TierMuxThe free tier, multiplexed.
TierMux verifies which providers still serve which models for free, then builds on that catalog: a free agentic coding assistant for VS Code that fails over mid-sentence, and a self-hosted gateway that runs on your own Cloudflare account — your keys, your quota, no telemetry.
One VSIX runs in all of them
One catalog. Three products.
Everything TierMux ships is free and built on the same verified data — the catalog is the common data layer.
TierMux for VS Code
A free agentic AI coding assistant. No subscription, no per-token bill.
- Ask / Plan / Agent modes, with approvals, diffs, checkpoints and undo
- Runs in VS Code and every VS Code fork — Cursor, Windsurf, Google Antigravity, VSCodium: one VSIX for all of them
- Keyless out of the box — Kilo Gateway, OpenCode Zen, OVH AI Endpoints and Pollinations: no key, no account, no config
- Post-edit verification runs your tests and feeds failures back into bounded fix rounds
- Keys stay in VS Code's encrypted storage; requests go VS Code → provider, directly
TierMux Gateway
A self-hostable LLM gateway. User ownership is the feature.
- Runs on your Cloudflare account (Workers + D1, one click) or Docker / Node + SQLite
- Speaks OpenAI, Responses and Anthropic over one endpoint — Claude Code, Codex CLI, Cline, Zed
- Key rotation, circuit breakers, aliases, and an
automodel that ranks on observed facts - Prompts and keys never leave your install — the only outbound call is the public catalog
The catalog
The data layer both products are built on.
- The whole free tier in one list — every provider is verified with real requests before it lands here; dead entries are removed, not cached
- Capability flags, context windows, rpm/rpd limits and prices per model
- Re-synced every five minutes, with a health guard that benches failing sources
- Public JSON API — CORS open, no key, already speaking the gateway's config format
What free actually gives you
Not a trial, and not credits that run down to zero. These are the providers' own free tiers — real, recurring quotas — and both products exist to spend them in the right order. Same catalog behind each; what differs is whose keys are in play.
In VS Code
- Your first message costs nothing and needs no account. Four providers answer keyless, so there is no signup between installing and typing.
- Add a free key and that provider's quota is yours too. Several keys per provider, rotated — unused ones simply rest.
- Spent quota is not a wait. A 429 moves the turn to the next provider mid-sentence instead of parking you until the window resets.
- Nothing is metered by us. The Usage tab counts your own lifetime tokens and what the same turns would have cost on a paid API.
Through your own endpoint
- The same catalog, but the keys are yours — and one pool serves every tool you point at it: Claude Code, Codex CLI, Cline, Zed.
- Headroom multiplies with the pool. A model capped at 1,000 a day with three keys behind it is 3,000 a day, and a spent key goes to the back before it costs a 429.
- A 429 parks the key, not the provider. Rate limiting is a property of the credential; circuits open per provider and per model, separately.
- No per-token bill, and no hosting bill — a gateway this small sits inside Cloudflare's free plan.
Limits are per key, per provider — adding one free key beyond the keyless four is the single biggest jump you can make, and a second key on whichever provider you lean on hardest is the next. How to add them →
Model directory
A live view of the catalog, frontier models first. Every listing is verified with real requests — capabilities are measured, not copied from a pricing page.
Showing — of — matching models · full directory in the docs · raw JSON at /models
Where it sits
Not a cheaper version of a paid assistant — a different premise. The free tier is the product, so everything is built to survive its limits.
| TierMux | Copilot | Cursor | Cline / Kilo | |
|---|---|---|---|---|
| Usable every day at $0 | yes | no1 | no2 | no3 |
| First message with zero keys or accounts | yes | no | no | no |
| Pools many providers' free tiers, auto-failover | yes | no | no | no |
| Several keys per provider, rotated | yes | no | no | no |
| Any OpenAI-compatible endpoint | yes | partial | partial | yes |
| Agent and plan modes, MCP | yes | partial | yes | yes |
| Open source, request goes straight to the provider | yes · MIT | no | no | yes |
1. Copilot Free caps at 50 chat requests a month, with usage-based billing since June 2026. 2. Proprietary fork; real usage sits behind paid tiers. 3. Free software, but every request bills your own API tokens.
What is the catch? Why is this free?
There is no TierMux account, no server of ours in the request path, and nothing to upsell. The extension talks straight to the provider using free tiers those providers publish themselves; this site only tells it which ones currently work.
The cost you carry is the free tier's own: slower first tokens, lower rate limits, and models weaker than the paid frontier. Pooling providers is what makes that workable, not a way around it.
Do I need an API key to start?
No. Four providers ship keyless — Kilo Gateway, OpenCode Zen, OVH AI Endpoints and Pollinations — so the first message works with no key, no account and no config.
Adding free keys for two or three more providers is the single biggest upgrade you can make: more keys means more headroom before a wall, and unused keys simply rest.
Where do my prompts and keys go?
Keys live in VS Code's encrypted secret storage. Requests go VS Code → provider, directly — there is no TierMux server in the path, and no telemetry to switch off because none was written.
Chats are JSON files in the extension's workspace storage on your own disk. The gateway keeps the same rule on your own infrastructure: prompts, responses and keys never leave the deployment.
Does it work in Cursor, Windsurf or VSCodium?
Yes — one VSIX runs in VS Code and every VS Code fork. The only wrinkle is which registry your editor reads: VS Code uses the Marketplace, the forks generally read Open VSX. Both serve the identical build, and Extensions: Install from VSIX takes it in any of them.
Extension or gateway — which one do I want?
The extension if you want to code in VS Code and be done. It needs nothing deployed.
The gateway if you want one endpoint your other tools can share — Claude Code, Codex CLI, Cline, Zed — with your keys pooled behind it and routing decided once, centrally. It runs on your own Cloudflare account or your own hardware. Plenty of people run both.
How current is the catalog?
It re-syncs every five minutes, with a health guard that benches a source that starts failing. Free tiers are volatile — a promotion can appear and vanish inside a day — so the catalog is built to remove dead entries rather than cache them.
Whatever this page shows came from the live API a moment ago. You can read the same data yourself at /models.