Point Claude Code, Codex or Cursor at one endpoint. merido pools every account and free tier you already pay for, routes each request to the cheapest model that can do the job, fails over when one gets throttled — and shows you the dollars it saved.
# install, then run — that's the whole thing
curl -fsSL https://merido.dev/install.sh | sh
merido start # → http://127.0.0.1:8788Works with your stack
Your AI coding spend leaks in three predictable places. merido plugs each one behind a single endpoint — here's the before and after.
Token spend balloons and you can't see which calls are wasteful or which model is overkill for the task.
Each request routes to the cheapest model that can do the job, and bloated tool_result output is compressed losslessly.
One account throttles and you're blocked mid-task, watching a retry timer while the work just waits.
Every account and free tier sits behind one name, and a four-tier fallback keeps you coding the moment one provider blinks.
Subscriptions, API keys and free tiers sit idle because your CLI only ever knows one endpoint at a time.
merido fans out across every provider and account you own, so the capacity you already pay for finally gets used.
The Token-Optimization Advisor reads your real traffic per CLI and project, finds the waste, and suggests concrete fixes you apply in one click. Every fix lands in a savings ledger, so you can see exactly how many tokens and dollars merido saved you, and whether it paid for itself.
tool_result output, losslessly.tool_result · claude-codeIllustrative example — merido reports the real numbers from your own traffic.
One static binary with embedded SQLite. Run two commands and the API plus the dashboard come up on a single local port — no toolchain, no dependencies.
Set your tool's base URL to merido and drop in a key. Claude Code, Codex, Cursor, Cline and Continue all work out of the box — nothing else changes.
merido pools your accounts, routes each call to the cheapest capable model, fails over automatically, and books every saving into the ledger.
Under the hood: one translation core, every wire format, streaming kept intact. Read the architecture →
Spread requests across every account and free tier you have. merido routes by cost and latency, balances load, dedupes identical calls, serves a semantic cache, and runs a four-tier fallback chain that keeps you coding when a provider blinks.
Circuit breakers, active & passive health checks, per-key rate limits, smart retry.
Images, embeddings, audio (TTS + STT), sandboxed MCP and WASM plugins.
OpenTelemetry, Prometheus, a queryable audit log, and a live request stream.
Encryption at rest for every credential, argon2 auth, and a BYOK vault. Your keys never leave your machine.
Define a client-callable model that fans out to an ordered list of real
provider/model targets. merido rotates and fails over on its own,
while your CLI keeps calling one name.
strategy: failover
targets:
- anthropic/claude-sonnet
- openrouter/claude-sonnet
- gemini/gemini-2.5-pro
# your CLI just calls:
model: "smart-sonnet"
One command installs the merido binary; the next starts it. A single
binary serves the API, the dashboard, and this page from one port — no toolchain,
no dependencies.
# 1 · install merido
curl -fsSL https://merido.dev/install.sh | sh
# 2 · run the gateway (embedded SQLite)
merido start
→ http://127.0.0.1:8788# 1 · install merido
irm https://merido.dev/install.ps1 | iex
# 2 · run the gateway (embedded SQLite)
merido start
→ http://127.0.0.1:8788
Then merido keys create, merido providers add, and
merido gain to see what you've saved.
Estimate your real monthly Claude Code spend and see how much merido can save you.
Proven tactics that cut Claude Code token spend without slowing you down.
Deep dives on AI cost control, routing strategy, and getting more from every token.
A local-first AI gateway written in Rust. It sits between your AI coding tools (Claude Code, Codex, Cursor, Cline, Continue) and 40+ upstream LLM providers. It translates formats, compresses tool output to save tokens, and fails over so one outage never stops you.
merido speaks an OpenAI-compatible API and routes to 40+ providers including OpenAI, Anthropic, Google Gemini, OpenRouter, GLM and Kiro. Any CLI targeting an OpenAI or Anthropic endpoint works out of the box.
No. merido is a proprietary product. You download a single static binary and self-host it, bringing your own provider keys and your own billing. Every feature ships in that one binary.
It analyzes your real usage, detects token waste per CLI and project, and recommends concrete changes like compressing tool results, switching to a cheaper model, or enabling prompt caching. You apply them behind a guarded probation window with auto-rollback.
Yes. The default local profile runs as a single binary with embedded SQLite, serving both API and dashboard from one port (8788). A cloud profile adds Postgres, Redis and multi-tenancy when you scale.
A clean-room reimplementation in Rust with a single translation core, a built-in Token-Optimization Advisor, and one self-hosted binary that serves both the API and a full operator dashboard. It's Helicone-header compatible, so existing clients point at it directly.
one binary · your keys, your billing
Every feature ships in a single binary you self-host. Bring your own keys.