For AI agents: the complete documentation index is available at https://aioproxy.dev/llms.txt, the full documentation bundle is available at https://aioproxy.dev/llms-full.txt, and this page is available as Markdown at https://aioproxy.dev/index.md.
AIO Proxy
  • English
      • Open source · Local-first model gateway

      Every model. Every client.One endpoint.

      AIO Proxy sits between your tools and your model providers. Keep the SDKs and coding agents you already use, plug in API keys or the subscriptions you already pay for, and get routing, failover, and full request traces from one local binary.

      Get startedStar on GitHub
      $brew install aio-proxy/tap/aio-proxy
      • Claude Code
        anthropic
      • Codex
        openai-response
      • Cursor
        openai-compatible
      • Gemini SDK
        gemini
      AIO Proxypassthrough
      • Anthropic
      • ChatGPT Pro
      • GitHub Copilot
      • OpenRouter
      • Google Gemini
      inbound API protocols
      7
      built-in subscription plugins
      10
      same-protocol passthrough overhead
      <1 ms
      lines of client code to change
      0

      Protocols

      Your clients keep speaking their language.

      OpenAI, Anthropic, and Gemini clients all hit the same endpoint. When the upstream speaks the same protocol, bytes stream straight through. When it does not, AIO Proxy translates messages, tools, reasoning, and streaming for you.

      Client speaks

      anthropicopenai-response

      Cross-protocol conversion

      Mapped through AI SDK model messages. Tool calls, thinking, multimodal input, and SSE chunks are translated both ways.

      Upstream serves

      Also: Embeddings, Realtime (WebRTC / WebSocket), Images & Audio

      Routing & failover

      Traffic that routes itself — and recovers on its own.

      Ask for a model name. AIO Proxy finds every Provider that serves it, tries higher-priority tiers first, spreads traffic within a tier by weight, and fails over until one answers.

      POST /v1/messagesmodel:claude-sonnet-4-6
      1. 1
        claude-team
        priority 100 · weight 3
      2. 2
        anthropic-key
        priority 100 · weight 1
      3. 3
        openrouter
        priority 0 · weight 1
        standby
      Model-first matching
      Aliases map one name to many upstreams.
      Priority tiers
      Higher tiers first, lower tiers as standby.
      Weighted spread
      Share traffic across keys in the same tier.
      Session affinity
      Long conversations stay on the same Provider.
      ~/.aio-proxy/config.jsonc
      {
  "providers": {
    "claude-team": {
      "kind": "oauth",
      "plugin": "@aio-proxy/plugin-claude-code",
      "capability": "claude",
      "priority": 100, "weight": 3
    },
    "anthropic-key": {
      "kind": "api", "protocol": "anthropic",
      "baseURL": "https://api.anthropic.com/v1",
      "apiKey": "{{env.ANTHROPIC_API_KEY}}",
      "models": ["claude-sonnet-4-6"],
      "priority": 100, "weight": 1
    },
    "openrouter": {
      "kind": "oauth",
      "plugin": "@aio-proxy/plugin-openrouter",
      "capability": "openrouter"
    }
  }
}

      Providers

      Bring the subscriptions you already pay for.

      Log in once with OAuth and turn ChatGPT, Claude, Copilot, and more into standard API endpoints — shared by every tool on your machine.

      ChatGPT
      Plus / Pro / Team
      Claude
      Pro / Team / Max
      GitHub Copilot
      Individual / Enterprise
      Google Antigravity
      Google account
      Cursor
      IDE subscription
      xAI Grok
      Device login
      Kimi Code
      Device login
      OpenRouter
      PKCE login
      OpenCode Go
      OAuth
      Muse Code
      Device login
      Any compatible API
      OpenAI, Anthropic, Gemini, or any gateway that speaks them.
      Any AI SDK provider
      Drop in an @ai-sdk/* package and configure it inline.

      Agents & clients

      One command to wire up your coding agents.

      aiop agent detects installed agents, injects the provider config, and issues a dedicated local token — no copying base URLs between config files.

      Codex, Grok Build, OpenCode, Pi, and OMP configured natively
      Session migration with a rollback journal
      Clean removal that leaves your preferences untouched

      Everything else — Claude Code, Cursor, Cline, Cherry Studio, your own SDK code — just needs a base URL.

      $ aiop agent configure codex
      ✓ Detected Codex CLI 0.42 · profile ~/.codex
      ✓ Injected provider "aio-proxy" into config.toml
      ✓ Issued local agent token (rotates automatically)
      ✓ Migrated 38 sessions · rollback journal saved
      $ aiop agent list
      codex connected opencode connected
      grok connected pi not installed

      Observability

      Know where every request went, and what it cost.

      The built-in Dashboard records every request and every Provider attempt: status, latency, tokens, and cost, with full traces when you need to dig in.

      Requests
      12,480+18%
      Success rate
      99.7%+0.4%
      Tokens
      48.2M+22%
      Cost
      $31.06−9%
      ModelProviderStatusLatencyTokensCost
      claude-sonnet-4-6claude-team2001.8 s14.2k$0.071
      gpt-5chatgpt-pro2002.4 s9.8k$0.000
      claude-sonnet-4-6anthropic-key2000.9 s3.1k$0.016
      gemini-2.5-progoogle2001.2 s6.4k$0.021
      gpt-5-miniopenrouter5020.3 s——
      Trace · claude-sonnet-4-6
      inbound · anthropic
      route · 3 candidates
      claude-team · 429
      anthropic-key · stream
      first token
      Export traces to any OpenTelemetry backend

      Quick start

      Running in under a minute.

      1. 1

        Install

        One binary. Homebrew, Bun, or curl.

        $ brew install aio-proxy/tap/aio-proxy
      2. 2

        Run

        Starts the API and opens the Dashboard.

        $ aiop run --open
      3. 3

        Point your client

        Swap the base URL. Keep the model name.

        $ export OPENAI_BASE_URL=http://127.0.0.1:9317/v1

      Stop juggling keys, endpoints, and SDKs.

      Free and open source. Runs on your machine.

      Get startedGitHub