Official SDKVerifiedAvailable2026-09-23
TypeSafe JavaScript SDK
A server-side Node.js integration using the official SDK, typed choice outputs, bounded retries, response validation, and a human-review threshold.
Install
npm install @typesafe-ai/sdkServer environment variables
TYPESAFE_API_KEYPrerequisites
- Node.js 20 or newer
- A TypeSafe API key stored on the server
- A documented fallback for low-confidence decisions
Implementation steps
- 1Install the official package and keep it on the server.
- 2Set TYPESAFE_API_KEY in the server environment.
- 3Define named choices with operationally distinct criteria.
- 4Set a per-attempt timeout and a bounded retry count.
- 5Validate the returned choice and confidence before routing work.
- 6Send low-confidence decisions to human review.
Server example
import { choice, TypeSafeClient } from '@typesafe-ai/sdk';
const apiKey = process.env.TYPESAFE_API_KEY;
if (!apiKey) throw new Error('TYPESAFE_API_KEY is required');
const client = new TypeSafeClient({
apiKey,
timeout: 10_000,
retry: { maxRetries: 2 },
});
const result = await client.systemOne({
model: 'jev-latest',
state: {
ticket: 'I was charged twice. Please fix this ASAP.',
},
questions: {
route: choice('Choose the support queue.', {
billing: 'Payments, invoices, refunds, or duplicate charges',
technical: 'Product behavior, errors, or integrations',
other: 'Everything else',
}),
},
});
const decision = result.answers.route;
if (
typeof decision.choice !== 'string' ||
!Number.isFinite(decision.confidence) ||
decision.confidence < 0 ||
decision.confidence > 1
) {
throw new Error('Unexpected decision response');
}
const action = decision.confidence < 0.8 ? 'human_review' : decision.choice;
console.log({ action, confidence: decision.confidence, model: result.model });Production checks
- Never ship the API key to browser code.
- The SDK timeout is per attempt, so total wall time can exceed ten seconds when retries occur.
- Do not log request bodies when they can contain personal or confidential data.
- Pin and test a package version before release.
Limitations and verification notes
The example demonstrates integration safety, not task accuracy. Calibrate the 0.8 review threshold on representative data before production use.
Sources
- TypeSafe AI JavaScript SDKTypeSafe AI · 2026-09-23Open source
- TypeSafe API referenceTypeSafe AI · 2026-09-23Open source
This guide is independently maintained by Type Safe. Check the provider’s current documentation, versions, and terms before use.