Playground

Give Decido a document and ask it questions. It answers each one with a probability, so your agent knows when to act and when to ask a person. No signup; inputs are not stored.

Input

Start from:

Example answers below ↓

Document text or JSON

Questions yes/no · pick one · scale

One request as JSON (the same format as decido serve and the Jev 1.13 API)

type: noul (yes/no), choice (criteria: option → description), score (criteria: levels, lowest first).

Batch files run locally: decido run decido-ai/decido-1-fast-1.7b --request file.json (one request per file), or Decido(...).predict_batch([...]) in Python.

Limits & uploads

Online demo limit: about 2,500 characters of English in total (document + questions). No limit when you run it locally.

Upload a file reads a local .txt, .md, .csv, .json, .eml or .html file (up to 64 KB) into the Document box; nothing is sent until you press Decide.

Answers

Act on its own when at least90% sureBelow that, it asks a person.

Pick an example or write your own and press Decide.

Response JSON · what your code gets back

Model: Decido-1-Fast (1.7B), the free online demo. Larger models (Decido-1, Decido-1-Think, Decido-1-Max) run on your own machine (see below).

Run it on your machine free, no limits

The same code runs every Decido model on your own machine: nothing leaves it. New here? 5-minute quickstart →

Call the hosted API free key, 25 requests/day

A free key to try the same hosted model from your code: 25 requests/day · Decido-1-Fast · demo limits (up to 5 questions, about 2,500 characters). It's for evaluation; for real workloads run Decido yourself with decido serve (same request format, also the Jev 1.13 API's). API docs →

Label (optional, for you)

Call it from your server, never from browser code (the key would be public).

Test a key · sends the request above once and uses 1 of today's 25 requests

Bigger models run on your laptop

The online demo runs the small Decido-1-Fast (six-benchmark score 73.0). For judgment calls use the larger open models locally, with the same code (speeds on an M4 Pro, 1k-token document):

Decido-1: 85.4 (Jev 1.13: 83.9); ~1.1 s per 5-question request on the Mac GPU, ~4 s on CPU: Decido("decido-ai/decido-1-26b-a4b")
Decido-1-Think: 88.7, reasons before answering; ~10 s per question on the Mac GPU: Decido("decido-ai/decido-1-think-26b-a4b")
Decido-1-Max: 89.2, our most accurate; ~11 s per question on the Mac GPU, ~28 s on CPU: Decido("decido-ai/decido-1-max-26b-a4b")