A new open-source tool called Agent Standup lets developers watch their AI coding agents work as if they were colleagues on a team video call. Published on a project page at atoll92.github.io and shared on Reddit by its creator, the tool turns Claude Code multi-agent sessions into a live view that shows which agent is making progress, which is waiting on another, and which needs a human decision.
The project is free and released under the MIT licence. It runs entirely on the user's own machine, and its makers say it adds no extra token cost to a session.
How the tool presents agent activity
Agent Standup borrows the format of a daily team check-in. Each agent appears as a participant in a video-call-style layout, and its tile reflects its current state. The tool recognises five statuses: working, waiting on another agent, blocked, needs approval, and done. Agents are sorted by urgency, so the ones that require attention appear first.
That ordering is the main practical idea. When several agents run in parallel, a developer often has to check each terminal to find out which one is stuck. A single view that ranks agents by how much they need help is designed to reduce that overhead.
Reading existing session files
According to the project description, the tool does not connect to a remote service or add its own model calls. Instead, it reads the session files that Claude Code already writes to disk. Because no extra prompts are sent, the project states that it adds zero tokens to usage, and because the data never leaves the machine, nothing is transmitted externally.
For teams that handle proprietary code, that local-only design may matter more than the visual novelty. Developers often hesitate to route session logs through third-party dashboards, and a tool that works offline from existing files avoids that question.
Where it runs and what else it offers
The project lists three ways to use it. It can run as an extension in VS Code or Cursor, as a command-line tool launched through npx, or as a desktop app on Mac, Windows, and Linux. This range means it can fit into a range of editor setups without a separate account or hosted service.
Beyond the live view, the project mentions several extras:
- Session replay with highlighted moments
- A context gauge to show how full an agent's working memory is
- Video export of a session
- Optional voices for the team-call effect
Context for developers
Multi-agent coding, where several AI agents split up tasks such as writing code, running tests, and reviewing changes, has become a common topic among developers using tools like Claude Code. As these workflows grow, the difficulty shifts from getting one agent to work to tracking many at once. Tools that make agent state visible are one response to that problem.
The project's author invited feedback and pull requests, which suggests an early-stage release that is still taking shape. Readers should check the project's documentation for current features and supported versions, since descriptions of a young open-source tool can change quickly.
What to watch next
The key questions for Agent Standup will be practical ones: how accurately it interprets session files as those formats change, and whether the urgency sorting holds up in large sessions with many agents. Community contributions will likely determine how far the tool extends beyond Claude Code.
For developers already running several agents at once, the appeal is straightforward. A glanceable, fully local view of who is working and who is waiting could make multi-agent sessions easier to supervise.
Related: Pineflow: Free Open-Source Guide to Shipping a Mobile App with AI · AdHelm: Free Local CLI Lets AI Agents Audit Your Ad Accounts
Comments (0)