Skip to content
This repository was archived by the owner on Apr 21, 2026. It is now read-only.

README.md

Latest commit

 

History

History
166 lines (125 loc) · 6.05 KB

README.md

File metadata and controls

166 lines (125 loc) · 6.05 KB

CLIRO

its end here, i have no time to maintain this

-- goodbye!.

CLIRO is a Wails desktop control plane for running a local OpenAI-compatible proxy across ChatGPT Codex and Kiro accounts. CLIRO stands for CLIrouter.

Current release: v0.4.0

Screenshot

Dashboard Accounts
Dashboard Accounts
API Route Usage
API Route Usage

Highlights

  • v0.4.0 architecture refresh:
    • proxy runtime restructured around internal/proxy/codex, internal/proxy/anthropic, internal/proxy/models, and internal/proxy/shared
    • old internal/contract, internal/gateway, internal/route, and internal/protocol/* layers removed
    • Codex runtime simplified into focused execution/payload/runtime files
    • Kiro runtime promoted from partial path into a live routed provider with payload building, event-stream parsing, host fallback, and catalog support
  • OpenAI-compatible and Anthropic-compatible local proxy endpoints:
    • POST /v1/responses
    • POST /v1/chat/completions
    • POST /v1/completions
    • POST /v1/messages
    • GET /v1/models
    • GET /health
    • GET /v1/stats
  • Automatic reasoning/thinking parameter injection via model name suffixes:
    • -low / -minimal → 4096 budget tokens (OpenAI: effort: "low")
    • -medium → 10000 budget tokens (OpenAI: effort: "medium")
    • -high → 16384 budget tokens (OpenAI: effort: "high")
    • -xhigh → 32768 budget tokens (OpenAI: effort: "xhigh")
    • Example: gpt-5.4-high automatically enables extended thinking
  • Cross-protocol reasoning/thinking conversion:
    • OpenAI reasoning.effort ↔ Anthropic thinking.budget_tokens bidirectional mapping
    • Automatic parameter filtering to prevent "Unknown parameter" errors
    • Universal reasoning_content field in responses for client compatibility
  • Multi-account routing for Codex and Kiro with availability-aware scheduling, circuit breaker steps, and cooldown handling.
  • OAuth flows for Codex plus Kiro device auth and Kiro social auth.
  • Token refresh, quota refresh, smart batch quota refresh, and force-refresh-all quota actions.
  • API Router controls for proxy runtime, security, routing policy, endpoint testing, and Cloudflared public access.
  • One-click CLI config sync in API Router for Claude Code, OpenCode, Kilo CLI, and Codex AI, including model selection from local /v1/models catalog.
  • Local CLI auth sync for Kilo, Opencode, and Codex CLI.

Supported Models

  • All models are listed directly in GET /v1/models.
  • Model name suffixes enable automatic reasoning/thinking:
    • Add -low, -medium, -high, or -xhigh to any model name
    • Example: gpt-5.4-high, claude-sonnet-4.5-high
  • Suffix is stripped during routing and converted to appropriate reasoning parameters
  • Current Kiro catalog includes examples such as:
    • auto
    • claude-opus-4.5
    • claude-opus-4.6
    • claude-sonnet-4
    • claude-sonnet-4.5
    • claude-sonnet-4.6
    • claude-haiku-4.5
    • claude-haiku-4.6
    • minimax-m2.5
    • qwen3-coder-next
    • deepseek-3.2

Local Data

CLIRO stores runtime state in ~/.cliro/:

  • config.json - proxy, auth, scheduling, Cloudflared, and UI-facing settings
  • accounts.json - connected account records with token/quota metadata
  • stats.json - usage counters for the local proxy
  • app.log - persistent application log file
  • bin/cloudflared(.exe) - downloaded Cloudflared binary when public access is installed

Development

Prerequisites:

  • Go 1.23+
  • Node.js 18+ and npm
  • Wails CLI v2.11+

Install frontend dependencies:

cd frontend
npm install
cd ..

Run desktop dev mode:

wails dev

Run validation:

cd frontend
npm run check
cd ..
go test . ./internal/...

Build

Build desktop app:

wails build

Windows output:

build/bin/CLIRO.exe

Proxy Base URL

Default base URL:

http://localhost:8095/v1

Example without reasoning:

curl -X POST http://localhost:8095/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.3-codex","messages":[{"role":"user","content":"Hello"}]}'

Example with automatic reasoning (using -high suffix):

curl -X POST http://localhost:8095/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-high","messages":[{"role":"user","content":"Explain quantum entanglement"}]}'

The response will include reasoning_content field with extended thinking.

Notes

  • Authorization mode requires the configured API key for all proxy routes.
  • Cloudflared public access is managed from the API Router tab and depends on the local proxy being online.
  • Smart Refresh All Quotas skips accounts still waiting for quota reset; Force Refresh All Quotas checks every configured account.
  • Kiro runtime uses q.us-east-1.amazonaws.com/generateAssistantResponse first and falls back to codewhisperer.us-east-1.amazonaws.com/generateAssistantResponse on runtime failure.
  • Reliability is now strongest on Codex/OpenAI paths; Kiro is live and routed, with the remaining risk concentrated around image handling, tool-result fidelity, and streaming edge cases.
  • Cross-protocol adapter audit and compatibility coverage are documented in docs/audit-adapter-cross-protocol.md.
  • Model aliasing behavior and examples are documented in docs/feature-model-aliasing.md.

Attribution

  • Codex and Kiro icons/marks remain the property of their respective owners.
  • The CLIRO route app icon uses Icons8 artwork: https://icons8.com/icons/set/route

Release Notes

See CHANGELOG.md for the full release history.