Forager
Open source · Cloudflare Workers

Free LLM tiers, foraged into one endpoint.

Forager gathers the free API tiers of 18 AI providers behind a single OpenAI-compatible URL on your own Cloudflare account, and counts every quota so you never pay a cent.

Deploy to Cloudflare Star– Read the field guide ↓

Runs on the Workers Free plan. You bring your own provider keys.

Specimen Nº 001POST /v1/chat
$ curl https://forager.you.workers.dev/v1/chat/completions \
  -H "Authorization: Bearer $FORAGER_KEY" \
  -d '{"model":"auto","messages":[{"role":"user","content":"hi"}]}'
HTTP 200 · x-routed-via: groq/openai/gpt-oss-120b
quota left this cycle 912 / 1,000 req
$0What it costs you
18Free providers in the catalog
1OpenAI-compatible URL
4Quota cycles tracked: min · hr · day · month
I. FIELD NOTES

How it works

Every free tier has its own limits, reset times and fine print. Forager keeps the ledger for you and only takes what's there.

Gather

Add a key for each free provider in the dashboard: Groq, Gemini, NVIDIA, Mistral, OpenRouter and more. Keys are encrypted and never shown again.

Route

Ask for auto, auto:coding or a specific model. Forager picks the best model with quota left and fails over on rate limits, outages and timeouts.

Stay at $0

It counts requests, tokens and dollars per key in every cycle, stops at 90% of each limit, blocks paid models, and parks any key that hits a billing error.

II. THE KIT

One key, every way in

Your Forager API key works everywhere the endpoint does: a chat on your phone, the coding tools on your laptop, and a dashboard that shows where every request went.

Chat

Open /chat and sign in with your key. Pick auto or any model, watch answers stream in with the provider that served them, and add it to your home screen: it installs like an app, and your chats can sync across devices.

Code

Point OpenCode, the OpenAI SDKs or any OpenAI-compatible tool at /v1. Ask for auto:coding, auto:vision or auto:structured and Forager picks a model that can do the job.

Watch

The dashboard shows each quota filling up, what's cooling down, and a 30-day history of every request: which key sent it, which model answered and what failed on the way. No message content is kept. It works on your phone too.

III. SPECIMENS

What's in the catalog

Free tiers change constantly. The catalog is data, not code: Forager refreshes OpenRouter's free list daily, and you can update limits without redeploying.

tested checked live, September 2026 · docs from provider docs, not yet tested live · card on file free tier unlocks with a payment method; you're never charged under quota

IV. THE TRAIL

Self-host in three steps

Forager runs entirely on your own Cloudflare account. Nobody else sees your keys or your prompts.

  1. Deploy to Cloudflare

    Click the button and pick a project name. Cloudflare copies the repo to your GitHub and creates the Worker and its Durable Object. There's nothing to fill in.

  2. Create your key, add providers

    Open /dashboard on your new Worker and click Create my API key. Save it, then add a key for each free provider you use. Each one is tested as soon as you add it.

  3. Point your apps at it

    Use any OpenAI SDK or tool with your Worker's /v1 URL, or open /chat. The x-routed-via header and the dashboard's request history show which provider answered.

Python · OpenAI SDK
from openai import OpenAI

client = OpenAI(
    base_url="https://forager.you.workers.dev/v1",
    api_key=FORAGER_KEY,
)
reply = client.chat.completions.create(
    model="auto:coding",
    messages=[{"role": "user", "content": "Write a haiku about moss"}],
)
Or from the command line
# clone and deploy, then open /dashboard
git clone https://github.com/volkanger/forager
cd forager && npm install
npx wrangler deploy
V. ETIQUETTE

Forage responsibly

Use your own keys

One key per provider, used only by Forager. Stacking free accounts to multiply quota usually breaks a provider's terms.

Free tiers move

Limits and free models change without notice. Forager learns from errors and rate-limit headers, but check each provider's console now and then.

Read the fine print on data

Some free tiers (Gemini, Mistral, Kilo, Token Harbor) may use prompts for training, and NVIDIA logs requests. Don't send anything private to them.

Keep billing off

No billing account on the Gemini project, auto-reload off on Ollama, and never buy credits on card-on-file providers. That's what keeps free at free.