For the last few months my daily work has been spread across a growing pile of AI coding tools — Claude Code in one terminal, Codex in another, Fable somewhere else, each with its own sessions, its own context, its own MCP servers and its own way of remembering (or forgetting) what I was doing. It worked, but I was the glue holding it all together. Every context switch cost me something.
So I built APX — and it's now live and running really well.
What APX actually is
APX is a local daemon that acts as my personal assistant for working with AI agents. Instead of me jumping between CLIs, APX sits in front of all of them and orchestrates the whole thing from one place.
In practice that means it can:
- Manage sessions across Claude, Codex and Fable — start them, route work to the right runtime, and keep track of what each one is doing.
- Schedule tasks so work runs when it should, not only when I happen to be sitting at the keyboard.
- Organize everything: projects, agents, MCP servers and artifacts, all in one coherent structure instead of scattered configs.
The tagline in my head was simple: stop being the glue. Let one assistant hold the context so I can stay on the actual problem.
The problem it solves
If you use more than one AI CLI seriously, you already know the pain:
- Each tool has its own notion of a "session," and none of them talk to each other.
- MCP servers get re-declared per tool, per project, with subtly different scopes.
- Your agents and prompts live in a dozen folders and README files.
- When you want something to run later — or run in a different runtime — you're doing it by hand.
APX turns that into a single control plane. One place to see the agents, one place to register MCPs, one place to hand a task to whichever runtime fits it best.
How I use it day to day
The part that changed my workflow the most is being able to delegate a task to any runtime without re-setting-up my whole environment. I can hand a task to the local super-agent, or route it to Claude, Codex or another CLI — same projects, same agents, same MCPs underneath.
On top of that, the scheduling piece means the boring recurring work (status checks, routine jobs, follow-ups) just happens. I stopped being the cron job.
And because projects, agents and artifacts are organized in one model, onboarding a new agent or spinning up a new project is minutes, not an afternoon of copy-pasting config.
It's open source
APX is open and you can dig into it here:
👉 https://github.com/agentprojectcontext/apx
If you're juggling multiple AI CLIs and feel like you are the integration layer, this is exactly the itch APX scratches. I'd genuinely love feedback, issues and ideas — this is the tool I use every single day, so it keeps getting better.
More coming soon. If you want to see it in action, I'm posting demos on my channels as Tecno Manu.
Top comments (0)