Documentation

Traza Cloud runs the open Traza engine for you. The first part of these docs covers what the cloud adds: accounts, keys, the gateway, alerts, the notary and billing. The second is the engine’s own documentation, imported from traza 0.28.0, which describes every route your project can use.

Traza Cloud

Start

• Quickstart From sign-in to a recorded run, a run fuse and a replay on Traza Cloud, in about five minutes.
• API keys Traza Cloud API key format, the ingest, read and full scopes, the routes each reaches, rate limits, and what the gateway never forwards.

Send data

• Claude Code & Codex Record every Claude Code and Codex session as runs with traza-server import, follow live ones with --watch, steer them with the hook pack, and see every developer's sessions in one view on Traza Cloud.
• OpenTelemetry Export traces to Traza Cloud from any OpenTelemetry SDK, and the three details that decide whether sessions, Guard and Replay work.
• Python Send traces from Python to Traza Cloud with the OpenTelemetry SDK, OpenLLMetry or OpenInference.
• TypeScript Send traces from Node.js to Traza Cloud with the OpenTelemetry Node SDK, or with OpenLLMetry for JavaScript.
• OpenAI Agents SDK Record OpenAI Agents SDK runs in Traza Cloud with the OpenInference instrumentation, one session per conversation.
• LangGraph Record LangGraph and LangChain runs in Traza Cloud with OpenLLMetry or OpenInference, one session per thread.
• Claude Code (OTel, hook) Put Traza's pre-call gate and nudges in front of every Claude Code tool call with the engine's hook pack, and optionally export Claude Code's own beta traces.
• curl Send spans to Traza Cloud without an SDK, in Traza's native JSON or OTLP/HTTP JSON, compressed or not.

Use

• Foresight Read your agents' Fate Maps, get a verdict for a running run, and stop runs whose path has historically failed with a predictive fuse, on Traza Cloud.
• Steer Mission Control for live runs across your projects, nudges that reach an agent at its next pre-call check, and approving, killing or nudging a run from your phone or Slack.
• Guard Run fuse policies (predictive ones included), the pre-call check, kill switch and approvals, and alert delivery to email, push, interactive Slack and signed webhooks on Traza Cloud.
• Replay Create replays of recorded runs and point the OpenAI or Anthropic SDK at Traza Cloud's gateway, with zero model calls.
• X-Ray Read a recorded model call as the context window it was, with each chunk traced to the step that introduced it, and find out why a prompt cache broke and what it cost.
• Prove Fetch evidence packs, check their digest, and verify Traza Cloud notary signatures on attestations and ledger anchors with OpenSSL 3 and Python.
• MCP Connect Claude Code or any MCP client to a Traza Cloud project, what each key reaches, and how stored text is kept from steering the model.

Account

• Billing How Traza Cloud meters spans, what happens at the cap, overage and spend caps, plan changes, failed payments and retention.
• Export Export every span from Traza Cloud at any time, on any plan, check that the export is complete, and move it into a self-hosted traza-server.

Engine reference

• Getting started Run the engine yourself and send a first span.
• HTTP API reference Every route, parameter, response shape and error.
• LLM semantics The OpenLLMetry, OpenInference and GenAI keys Traza reads.
• Foresight The Fate Map, verdicts for one run, predictive fuses, and their limits.
• Steer Live runs, nudges delivered at the pre-call check, and run topology.
• X-Ray and Cache Autopsy The context window a call saw, and why its prompt cache broke.
• Recording coding agents Claude Code and Codex transcripts as runs, and Agent Blame.
• Run fuses Runs, usage, verdicts, and why a post-ack stop is one step late.
• Replay Tapes, the replay proxy, strict matching, diffs and recall.
• Fork & Rewind Fork a run at any step, rewind a long one, tape tests.
• Ledger and evidence packs The hash chain, its verification, and what it proves.

All engine documentation · Changelog