Welcome to Trajectory IR

A next-generation durable execution system that makes it possible to effortlessly compose, seal, and observe every AI agent tool execution in real-time for the first time ever.

The Problem

Modern AI agents execute multi-step workflows across LLMs, APIs, and databases. When an agent fails mid-execution:

• Unsealed side-effects re-execute, causing duplicate API calls, double charges, or corrupt state.
• Execution context is wiped, requiring expensive re-prompting.

The Solution

Trajectory IR seals every decision using RFC 8785 (JCS) and SHA256 before execution, guaranteeing 100% crash recovery and zero duplicate side-effects.

from trajectory_ir.runtime import Trajectory
from trajectory_ir.effects import EffectClass
from dbos import DBOS

DBOS.launch()

@Trajectory.tool(effect_class=EffectClass.NON_IDEMPOTENT_WRITE)
def deploy_server(name: str):
    return f"Deployed {name}"

@DBOS.workflow()
def run_agent():
    traj = Trajectory.start(tenant_id="prod")
    result = deploy_server("cluster-01")
    traj.append_observation(result)

Explore the Docs