Handoffs in one click
Move a task from Claude Code to Codex, or start a fresh session when the context window fills up. Nothing gets explained twice.
For Claude Code, Codex and Gemini CLI
Junxion runs the agents you already use on your Mac, learns what matters in each project, and hands the next agent a short brief, so it picks up where the last one left off.
How it works
Start Claude Code, Codex or Gemini CLI in any project. They sign in with your own accounts and keep their settings, plugins and connectors.
A local model reads each session and keeps the decisions, rules, preferences and open work, each with where it came from.
Switch agents, start fresh when the context fills up, or come back next week: the new agent starts with what it needs, and can ask Junxion for more.
Live demo
A sample to-do project with recorded Claude Code and Codex sessions. Start a session, answer the agent's question, hand the work to Codex, then see what Junxion learned. No sign-in.
Measured, not guessed
On our handoff benchmark (recorded sessions, a 600-token budget), Junxion's brief carried the key facts a new agent needed far better than the last 600 tokens of the conversation.
What you get
Move a task from Claude Code to Codex, or start a fresh session when the context window fills up. Nothing gets explained twice.
See everything Junxion knows about a project as a map, a list and a history. Correct anything, or add your own.
Agents use their own sign-in (or an API key you add), with their own connectors and plugins. Nothing to set up again.
Your agents and projects as a small city: see who's working where, and what's waiting on you.
Sessions and knowledge live on your Mac. Learning runs on a local model, so Junxion never sends your code anywhere.
Optional sync keeps your knowledge the same on all your Macs, sealed on your device so no one else can read it.
Questions
Claude Code, Codex and Gemini CLI today, through the open Agent Client Protocol. If you already use Claude Code or Codex in a terminal, Junxion uses that same setup.
Junxion runs on your Mac and keeps everything there. Your agents talk to their own providers, as they always do. Junxion's servers only hold your account and, if you turn on sync, data that's encrypted on your device.
No. Agents use the sign-in you already have (your Claude or ChatGPT account). You can add an API key instead if you prefer.
macOS, the agents you want to use, and Node.js. Learning uses a local model through Ollama; without it, sessions are still saved and searchable.