AI NewsDev toolsAnnouncement

PandaDOS puts coding agents behind a browser work queue

PandaDOS is a self-hosted control plane for Claude Code and Codex. It turns issues into a queue of coding jobs you can watch in a browser.

AI News

Editorial2 min read

LinkedInX
PandaDOS GitHub repository social card

Image: GitHub

Why it mattersA team can hand many coding agents many small jobs. They can then check progress from one tab instead of many open terminals.

A new GitHub repository called PandaDOS reached 652 stars and ranked 21st on Trendshift's daily list on 8 November 2026. Its own line is this: "Your self-hosted control plane for Claude Code and Codex. Turn issues into queued, observable, approval-aware coding workflows, from any browser." The project is open source under Apache 2.0.

PandaDOS sits in front of a team's existing coding agents. A developer or a product manager files an issue. PandaDOS picks it up, runs the agent on it, and shows the run as a job card in a browser. The team watches terminal output, file changes and Git operations through one page.

What the queue is for

The README names four jobs for the control plane. It automates the routine approvals a coding agent asks for during a run. It keeps context across tasks. A second job on the same codebase does not rebuild its picture from scratch. It makes the work visible from a browser. And it moves ad-hoc chat into tracked, planned work with stages.

The last one is the clearest change. A conversation with an agent is hard to share across a team. A queue of issues with status is not.

The approval policy

PandaDOS has three approval policies. They are called cautious, balanced and automatic. In cautious, a person signs off on nearly every step. In automatic, safe and reversible edits run without a hand on the keyboard. Destructive or ambiguous edits escalate back to a person in every mode.

A team using Claude Code or Codex locally today will know this is the hole. The agent asks for permission on every tool call. Running ten jobs at once is ten streams of permission prompts. PandaDOS puts one policy on top and routes only the edge cases to a person. Whether the edges are the right edges is the policy's job.

Where it fits

PandaDOS is self-hosted. The team runs it in its own environment. It does not replace Claude Code or Codex. It sits above them, calling their public interfaces.

Imagine a team already pays for Claude Code or Codex. Its choice is between adding PandaDOS and adding more open terminals. A browser with a queue is a cheaper way to see the work. The three-level policy has to fit how the team reviews code.

The traction is concrete. The repository has 652 stars on GitHub on 8 November 2026. It ranks 21st on Trendshift's daily list that day. A small team at uniquechao publishes it. The maintainers tag releases often. The most recent is v2.6.5 on 6 November 2026, two days before this item ran.

A coding manager running three or four agents this quarter has one question for a tool like this. Does it hide failures, or does it surface them? The README shows file changes and terminal output on the job card. A team can read for itself how an agent lost the thread.

Source

uniquechao/PandaDOS on GitHub.

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project