agent-thread

Share, inspect, import, and locally convert Claude Code and Codex chat sessions.

agent-thread is a TypeScript CLI and Next.js share viewer for publishing local agent chats as public links. It uses Bun for development, tests, and builds, while the published CLI is Node-compatible. It can also import shared threads back into Claude Code or Codex, and convert local Claude/Codex histories without uploading them.

Features

• Claude Code session discovery from ~/.claude/projects
• Codex thread discovery from ~/.codex/sessions
• Interactive session picker scoped to the current project
• Public share pages backed by Cloudflare D1 and R2
• Shared thread imports back into Claude Code or Codex
• Local Claude Code <-> Codex chat conversion without uploading
• bunx agent-thread CLI entrypoint through the agent-thread binary

CLI Usage

Upload the latest Claude Code session to the default hosted service:

bunx agent-thread --latest

Upload Codex threads instead:

bunx agent-thread --codex

Point the CLI at a local Cloudflare preview or self-hosted server:

AGENT_THREAD_SERVER_URL=http://127.0.0.1:<preview-port> bunx agent-thread
AGENT_THREAD_SERVER_URL=https://your-domain.example bunx agent-thread --codex

Import a shared thread into Claude Code or Codex:

bunx agent-thread --import 0c5a0y4a406r
bunx agent-thread --import https://agent-thread.com/t/0c5a0y4a406r --to codex

If --workspace is omitted, imports are attached to the current directory. Use --dry-run to inspect target paths without writing files and --force to overwrite an existing local import.

Convert a local chat into the opposite app without uploading:

bunx agent-thread --local
bunx agent-thread --codex --local
bunx agent-thread --local --latest --dry-run
bunx agent-thread --codex --local --workspace /path/to/project

Local conversion always targets the opposite app. --to is optional and must match that opposite target when provided. It does not require a web server or Cloudflare resources.

Requirements

• Bun
• A local Claude Code or Codex history if you want to upload or convert chats
• Cloudflare account, D1 database, and R2 bucket if you want to self-host the web app

Local Setup

Install dependencies:

bun install

Run type checks and tests:

bun run check
bun test

Start the Next.js dev server:

bun run dev

The dev server is useful for UI work. Upload, export, and share APIs require Cloudflare bindings for D1 and R2, so use bun run preview or a deployed Worker when you need the full hosted flow.

Self-Hosting

agent-thread deploys to Cloudflare Workers through OpenNext. The hosted app needs:

• D1 binding named DB
• R2 binding named SESSIONS_BUCKET
• PUBLIC_BASE_URL set to your public app URL
• AGENT_THREAD_SERVER_URL set to the same URL for CLI defaults in the deployed app

Create Cloudflare resources:

bunx wrangler d1 create agent-thread
bunx wrangler r2 bucket create agent-thread-sessions

Update wrangler.toml with your own values:

• database_id from the D1 create output
• bucket_name for your R2 bucket
• routes for your domain, or remove the [[routes]] block if you deploy to a workers.dev subdomain
• PUBLIC_BASE_URL and AGENT_THREAD_SERVER_URL

Apply the D1 migration:

bunx wrangler d1 migrations apply agent-thread

Build and test the Worker locally with Cloudflare bindings:

bun run preview

Deploy:

bun run deploy

After deployment, point the CLI at your instance:

AGENT_THREAD_SERVER_URL=https://your-domain.example bunx agent-thread --latest

Project Scripts

bun run cli        # run the TypeScript CLI locally
bun run dev        # run Next.js dev server
bun run preview    # build and preview the Cloudflare Worker
bun run deploy     # build and deploy to Cloudflare
bun run build      # production build plus CLI bundle
bun run check      # TypeScript check
bun test           # test suite

Data Model

Uploads store metadata in D1 and session JSONL payloads in R2. The first migration creates the uploads table and indexes public IDs and source session IDs.

The app stores raw source files alongside a normalized transcript representation so shared threads can be rendered in the browser and imported back into either supported local app.