Get started

5-minute quickstart.

Predict on real data, apply a correction, watch the audit chain extend — without leaving this page.

Sign up + create a namespace

Sign up at /signup. Magic-link email; no password. Your account starts on the free tier with 5,000 predictions/month included.

From the dashboard, create a namespace. Pick an encoder: managed_bge (we host), or openai/cohere/voyage (you supply the API key — free tier defaults here).

Get an API key

From /dashboard/keys generate an sst_live_* key. This is shown ONCE — copy it now.

# Save your key + namespace as env vars
export SEEDSTRATE_API_KEY="sst_live_..."
export SEEDSTRATE_NAMESPACE="ns_acme"

First predict (curl)

curl -X POST https://seedstrate.ai/api/v1/predict \
  -H "Authorization: Bearer $SEEDSTRATE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "namespace": "'$SEEDSTRATE_NAMESPACE'",
    "input": "I want to cancel my card"
  }'

Response includes prediction_id, label, confidence, mode_used, and a horizon_trace showing what each substrate component contributed.

Apply a correction

curl -X POST https://seedstrate.ai/api/v1/correct \
  -H "Authorization: Bearer $SEEDSTRATE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "namespace": "'$SEEDSTRATE_NAMESPACE'",
    "prediction_id": "pred_xxx",
    "correct_label": "card_cancellation",
    "reason": "agent reviewed"
  }'

No retraining. The next prediction with similar input lands on card_cancellation immediately.

Verify the audit chain

curl https://seedstrate.ai/api/v1/audit/pred_xxx \
  -H "Authorization: Bearer $SEEDSTRATE_API_KEY"

Returns the hash-chained ledger for that prediction + every correction applied to it. verified: true means every entry's content hash and chain link check out.

Use the SDKs (optional)

# TypeScript
npm install @seedstrate/sdk

import { Seedstrate } from "@seedstrate/sdk";
const sst = new Seedstrate({ apiKey: process.env.SEEDSTRATE_API_KEY!, namespace: "ns_acme" });
const r = await sst.predict({ input: "I want to cancel my card" });

# Python
pip install seedstrate

from seedstrate import Seedstrate
sst = Seedstrate(api_key="sst_live_...", namespace="ns_acme")
r = sst.predict(input="I want to cancel my card")

Latency tip

Use a persistent HTTP client with connection pooling. The official SDKs already do this — they reuse one connection across many calls, so steady-state latency lands around ~100 ms p50 via the tunnel. Naïve curl-per-request or requests.post-per-request without a session pays a TCP + TLS handshake (and on Windows, an IPv6→IPv4 fallback) every call, which adds 200–2000 ms per request.

For serverless (Vercel, Lambda) where each invocation can be a fresh process, instantiate the SDK client in module scope and reuse across handler invocations on warm functions.

What's next?

· Full API reference — every endpoint, every field, every error code.
· /dashboard/playground — run predicts + corrections + see the audit chain inline without writing code.
· /dashboard/review-queue — operate the trust pipeline. Promote end-user corrections to shared.
· /dashboard/training — fine-tune a custom encoder for your domain (Pro tier).
· /dashboard/webhooks — get notified of every predict / correct / flag / promote / revoke.

curl -X POST https://seedstrate.ai/api/v1/predict \ -H "Authorization: Bearer $SEEDSTRATE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "namespace": "'$SEEDSTRATE_NAMESPACE'", "input": "I want to cancel my card" }'

curl -X POST https://seedstrate.ai/api/v1/correct \ -H "Authorization: Bearer $SEEDSTRATE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "namespace": "'$SEEDSTRATE_NAMESPACE'", "prediction_id": "pred_xxx", "correct_label": "card_cancellation", "reason": "agent reviewed" }'

# TypeScript npm install @seedstrate/sdk import { Seedstrate } from "@seedstrate/sdk"; const sst = new Seedstrate({ apiKey: process.env.SEEDSTRATE_API_KEY!, namespace: "ns_acme" }); const r = await sst.predict({ input: "I want to cancel my card" }); # Python pip install seedstrate from seedstrate import Seedstrate sst = Seedstrate(api_key="sst_live_...", namespace="ns_acme") r = sst.predict(input="I want to cancel my card")