retrace

// sessions

See the whole multi-agent system

Group multi-turn runs into a single session, see how your agents hand off work in a topology graph, and automatically catch failure modes that only appear between agents.

Start freeRead the docs →

Sessions

Runs that belong to the same conversation are grouped into a session, with totals for cost, tokens, and duration, so you can replay an entire multi-turn interaction.

Agent topology

Tag spans with an agent id and role and Retrace builds a topology graph showing how work is ordered and handed off between agents.

Cross-run registry

Each agent gets a profile across all your runs, so you can see how a given agent behaves over time, not just inside one trace.

Inter-agent detectors

Automatically flag reasoning–action mismatch and task derailment between agents, plus deterministic ping-pong loops.

Trajectory checks

Catch goal-drift and dropped early-turn context, failures that end-state checks miss entirely.

Single- and multi-agent

Single-agent runs are untouched; the multi-agent analysis only engages when a run actually involves two or more agents.

agent.py
import retrace

with retrace.agent("researcher", role="research"):
    findings = research_agent.run(task)

with retrace.agent("writer", role="writer", parent="researcher"):
    draft = writer_agent.run(findings)

Tag each agent's work to build the topology graph.

agent.ts
import { withAgent } from "retrace-sdk";

withAgent({ id: "planner", role: "Planner" }, () => {
  return planAgent.run(task);
});

TypeScript, withAgent scopes spans to an agent.

REST API

  • GET /api/v1/sessions: list sessions
  • GET /api/v1/sessions/:id: session detail
  • GET /api/v1/sessions/:id/graph: agent topology graph

More of the platform

// recordRecordOne decorator turns every run into a regression test// fork & replayFork & ReplayFork the failed step and re-run, debug the cause, not the symptom// prove the fixProve the FixRe-run a fix and get a verdict before you ship// eval gatesEval GatesFail the build when agent behavior regresses// guardrailsGuardrailsHalt a run when it blows a budget, loops, or stalls// enforcementEnforcementBlock a bad action before it runs, not diagnose it after// failure taxonomyFailure TaxonomyAuto-classify failures so you fix patterns, not one-offs