trajectory by letta-ai: developer tool for Hermes Agent

TypeScript library that normalizes agent session transcripts from Hermes and other runtimes into one format

What trajectory does

trajectory gives agent tools one TypeScript API for a problem each of them solves differently: representing messages, reasoning, tool calls and tool results. The caller passes a transcript string and its source name to normalizeTranscript and gets back records plus diagnostics, which is empty when no cleanup was needed. Records carry roles such as meta, user, assistant and tool, with timestamps and tool call ids. Deep Agents is the exception, because normalizeCheckpoint reads its LangGraph SQLite store by thread ID.

The supported sources are ATIF, Claude Code, Codex, Copilot CLI, Cursor, Droid, Gemini CLI, Hermes, Letta Code, Oh My Pi, OpenClaw, OpenCode, OpenHands, pi and Deep Agents. The Hermes adapter accepts a session-store message-row array or an envelope with a session object and a messages array. A Python wrapper is published as agent-trajectory and imports as trajectory.

Key features

One normalizeTranscript API for 15 agent transcript formats
Hermes adapter for session-store rows or a session-plus-messages envelope
Deterministic, validated records with diagnostics for recoverable cleanup
TypeScript package @letta-ai/trajectory on npm
Python wrapper agent-trajectory on PyPI

When to use it

Feeding Hermes session history into a memory or dreaming pipeline
Comparing transcripts from several coding agents in one format
Preparing agent runs for evaluation or training data

Who it is for: Developers building memory, search or evaluation tools on top of transcripts from several agent runtimes.

How it fits with Hermes Agent

Hermes Agent is one of the supported sources, and the repository carries the hermes-agent topic alongside Claude Code, Codex and OpenClaw.

How to install trajectory

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

npm install @letta-ai/trajectory
pip install agent-trajectory

FAQ

What is trajectory?

trajectory is a library that normalizes agent transcripts from different runtimes into one validated record format. It is designed to be consumed by agents, for example for memory formation, dreaming and search.

Does trajectory work with Hermes Agent?

Yes, it has a hermes source adapter. The adapter accepts a session-store message-row array or an envelope with a session object and a messages array.

How do I install trajectory?

Run npm install @letta-ai/trajectory for the TypeScript package, or pip install agent-trajectory for the Python wrapper, which imports as trajectory.