Free APIs & Chaining

The goal: effectively unlimited compute, for free. No single provider gives you infinite free usage — but if you chain several free providers, Orb falls through to the next one whenever one runs out. Stack three or four and you rarely hit a wall.

Free tiers, rate limits, and model names change constantly. Treat every list on this page as a starting point and check each provider’s own docs for current limits. This is a community wiki — if something’s out of date, please send a PR.

How chaining works

Orb tries each brain in order and skips any that are rate-limited or down (for a short cooldown), then uses the next. The on-device model is always last, so you always get an answer:

free API #1  →  free API #2  →  free API #3  →  on-device floor

Order them by what you want first (fastest, or highest-quality) — Orb handles the rest.

Where you configure it

In the app: Settings → add a provider, paste a key, set the order. Keys are stored in your device keychain. (Easiest — start here.)
On your own server: set the ORB_FREE_BACKENDS environment variable (a JSON list). See the format below. (For self-hosters — see Self-Hosting.)

Any provider that speaks the standard OpenAI Chat Completions format works — which is almost all of them.

Commonly used free / free-tier providers

Provider	OpenAI-compatible base URL	Notes
Groq	`https://api.groq.com/openai/v1`	Very fast; generous free tier; open models (Llama, etc.)
Google AI Studio (Gemini)	`https://generativelanguage.googleapis.com/v1beta/openai/`	Large free tier; capable models
Cerebras	`https://api.cerebras.ai/v1`	Extremely fast; free tier
OpenRouter	`https://openrouter.ai/api/v1`	Aggregator; many models tagged `:free`
Mistral	`https://api.mistral.ai/v1`	Free developer tier
GitHub Models	see GitHub Models docs	Free for developers
Cloudflare Workers AI	see Cloudflare docs	Free daily allowance
Local (your own hardware)	`http://<your-pc>:<port>/v1`	Truly free + private — see Local Inference

Get a key from the provider (usually a free signup), then add it in the app or on your server.

Server config format (`ORB_FREE_BACKENDS`)

A JSON list, in priority order. Each entry:

[
  {
    "name": "groq",
    "base_url": "https://api.groq.com/openai/v1",
    "model": "llama-3.3-70b-versatile",
    "api_key_env": "GROQ_API_KEY"
  },
  {
    "name": "gemini",
    "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
    "model": "gemini-2.0-flash",
    "api_key_env": "GEMINI_API_KEY"
  }
]

name — a label for logs.
base_url — the provider’s OpenAI-compatible endpoint (no trailing /chat/completions; Orb appends it).
model — the model id to request.
api_key_env — the name of an environment variable that holds the key (so the key itself never goes in the config). Set e.g. GROQ_API_KEY=... alongside it. Omit this field entirely for a keyless local endpoint.

A provider whose key isn’t set is automatically skipped — so you can list several and only fill in the ones you’ve signed up for.

Tips for the most free compute

Mix fast + capable: put a fast provider (Groq, Cerebras) first for snappy replies, a more capable one next for harder questions.
Add a local endpoint last (before on-device): if you have a PC that can run a model, it’s free and private — see Local Inference.
Spread across providers: several small free tiers chained together outlast any single one.
Keep keys safe: never paste a key into a public place. In the app they live in the keychain; on a server they live in environment variables, never in the repo.