Welcome to Trajectory IR

Portable, hash-verifiable intermediate representation for agent execution trajectories.

Portable, hash-verifiable intermediate representation for agent execution trajectories.

Trajectory IR is the flight recorder and safety switch for agent runs: seal the model’s plan before side effects, classify tool effects, resume without re-inventing the plan, and export a runtime-independent .tir package anyone can verify by content hash. It sits on top of durable execution engines (Temporal, DBOS, Restate) — it does not replace them, and it is not another agent framework.

Rendering diagram...

  Agent host / framework     (LangGraph, custom loops, MCP hosts, …)
            │
            ▼
  ┌─────────────────────────────┐
  │       Trajectory IR         │  seals · effect classes · .tir · sandbox
  └─────────────────────────────┘
            │
            ▼
  Durable backend              (Temporal · DBOS · Restate)

Failure today → what Trajectory IR adds

Failure today	What Trajectory IR adds
Crash mid-tool → retry double-fires a write	Block-and-gate + durable backend memoization
Resume re-asks the model → silent plan drift	Sealed `DECISION` is the plan; resume replays it
History locked in one framework	Portable thin/fat `.tir` with hash-checked node IDs
Demo can charge / email / deploy for real	Sandbox mode blocks dangerous effect classes

Who is this for?

Audience	Start here
Students / newcomers	Go Quickstart → Sandbox demo
Agent / platform engineers	How it fits → Concepts
Security / compliance	Seals + `.tir` + Security
Contributors	Contributing · Go first for new features
AI coding agents	Engine master spec — implement the spec, don’t invent APIs