A coding task is running on your computer. You leave your desk, then want to check its progress, read the analysis, and decide whether to implement the plan—all from your phone.
The useful part is keeping the same session and its context, rather than starting a separate conversation. That is what this unofficial Telegram extension for Atomic Agent adds.
A short demo
Watch the 46-second demo on GitHub.
The video uses recreated interfaces and fictional content. It is an illustrative workflow with condensed timing, not an unedited screen recording.
The workflow
Start a task in the patched Atomic terminal, for example:
Analyse the settings module and prepare a refactoring plan. Do not implement it yet.
In your paired private Telegram chat:
/desktop
How far along are you?
Show me the result of your analysis
/review
The review shows the latest completed answer as the proposed plan, current Git changes, and any retained verification evidence. Approve plan queues implementation of that exact plan. Reject plan dismisses the review; it does not undo existing edits.
Other controls include:
/queue Run the relevant tests and review the changes
/status
/review diff
/stop
/review diff includes the tracked-file patch. /stop requests cancellation of the current desktop task and clears outstanding Telegram requests.
How it works
Atomic already has a Telegram integration. This extension adds continuation of eligible terminal sessions through the same process and session store.
Status and result queries read saved state without another model request. Follow-ups use Atomic's per-session execution queue and reload the latest context when they start.
Plan-review buttons are restricted to the paired owner's private chat and bound to the original message and session. They work once and expire after 30 minutes. The bridge checks session and Git freshness again on approval and at the execution boundary. If the reviewed state changes, a new review is required. Normal tool permissions still apply.
Try the beta
Source, installation and Telegram setup
The repository provides a source installer for a pinned Atomic v0.8.0 revision, setup instructions with links to Telegram's official documentation, and regression tests. CI checks installation and runtime behaviour on Linux, Windows and macOS Intel. The workflow was also exercised locally on Apple Silicon, including a live Telegram approval and test run.
A few practical limits:
- This is an unofficial source patch, not an official Atomic plugin.
- The terminal process must stay running. A second Atomic process cannot control its in-memory queue.
- Pending follow-ups live in memory and do not survive a restart.
- Cancellation is cooperative; completed edits and external effects remain.
- Plan reviews require an idle linked session in a Git worktree. Large repositories, symlinks and submodules have documented review limits.
- Status reflects the last saved checkpoint. Recorded test evidence is not a fresh test run or an independent guarantee that tests passed.
- Messages and requested results travel through Telegram, even when the model runs locally.
What would make this useful for you?
The next improvement I would prioritise is a restart-safe queue. I would also like feedback on reviewing plans from a phone: is the diff and recorded test output enough, or would you want a different summary before approving?
This project and write-up were developed with AI assistance. The demo content is simulated; the bridge itself has automated tests and a live workflow check.
Top comments (0)