# Gödel Machines Inference

> An OpenAI-compatible API with free, student and paid access for DeepSeek V4.1 Flash, served from our own GPUs in Hyderabad, India.

- Base URL: https://api.goedelmachines.com/v1
- Model id: deepseek/deepseek-v4.1-flash (deepseek-v4.1-flash is also accepted)
- Context: 1M-token context (1,048,576 tokens). Input and output share this window; plan output limits are listed below.
- Endpoints: POST /chat/completions (streaming, tool calling, reasoning), GET /models
- Auth: Authorization: Bearer <your API key> (free and paid keys use the same endpoint)

Free tier data policy: prompts and completions sent with free keys are kept, people at Gödel Machines may read
them, and they may be used (separated from the account) to train models. Do not send secrets or personal data.
Policy: https://inference.goedelmachines.com/legal#free-tier

## Plans, prices and retention

Free and Student keys retain prompts/completions; they may be read and used for training.
Student pass: eligible university email accounts; shared free-tier capacity and priority.
Paid keys: zero prompt/output retention; metadata-only accounting. Buyers in India can add prepaid credit
from ₹100 before GST; credit never expires. Paid access outside India is coming soon.
INR per million tokens: input ₹12, cached input ₹0.25, output ₹50, plus 18% GST.
- free: 2 concurrent requests, 60 requests/minute; 10,000,000 new input tokens/day, 1,000,000 output tokens/day; max output 235,929 tokens/request; prompts and outputs retained.
- paid: 8 concurrent requests, 300 requests/minute; no daily cap new input tokens/day, no daily cap output tokens/day; max output 235,929 tokens/request; zero prompt/output retention.
- student: 4 concurrent requests, 120 requests/minute; 30,000,000 new input tokens/day, 3,000,000 output tokens/day; max output 235,929 tokens/request; prompts and outputs retained.
Cached input does not consume the new-input allowance. Daily limits reset at 00:00 UTC.
Compare plans without signing in: https://inference.goedelmachines.com/pricing

## Connect a coding agent in one command

curl -fsSL https://inference.goedelmachines.com/install.sh | sh

Configures OpenCode, Codex CLI, Claude Code, Aider, Crush, Goose, Continue, oh-my-pi and pi (and prints steps for
Kilo Code, Cline, Roo Code), getting a key (free or paid, chosen in the browser) by device login. Flags: --dry-run, --list, --only h[,h], --yes,
--new-key (sign in again for a new key, e.g. paid), --uninstall. The key type (free or paid) shows in the model name
(OpenCode, Claude Code, Crush, Continue, oh-my-pi, pi) or the provider name (Codex CLI, Goose); not in Aider, Kilo Code,
Cline or Roo Code. Claude Code runs with its own settings file: claude --settings ~/.config/goedel/claude-code.json

## Get a key without a browser (device login)

An agent can get a key for its user in three steps, like `gh auth login`. The user must approve in a browser.

1. POST https://inference.goedelmachines.com/api/device/code
   Body (optional): {"client_name": "my-agent"}   (1-60 characters; shown to the user)
   200: {"device_code": "...", "user_code": "ABCD-1234", "verification_uri": "https://inference.goedelmachines.com/device",
         "verification_uri_complete": "https://inference.goedelmachines.com/device?code=ABCD-1234", "expires_in": 600, "interval": 5}
   Keep device_code secret. Show the user verification_uri_complete (or verification_uri and user_code).

2. The user opens the link, signs in (Google or GitHub), and approves a free key (prompts stored; accepting the
   free-tier terms if new) or a paid key (zero data retention; needs a positive credit balance).

3. POST https://inference.goedelmachines.com/api/device/token every `interval` seconds
   Body: {"device_code": "..."}
   200: {"api_key": "sk-gm-free-...", "base_url": "https://api.goedelmachines.com/v1",
         "model": "deepseek/deepseek-v4.1-flash", "key_name": "my-agent", "tier": "free"}
   tier is "free" or "paid", whichever the user approved; the key's prefix matches (sk-gm-free- / sk-gm-paid-).
   The key is returned once. Store it; it cannot be fetched again.
   400 {"error": ..., "error_description": ...} while not done:
     authorization_pending  keep polling at the interval
     slow_down              polling too fast; use the "interval" in the answer (it grows by 5 s)
     key_limit              the user has 5 keys; they must revoke one at https://inference.goedelmachines.com/keys (polling may continue)
     no_credit              a paid key was approved but the balance ran out; they must add credit at https://inference.goedelmachines.com/billing
                            (polling may continue)
     access_denied          the user declined; stop
     expired_token          10 minutes passed; start again
     invalid_grant          unknown code or key already collected; start again
     invalid_request        malformed body
   429 rate_limited (honour Retry-After). Show error_description to the user.

Shell:

    SITE=https://inference.goedelmachines.com
    resp=$(curl -s -X POST $SITE/api/device/code \
      -H "Content-Type: application/json" -d '{"client_name": "my-agent"}')
    echo "Open $(echo "$resp" | jq -r .verification_uri_complete) and approve"
    code=$(echo "$resp" | jq -r .device_code)
    wait=5
    while sleep $wait; do
      out=$(curl -s -X POST $SITE/api/device/token \
        -H "Content-Type: application/json" -d "{\"device_code\": \"$code\"}")
      key=$(echo "$out" | jq -r '.api_key // empty')
      [ -n "$key" ] && break
      case $(echo "$out" | jq -r .error) in
        authorization_pending|key_limit|no_credit) ;;
        slow_down) wait=$(echo "$out" | jq -r .interval) ;;
        *) echo "$out" | jq -r .error_description; exit 1 ;;
      esac
    done
    export GOEDEL_API_KEY=$key

## Calling the API

    curl https://api.goedelmachines.com/v1/chat/completions \
      -H "Authorization: Bearer $GOEDEL_API_KEY" -H "Content-Type: application/json" \
      -d '{"model": "deepseek/deepseek-v4.1-flash", "messages": [{"role": "user", "content": "Hello!"}]}'

OpenAI SDK: OpenAI(base_url="https://api.goedelmachines.com/v1", api_key=key), model="deepseek/deepseek-v4.1-flash".

Reasoning is on by default and arrives in message.reasoning (delta.reasoning when streaming). Turn it off with
"chat_template_kwargs": {"thinking": false} in the request body (extra_body in the OpenAI SDK).

## Limits and errors

Plan limits are public at https://inference.goedelmachines.com/pricing#limits; your own usage is available after sign-in. 429 means over a limit or at capacity:
wait for Retry-After, then retry with backoff. 401: missing or invalid key. 413: body over 32 MiB (33,554,432 bytes) on the public endpoint (full JSON including base64 images).
The browser Playground has a separate 4,000,000-byte request limit. Use the API directly for larger requests.
Cached DNS connections can temporarily retain the previous API limit after an update; reconnect and allow DNS to refresh.

## More

- Docs: https://inference.goedelmachines.com/docs
- Models (JSON): https://api.goedelmachines.com/v1/models
- Help: hi@goedelmachines.com