Traceform

The black box for
your AI agents.

Capture every trace, span, and tool call from your AI agents — without changing your architecture.

Launching Q2 2026 · Built by Northstack

Drop-in SDK

Install in 2 lines. Works with any agent framework — LangChain, Vercel AI SDK, raw OpenAI. Zero lock-in.

npm install @traceform/sdk
import { trace } from "@traceform/sdk";

See every LLM call, tool execution, and retrieval step. Waterfall timeline. Input/output at every node.

LLM call · 340ms

Tool: search · 120ms

Retrieval · 85ms

LLM call · 290ms

Token spend by model, latency by step, error rate over time. Know what your agents cost before your bill does.

142k

tokens

1.4s

p95 latency

0.3%

error rate