swarm-board · V0.2

Observability for a swarm of AI coding agents

Dozens of agents work in parallel on one project. swarm-board reads their prose messages and turns them into hourly summaries, a merged summary for any slice of the timeline, and a live project state graph, so an operator sees what happened, who is blocked and why, without reading thousands of messages.

Open the timeline See the evals How it was trained

Headline numbers · same audited figures as the eval Summary tab

loading…

Benchmark board: every model on every stage

Three summary schemas

Stage 1 · Prose v1

Hourly bullets and message labels

A small model (Qwen3-30B-A3B) reads every agent message once. Per agent and hour it writes four bullet lists (started, finished, blocked, key decisions) and tags each message with intent, role, blocker and urgency.

Stage 2 · Prose v2

2D merge tree

A larger model (Qwen3-235B) merges the hourly bullets up two axes: one agent over time (1h, 4h, day) and all agents in one time bucket (1h, 4h, day). Any cell, row or column of the timeline is answered by one merged node, with every bullet linked to its sources.

State graph · v0.3 lean

Project state, turn by turn

For every agent turn, in order, a fine-tuned Qwen3-30B-A3B LoRA emits a JSON patch (tasks, status, relationships, key decisions). The patch is validated against the schema and quotes, then applied, so the project graph accumulates across the whole fleet.