@swisspra2
Swiss PranDeveloper▲5
AI Platform Architecture, Applied AI Founder, Independent
47m ago · New project · Agents · Bangkok

Shipped: Pursers: A Control Plane for Autonomous AI Agent Fleets

Why I built Pursers
The more coding agents I ran, the more time I spent moving context between tools, chasing stalled work, and checking what “done” meant. I built Pursers so I could direct the fleet without managing every handoff.

What it does
Pursers is an open-source, self-hosted control plane for running AI agents across tools and projects.

Basically, my earlier project ‘On_board’ was 1:1.

Pursers, is 1 to Many.

How it works
A coordinator breaks a goal into tickets. Workers claim them and build in parallel. A separate reviewer checks the exact commit and test evidence before the work comes back to me to merge.

How it helps me
With Pursers, I can run several projects at once without babysitting 20 or 30 agents or reviewing every change myself. I talk to one coordinator per project. Workers write the code, and independent reviewers check it. That makes it practical to use a premium model as the coordinator: its tokens go toward planning and decisions, not writing code.

1. Pursers built itself. I wrote zero lines of code. I set the direction and made the final calls, but I didn’t write code or make a single commit myself. From the first commit on August 22 through f2ebf456 on November 18, 2026, the full public main history contains 1,848 commits: 1,400 non-merge commits and 448 merges. That’s the full history, not a handpicked window.

Possibility is limitless, check this out

2. 200 SonarQube issues fixed overnight for under $10 total On an enterprise production cloud VM, I switched Butler to active mode and gave it scoped permissions to take in findings and supervise the fleet. Pursers ran the same work-and-review loop automatically, from SonarQube intake to opening a PR against the dev branch. The developers still run final tests and decide what merges into main.

3. One place to register enterprise MCP tools and A2A agents In another production deployment, I added a registration page where developers can connect MCP services or A2A agents. Pursers automatically configures a worker as a dedicated expert for each application. A user’s agent can delegate to that expert instead of spending time and tokens figuring out the same tools and workflows on every task.

github.com/swisspra/Pursers

Pursers: A Control Plane for Autonomous AI Agent Fleets logoPursers: A Control Plane for Autonomous AI Agent Fleets@swisspra/pursers · pursers.app
AI AgentsMulti-Agent SystemsAgent OrchestrationA2AMCPACPOpen Source

Comments

No comments yet
Start the conversation.
Log in to join the thread.