meridoThe local-first AI gateway

Route any coding CLI to 40+ LLM providers. Save tokens, fall back across accounts, and watch every request — from one Rust binary.

Quick start

What is merido?

OpenAI-compatible, drop-in

Point Claude Code, Codex, Cursor — anything that speaks the OpenAI, Anthropic, or Gemini wire format — at merido. It translates every format through one canonical representation.

One binary, batteries included

The API, the React dashboard, and this documentation site all ship inside a single executable on one port. No sidecars, no external services for the local profile.

Token saving that pays for itself

RTK-style tool-output compression and optional Caveman mode cut tokens on every request, with a savings ledger that proves the gateway's worth.

Routing & fallback

Virtual models pick the best capable target by cost, latency, and health, and fall back across accounts and providers when one fails.

Caching, two ways

Exact request dedup plus an opt-in semantic cache with a precision guard, so a fuzzy hit can never quietly return the wrong answer.

The Token-Optimization Advisor

merido watches your traffic and tells you where you are overspending — then applies the fix behind a probation window with auto-rollback.