# Inferiq

> An OpenAI-compatible gateway for coding tools with fix agents that turn errors into a cause and a patch, paid per call from a prepaid balance.

Inferiq provides one API key and an OpenAI-compatible endpoint for coding tools, plus fix agents that take an error and return a cause and a patch. It is in early access, with the gateway and the Quick Fix and Build Doctor agents live. Usage is paid per call from a prepaid US dollar balance, and every charge produces a receipt.

## What It Is

Inferiq sits between a coding tool (Cline, OpenCode, Cursor, Claude Code or a custom script) and the model. The tool only needs a base URL and an Inferiq key. The gateway checks the key, its allowed models, rate and daily limits and the balance, then routes the request, retries a failed attempt once, meters the usage and writes a receipt. It serves Qwen3.8-27B hosted by Inferiq through three aliases (inferiq/fast, inferiq/balanced, inferiq/best) that set reasoning depth.

## How Fix Agents Work

Fix agents are specialists for particular failures. Quick Fix takes an error message, stack trace and related files and returns the cause, a patch and a way to check it. Build Doctor handles install, build and deploy failures (npm, pnpm, Next.js, Vite, Docker, Vercel) using the build log, package.json, lockfile and config files. Agents return a patch and never run commands on the user's machine. Each run is quoted with a hard maximum that the user approves first. Type Fixer, Dependency Doctor, Test Fixer and Deep Debug are listed as upcoming.

## Receipts and Balance

Users deposit ETH or USDG on Robinhood Chain, or ETH or USDC on Base, into one balance. Each gateway request and fix writes a receipt with model, tokens, cost and balance before and after, fetchable by ID via GET /v1/receipts/{id}. Keys are free, can be limited per tool to chosen models, and can carry daily limits. According to the site, files are kept only until a run ends and results are deleted after 24 hours.

## Connecting Tools

OpenAI-compatible tools use the /v1/chat/completions endpoint with streaming and tool calls. Claude Code connects through an Anthropic-format /v1/messages endpoint, and editors can call fix agents through an MCP server at /v1/mcp. A CLI and more agents are listed as in progress on the roadmap.

## Features
- OpenAI-compatible /v1/chat/completions with streaming and tool calls
- Anthropic-format endpoint for Claude Code
- Fast, balanced and best reasoning aliases on Qwen3.8-27B
- Fix agents: Quick Fix and Build Doctor
- Quote with hard maximum before each fix run
- Receipt for every request and fix
- Free per-tool API keys limited to chosen models
- Per-key daily limits and monthly budgets
- Prepaid crypto balance in USD
- MCP server for calling fix agents from editors
- Model fallback and one retry on failed attempts

## Integrations
Cline, OpenCode, Cursor, Claude Code, OpenAI SDK, MCP, Robinhood Chain, Base

## Platforms
WEB, API, CLI

## Pricing
Paid

## Links
- Website: https://inferiq.run
- Documentation: https://inferiq.run/docs
- EveryDev.ai: https://www.everydev.ai/tools/inferiq