Decido API

Free tier · Decido-1-Fast · last updated 7 November 2026

Send a document and typed questions, get a probability for every answer. The request and response use the same format as decido serve and the Jev 1.13 API, so the same code runs against this free hosted API, against your own server, or locally with pip install decido-ai.

The hosted API serves one model, Decido-1-Fast. The other five Decido models are open weights: run any of them on your own machine with pip install decido-ai, or email founders@getdecido.ai for hosted access to the larger models.

Get a free API key

New to Decido? The 5-minute quickstart covers the Python client, your own server, the hosted API and agents (MCP, LangChain, OpenAI tools).

The model

Endpoint

POST https://getdecido.ai/v1/systemone
Authorization: Bearer dk_…
Content-Type: application/json

GET https://getdecido.ai/v1/models (no key needed) returns models, the models this API serves (today Decido-1-Fast), and open_weights, every Decido model with its download link, to run on your own machine with pip install decido-ai.

Authentication

• Send your key as Authorization: Bearer dk_…. That is the header the Jev / TypeSafe SDKs send.
• Get a key on the playground: sign in with Google (we ask only for your email address from it), one account per person, one active key per account. The key is shown once; we store only its SHA-256 hash, so a lost key cannot be recovered, so create a new one (the old one stops working; up to 5 new keys a day).
• The free API is for evaluation: no SLA, small daily limits. For real workloads run Decido yourself (pip install decido-ai && decido serve: same request format, no limits) or contact founders@getdecido.ai.
• Keys created before sign-in existed (without an account) stop working on 23 November 2026.
• Revoked or unused keys are deleted 90 days after last use. Keys used for abuse may be revoked.
• Call the API from your server only. There are no CORS headers on purpose: a key in browser code is public.

Request

The free API takes plain-text instructions; structured (object) instructions return 422. They work with decido serve.

curl -s https://getdecido.ai/v1/systemone \
  -H "Authorization: Bearer $DECIDO_API_KEY" -H "Content-Type: application/json" -d @- <<'EOF'
{
  "state": "Customer: I was charged twice for order 4821. Please refund one of the charges.",
  "questions": {
    "refund": {"type": "noul", "instructions": "Is the customer asking for a refund?"},
    "team": {"type": "choice", "instructions": "Which team should handle it?",
             "criteria": {"billing": "Charges, refunds, invoices", "shipping": "Deliveries", "tech": "Bugs and outages"}},
    "urgency": {"type": "score", "instructions": "How urgent is it?", "criteria": ["low", "normal", "urgent"]}
  }
}
EOF

Response

{
  "model": "decido-1-fast",
  "answers": {
    "refund":  {"type": "noul", "noul": 0.97},
    "team":    {"type": "choice", "choice": "billing", "probabilities": {"billing": 0.92, "shipping": 0.03, "tech": 0.05}, "confidence": 0.9},
    "urgency": {"type": "score", "score": 0.94, "legend": {"0": "low", "1": "normal", "2": "urgent"},
                "probabilities": {"0": 0.3, "1": 0.47, "2": 0.24}, "confidence": 0.2}
  },
  "usage": {"input_tokens": 212, "output_tokens": 0}
}

noul is P(yes). choice is the most likely option, with a probability for every option. score is the expected level (0 = first level), with a probability per level. Example values; yours will differ.

Limits (free tier)

Successful responses, and 429 responses for the daily quota or the per-minute rate, carry X-RateLimit-Limit, X-RateLimit-Remaining (requests left today) and X-RateLimit-Reset (Unix time of the next reset). 429 and 503 responses carry Retry-After (seconds).

Errors

Errors are JSON: {"detail": "…"}.

Jev / TypeSafe SDK

Point the client at Decido and use your Decido key; the SDK's zstd-compressed bodies (it compresses requests of about 1 KB and up) are accepted. Model names such as jev-latest run on Decido-1-Fast.

export TYPESAFE_API_KEY="$DECIDO_API_KEY"

from typesafe_sdk import TypeSafeClient, Noul
with TypeSafeClient(base_url="https://getdecido.ai") as client:
    r = client.system_one(state="Customer: I was charged twice for order 4821.",
                          questions={"refund": Noul(instructions="Is the customer asking for a refund?")})
    print(r.answers["refund"].noul)

Privacy

Documents, questions and answers are processed in memory and not stored or used for training. Request metadata (time, status, size, key prefix, country) is kept for up to 30 days. For your account we keep the sign-in provider, its user id and your email address; for each key its SHA-256 hash, a short display prefix, your label, and when it was created and last used. Details: privacy, terms.

Production use

The free tier has no SLA and small limits. For production volume, bigger models or sensitive data, run Decido yourself. It takes the same requests, has no limits, and nothing leaves your machine:

pip install decido-ai
decido serve decido-ai/decido-1-26b-a4b --port 8080      # or decido-1-fast-1.7b, decido-1-max-26b-a4b, …
curl -s localhost:8080/v1/systemone -H 'content-type: application/json' -d @request.json