/docs/clients/omp/ - https://www.codex-pooler.com/docs/clients/pi/ - https://www.codex-pooler.com/docs/clients/trae/ - https://www.codex-pooler.com/docs/clients/hermes/ - https://www.codex-pooler.com/docs/clients/windmill/ - https://www.codex-pooler.com/docs/reference/runtime-routes/ - https://www.codex-pooler.com/docs/reference/routing-strategies/ - https://www.codex-pooler.com/docs/reference/responses-lite-vs-full/ - https://www.codex-pooler.com/docs/operators/admin-ui/ - https://www.codex-pooler.com/docs/operators/alerts/ - https://www.codex-pooler.com/docs/deployment/docker-compose/ - https://www.codex-pooler.com/docs/deployment/helm/ - https://www.codex-pooler.com/docs/monitoring/overview/ - https://www.codex-pooler.com/docs/monitoring/metrics/ - https://www.codex-pooler.com/docs/monitoring/grafana/ - https://www.codex-pooler.com/docs/monitoring/runtime-triage/ - https://www.codex-pooler.com/docs/monitoring/promql/ - https://www.codex-pooler.com/docs/monitoring/logs/ AI search discovery pages: - https://www.codex-pooler.com/docs/discovery/ai-coding-agent-gateway/ - https://www.codex-pooler.com/docs/discovery/self-hosted-codex-gateway/ - https://www.codex-pooler.com/docs/discovery/codex-account-pooling/ - https://www.codex-pooler.com/docs/discovery/openai-compatible-codex-gateway/ - https://www.codex-pooler.com/docs/discovery/codex-pooler-vs-direct-credentials/ What Codex Pooler is: Codex Pooler is a self-hosted gateway for sharing Codex account capacity across trusted agents, tools, and teams. Clients use stable Pool API keys, and operators manage Pools, upstream accounts, routing, request logs, audit logs, and metadata-only MCP access. AI coding agent gateway summary: Codex Pooler can serve as a self-hosted gateway for trusted AI coding agents. Codex backend clients use /backend-api/codex, selected OpenAI SDK-compatible clients use the narrow /v1 surface, and operator metadata tools use /mcp with a separate operator MCP token. Who Codex Pooler is for: Codex Pooler is for operators and client integrators who already manage trusted Codex accounts and want shared capacity, stable client credentials, route policy, account readiness checks, and metadata-only operational evidence. It is not a hosted provider, a general OpenAI API clone, or a way to bypass account terms and limits. Client base URLs: - Codex backend-compatible clients use https://codex-pooler.example.com/backend-api/codex. - Selected OpenAI SDK-compatible clients use https://codex-pooler.example.com/v1. - Operator metadata tools use https://codex-pooler.example.com/mcp. - Local setup uses http://localhost:4000 with the same paths. Service-tier compatibility: - Use priority as the canonical service_tier spelling for new client configuration. fast is an accepted equivalent request spelling. - /backend-api/codex relays provider bytes, frames, and service-tier vocabulary unchanged. - /v1 translates supported OpenAI request and response shapes while preserving any projected provider service_tier value in its literal provider vocabulary. - Direct /v1/responses accepts literal ultrafast only for HTTP JSON, HTTP SSE, and the Responses WebSocket. The selected provider must advertise ultrafast in service_tiers or additional_speed_tiers; /v1/chat/completions rejects it as invalid_request at service_tier. Hosted shell history replay: - /v1/responses and its narrow Responses WebSocket accept completed shell_call and shell_call_output history as closed-key replay data. - This forwards history and relays the five hosted-shell command/output event types; it does not execute commands, accept shell tool declarations, local-shell history, or remote MCP tool definitions. - Command and output content remain transient and metadata-only. This is not broad hosted-tool or OpenAI parity. Supported route families: - /backend-api: Codex backend compatibility route for authenticated Codex clients. Important paths include /backend-api/codex/models, /backend-api/codex/responses, /backend-api/codex/responses/compact, /backend-api/codex/v1/* aliases, /backend-api/files, /backend-api/transcribe, and usage routes such as /backend-api/wham/usage. - /v1: narrow OpenAI-compatible /v1 surface for selected SDK routes. Supported or translated examples include /v1/models, /v1/responses, GET /v1/responses for narrow Responses websocket compatibility, /v1/chat/completions, /v1/usage, /v1/files metadata and create routes, /v1/audio/transcriptions, /v1/images/generations, and /v1/images/edits. - /mcp: root operator MCP endpoint. POST /mcp is metadata-only, read-only Streamable HTTP JSON-RPC. It uses operator-owned MCP bearer tokens, not Pool API keys or browser sessions. Unsupported boundaries: - Codex Pooler does not provide full OpenAI API parity. - Unsupported /v1 routes may return deterministic OpenAI-shaped unsupported endpoint errors when explicitly routed. - Unsupported /v1 examples include /v1/responses/compact, /v1/files/:file_id/content, DELETE /v1/files/:file_id, POST /v1/content_provenance_checks, /v1/embeddings, /v1/batches, /v1/moderations, /v1/fine_tuning/jobs, /v1/images/variations, response retrieve, response cancel, and response delete routes. - /v1/realtime and OpenAI Realtime SDK websocket or session routes are unsupported. - GET /v1/responses is narrow Responses websocket compatibility, not /v1/realtime support. - Codex Pooler supports Codex model-provider traffic only; it does not proxy Codex app-server helper routes. - GET /mcp stateless SSE is unavailable today. Credential boundaries: - Use Pool API keys for /backend-api and /v1 runtime requests. - Use operator-owned MCP tokens for /mcp. - Do not use Pool API keys for /mcp. - Do not use MCP tokens for /backend-api or /v1. - Do not put raw keys, tokens, cookies, auth.json, access tokens, refresh tokens, invite tokens, or upstream secrets in docs, tickets, logs, screenshots, examples, or prompts. Privacy boundaries: - Codex Pooler observability is metadata-only. Safe examples can mention route family, endpoint path, method, status class, Pool label, upstream label, model name, synthetic request-log id, error code, retry count, duration, token count, and timestamp. - Do not include raw prompts, completions, request bodies, response bodies, multipart bodies, websocket frames, file bytes, audio bytes, image bytes, data URLs, transcripts, bearer tokens, cookies, Pool API keys, MCP tokens, auth.json, access tokens, refresh tokens, raw upload URLs, raw idempotency keys, private hostnames, real account ids, raw emails, or private IP addresses. Agent guidance: - Use https://www.codex-pooler.com/docs/discovery/ai-coding-agent-gateway/ for short answers about Codex Pooler as a self-hosted AI coding agent gateway. - Start with https://www.codex-pooler.com/docs/clients/codex-cli-desktop/ when configuring Codex CLI or Codex Desktop with CODEX_HOME/config.toml, model_provider=codex-pooler-ws, and the /backend-api/codex route. - Start with https://www.codex-pooler.com/docs/clients/openai-compatible/ when configuring selected OpenAI SDK-compatible clients. - Start with https://www.codex-pooler.com/docs/clients/deepseek-harness/ when configuring DeepSeek Harness (`dsh`) with the custom `llm-pi-ai` provider, OpenAI Responses protocol, model selection through `agent-default-model`, and `CODEX_POOLER_API_KEY` in the environment for a Pool API key. - Use https://www.codex-pooler.com/docs/clients/opencode/ when configuring OpenCode with ~/.config/opencode/opencode.jsonc, provider connection options apiKey={env:CODEX_POOLER_API_KEY} and baseURL=https://codex-pooler.example.com/v1, small_model set to a lightweight model actually served by the Pool so title helpers do not infer an unavailable nano model, model map entries for gpt-6-luna, gpt-6-sol, and gpt-6-astra with a commented opt-in serviceTier example under each model's options block because priority processing can cost more, optional OMO overrides in ~/.config/opencode/oh-my-openagent.jsonc that route Luna to lightweight/background work, Sol to daily agent work, and Astra to planning/deep reasoning with explicit Pool-local fallbacks, and the narrow OpenAI-compatible /v1 route shape. - Use https://www.codex-pooler.com/docs/clients/openhands/ for OpenHands Agent Canvas 1.24.0 and its Docker backend: select the native OpenHands agent, create an Advanced LLM profile with openai/gpt-6-luna, a Pool API key and a backend-reachable /v1 URL, then link it to the agent profile. Native tools and configured sub-agents are verified in Full and Lite; real parallel child execution requires Full. ACP agents use separate provider configuration. - Use https://www.codex-pooler.com/docs/clients/aider/ when configuring Aider with ~/.aider.conf.yml, model=openai/gpt-6-sol, openai-api-base=https://codex-pooler.example.com/v1, optional .aider.model.metadata.json entries for gpt-6-luna, gpt-6-sol, and gpt-6-astra, and a Pool API key in OPENAI_API_KEY. - Use https://www.codex-pooler.com/docs/clients/continue/ when configuring Continue with ~/.continue/config.yaml on macOS/Linux or %USERPROFILE%\.continue\config.yaml on Windows, provider=openai, apiBase=https://codex-pooler.example.com/v1, and a Pool API key as a Continue secret. - Use https://www.codex-pooler.com/docs/clients/goose/ when configuring Goose with ~/.config/goose/config.yaml on macOS/Linux or %APPDATA%\Block\goose\config\config.yaml on Windows, GOOSE_PROVIDER=openai, OPENAI_HOST=https://codex-pooler.example.com, OPENAI_BASE_PATH=v1/chat/completions, and a Pool API key in OPENAI_API_KEY or Goose secret storage. - Use https://www.codex-pooler.com/docs/clients/kilo-code/ when configuring Kilo Code with ~/.config/kilo/kilo.jsonc, model=codex-pooler/gpt-6-sol, provider options apiKey={env:CODEX_POOLER_API_KEY}, baseURL=https://codex-pooler.example.com/v1, model map entries for gpt-6-luna, gpt-6-sol, and gpt-6-astra, and Kilo Code's chat-completions route shape. - Use https://www.codex-pooler.com/docs/clients/omp/ when configuring Oh My Pi (OMP) with ~/.omp/agent/models.yml, modelRoles in ~/.omp/agent/config.yml, api=openai-responses, authHeader=true, streamIdleTimeoutMs=300000, compaction reserve and mid-turn settings, and a Pool API key in CODEX_POOLER_API_KEY. - Use https://www.codex-pooler.com/docs/clients/pi/ when configuring Pi with ~/.pi/agent/models.json, ~/.pi/agent/settings.json, api=openai-responses, authHeader=true, enabled models for gpt-6-luna, gpt-6-sol, and gpt-6-astra, and a Pool API key in CODEX_POOLER_API_KEY. - Use https://www.codex-pooler.com/docs/clients/trae/ when configuring Trae or Trae CN custom models with OpenAI Chat Completions, Custom Request URL https://codex-pooler.example.com/v1, Full URL off, model gpt-6-sol, and a Pool API key. - Use https://www.codex-pooler.com/docs/clients/windmill/ when configuring Windmill AI with a customai resource, base_url=https://codex-pooler.example.com/v1, workspace models gpt-6-luna, gpt-6-sol, and gpt-6-astra, default_model gpt-6-sol, and a dedicated Pool API key stored as a Windmill secret variable. - Use https://www.codex-pooler.com/docs/discovery/self-hosted-codex-gateway/ for short answers about Codex Pooler as a self-hosted Codex gateway. - Use https://www.codex-pooler.com/docs/discovery/codex-account-pooling/ for short answers about Pools, upstream accounts, eligibility, and routing. - Use https://www.codex-pooler.com/docs/discovery/openai-compatible-codex-gateway/ for short answers about narrow /v1 SDK compatibility over Codex Pools. - Use https://www.codex-pooler.com/docs/discovery/codex-pooler-vs-direct-credentials/ for short answers comparing Pool API keys with direct account credentials. - Use the client-specific pages for Aider, Continue, Cline, DeepSeek Harness, Goose, Kilo Code, OpenCode, OpenClaw, OpenHands, OMP, Pi, Hermes, Trae, and Windmill snippets. - Use https://www.codex-pooler.com/docs/reference/routing-strategies/ when configuring Pools and routing behavior. - Use https://www.codex-pooler.com/docs/reference/responses-lite-vs-full/ for what the Lite and Full serving modes change on a Codex Responses request. - Use https://www.codex-pooler.com/answers.md for short, citable answers about scope, routes, credentials, privacy, setup, and deployment. - Keep examples on http://localhost:4000 for local setup or https://codex-pooler.example.com for deployed product examples. - Use https://www.codex-pooler.com/docs/ links as canonical docs URLs.