Introducing Plannotator Inbox. Making decisions is the new bottleneck, while your agents work on many things across many threads over longer periods of time. Inbox is how you stay in-the-loop, organized, alongside the agent harnesses you already prefer to use. Try today, free, local, open source.

We decided to ship this before a bigger commercial product. Coming soon to iPhone and Android.

btw, the video may not make it clear but this is NOT a way to wrap or orchestrate agents - you still use your harnesses, ADEs, whatever. This is more of an (optional) way for agents to send you the markdown/html/diffs to an organized place where you can reply to and send feedback using the original annotation UXs - in an async manner outside of threads

This is the exact problem I was thinking about today - only I want access on my phone. I'm building something this weekend integrated with this:

lmk how opentunnel goes or if anything I can do for official support. It should work there or through tailscale. The webpage is responsive if you get the tunnels working., might be helpful: I am shipping a mobile app but otherwise

I have Tailscale access at home - so definitely an option and my first choice. I have some other constraints so opentunnel may be the only option for me in certain situations. Will try to take a more in depth look at this over the weekend. Thanks for releasing this!

finally an inbox for multi agent workflows, love the open source angle

nice, consolidating context saves friction when agents span multiple threads

How does Inbox rank which thread needs a decision first?

you never miss bro

Super cool

This is amazing! 🎉🎉🎊🎉

Kind of cool yea but also getting worried plannotator gets too complicated and bloated. Too many features too many things…it’s great at one thing why not keep on improving that instead of turning it into another super orchestrator thingy.

Seems like it's ignoring remote and port settings. Intentional? I have to be able to access inbox over my tailnet

Im creating a helper function so you can just do `plannotator inbox --tailscale` But immediate workaround for now is in:

so…i don’t have to use random ports anymore!?

One case from building Turnfeed with Codex: I change direction while an earlier question is still waiting. Can Inbox mark that question as superseded, so I don’t answer an outdated request? The version-pinned review in your docs looks useful here.

Agents working across threads makes decisions the bottleneck Inbox for staying in the loop is exactly the right layer

This is super needed. Decision bottleneck is real when you have agents running everywhere. Love the stay in-the-loop approach.

Compelling. If it works well I'm hooked. Trying now 🙏

thanks for trying. cleaning up the install instructions/prompt to make the initial onboarding smooth in short: `plannotator inbox` and it opens in your browser, with agent connection instructions (plugins or MCP) on the screen

So far I don't understand at all. My agents are not writing to it and I had to install so many separate things already. I was hoping for a chat-like app where I read and write to agents like facebook messenger. I just see even more setup instructions even though both claude and codex are already setup

yea it's not a direct message, instead an async surface they send you things like questions to answer, docs to review, etc. and from there is treated like email where your replies do go to the agent claude code requires restart after initial install. then say something like "ask your next question in Inbox, send your html design to inbox" ty for feedback though and sorry for early frustration. carrying it over into next iterations. direct message can come but will need to think that out more.

I just misunderstood the product, thanks!

A lightweight, open-source inbox could make managing multiple agents much more seamless

Keeping humans in the loop gets harder as agents scale, this is a smart solution

Which agent harnesses does Inbox plug into out of the box?

best UX: Claude, Pi, and OpenCode, via mods/extensions and the bi-directional event syncing (wake an agent) otherwise mcp for any other agent

Inbox needs a machine-readable decision receipt, not just a human queue: thread_id, proposed_action, required_role, expiry, chosen_branch, denial_reason, and downstream jobs unblocked. That lets an agent harness prove why it waited instead of silently stalling.

this is awesome. thank you