Quickstart

Sieve serves the same System One API as TypeSafe. If you already use a TypeSafe SDK, you change two things:

• Base URL: https://api.sieve.one
• Key: your Sieve API key, sent as Authorization: Bearer <key>

You need: an email address or a Google, GitHub or Hugging Face account, and one of: Python 3.10 or newer, Node.js 22 or newer, or curl. The versions tested against Sieve are typesafe-sdk 0.7.5 (Python) and @typesafe-ai/sdk 0.7.0 (JavaScript).

1. Get a keyCopy Anchor Link

Sign up at dashboard.sieve.one with an email and password, or with Google, GitHub or Hugging Face. A new account goes through these screens once:

1. Verify your email. Open the link we send. It works for 24 hours.
2. Review our terms. Agree to the Master Customer Agreement and the Privacy Policy.
3. Let's create your org. Enter the organisation name, billing country (it sets the currency and cannot be changed later), your GSTIN if you are in India (optional), and company size.
4. How did you hear about us? Optional. Skip it or submit.

The console then opens. Go to API keys, create a key and copy it. The full key is shown once. See Authentication, and Sign-in and account security to turn on two-factor authentication or add a passkey.

export TYPESAFE_API_KEY="sieve_app_live_..."
export TYPESAFE_BASE_URL="https://api.sieve.one"

Both TypeSafe SDKs read these two variables, so the code below needs no key or URL in it.

2. Pick a modelCopy Anchor Link

curl https://api.sieve.one/v1/models \
  -H "Authorization: Bearer $TYPESAFE_API_KEY"

Use a name from the list. sieve-latest is an alias that moves to the newest recommended version, and sieve-auto runs the cascade. The response always reports the versioned model that answered. See the model cards.

3. Ask a questionCopy Anchor Link

curl https://api.sieve.one/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sieve-latest",
    "state": "I was charged twice. Please fix this ASAP.",
    "questions": {
      "billing": { "type": "noul", "instructions": "Is this message about billing?" }
    }
  }'

4. Read the responseCopy Anchor Link

{
  "model": "sieve-9b-v2",
  "answers": {
    "billing": { "type": "noul", "noul": 0.98, "stats": {} }
  },
  "usage": { "input_tokens": 120, "output_tokens": 12 },
  "assets_used": null
}

• model: the versioned model that answered, never the alias you sent (the value shown is an example).
• answers: one answer per question name. Each answer's type matches its question.
• usage: input_tokens are billed. output_tokens are free.
• The x-typesafe-request-id response header carries the request ID. Quote it when you contact support.

CompressionCopy Anchor Link

Both TypeSafe SDKs compress request bodies of 1 KiB or more with zstd by default, and the gateway accepts it.

If a proxy in front of you cannot read zstd, turn compression off. LiteLLM proxy cannot read zstd request bodies (it fails with a 500), so behind LiteLLM pass compression=False in Python or compression: false in JavaScript.

Next stepsCopy Anchor Link