How Trajectory IR fits
Honest comparison — what Trajectory IR is, what it is not, and how it relates to Temporal, LangGraph, MCP, and memory products.
What it is / is not
| Trajectory IR is | Trajectory IR is not |
|---|---|
A portable IR + .tir package format | A replacement for Temporal / DBOS / Restate |
| Seals, effect classes, resume semantics | An agent orchestration framework (not LangGraph) |
| A thin layer over pluggable durable backends | A long-term memory product (not Mem0 / Zep) |
| Open source libraries (Apache-2.0) | A hosted multi-tenant SaaS |
How it relates to other systems
| System | What it solves | What Trajectory IR adds |
|---|---|---|
| Temporal / DBOS / Restate | Crash safety, retries, leases | Agent seals, effect classes, portable .tir |
| LangGraph / CrewAI / ADK | Orchestration & checkpoints | Framework-agnostic export of what ran |
| MCP tool hints | Basic tool safety vocabulary | Mapping into six effect classes + resume matrix |
| Mem0 / Zep | Long-term memory recall | Optional LTM node shapes only — not a recall product |
Trajectory IR sits between your agent host and a durable backend. You keep Temporal (or DBOS / Restate) for execution durability; you add seals and portable packages for agent-specific safety and auditability.
Industry context (not product metrics)
These numbers describe the broader agent ecosystem. They are not Trajectory IR latency, throughput, or win-rate claims.
In LangSmith usage data published by LangChain (State of AI 2024), traces with tool calls rose from 0.5% to 21.9%, and average steps per trace rose from 2.8 to 7.7.
A VentureBeat Pulse (July 2026, n=108 organizations with 100+ employees) reported that 49% had shipped an agent that passed internal evaluations and then failed with customers.
Always treat the above as industry context with citations and caveats.