Docs 29 guides · TypeScript · Python

Build agents you can explain.

Trace decisions, measure outcomes, and promote the versions that prove themselves in production.

Start tracing Install Docs MCP
~/my-agentzsh
  • $ npm install tracify-sdk

The path

From npm install to release gate.

01 / 04

Install.

Add the SDK, drop in your API key, and you are one wrapper away from a trace.

Quickstart →TypeScript SDK →Python SDK →
1 · install

.env

TRACIFY_API_KEY=tracify_sk_live_…
02 / 04

Instrument.

Wrap a run. Model calls, tool calls and retries become spans with inputs, outputs, tokens and timing.

Tool calls →Integrations →
2 · instrument
import { traceAgent } from "tracify-sdk";

const runAgent = traceAgent(
  async (question: string) => callModel(question),
  { apiKey: process.env.TRACIFY_API_KEY },
);

await runAgent("Where is my order?");
03 / 04

Inspect.

Open the run. See the whole path, the cost of every step, and the first place it went wrong.

Observability →Costs →
3 · inspect · run_9f3a
  • support-agentAGENT
  • chatLLM
  • order.lookupTOOL
  • refund.issueFAIL
  • chat · retryLLM

First failure: refund.issue · 502

04 / 04

Improve.

Turn what you saw in production into evaluations, versioned prompts and release gates.

Evaluation →Prompts →Release gates →
4 · improve · release gate (example)
  • Pass rate on the eval set ≥ 95%PASS
  • p95 latency under 4 sPASS
  • Cost per task under $0.05BLOCKED

Ship the version that proves itself in production.

All guides

Read by intent.