Agent Session Bridge

Agent Session Bridge

Local-first session portability for Pi, Claude Code, and Codex.

agent-session-bridge mirrors resumable conversations between the three local coding agents so you can start in one tool, switch tools in the same folder, and keep going from the same thread.

https://github.com/user-attachments/assets/d1b151a4-78e9-42ef-98ee-bb5045809f51

Why

Each agent stores sessions in a different place and in a different format:

- Pi uses session trees in ~/.pi/agent/sessions
- Claude Code uses JSONL transcripts in ~/.claude/projects
- Codex uses rollout event logs in ~/.codex/sessions

This project sits between them and keeps those local stores in sync.

The goal is simple:

1. talk to Pi in a repo
2. open Claude Code or Codex in that same repo
3. resume the same conversation instead of starting over

What Makes It Different

This repo is not a session viewer, a skills sync, or a live agent-to-agent bridge.

It is focused on one job:

- parse native session formats
- convert them into the other tools' native formats
- keep replay-safe mirror state on disk
- repair older imported sessions when history got messy

The distinctive angle is local-first durability.

- The bridge writes directly into each tool's own session store.
- The CLI installs self-contained runtime assets into ~/.agent-session-bridge/runtime.
- Sync, repair, and validation are all built around real local agent behavior.

Install

Public npm install:

Or install globally:

Tarball flow also works if you want to test an unpublished local build:

Local checkout flow:

CLI

Main commands:

- setup: install Pi, Claude Code, and Codex integration for the current repo
- enable: enable sync for the current repo or globally
- doctor: read-only health check for bridge config, installed integrations,…