RepoSwarm: Giving AI Agents Architecture Context Across All Your Repos
summary
RepoSwarm automatically clones hundreds of repositories, runs Cloud Code / agent SDK prompts inside a Temporal orchestration, and commits an architecture hub repository of purely markdown files that get continuous regeneration (daily). Each repo analysis takes about 10–15 minutes and roughly $0.30–$1, produces 80–90% correct monitoring, dependency and ADR findings, and creates an auditable architecture history that saves days on tasks like monitoring consolidation and service deprecation.
AI-written summary · about
chapters (9)
- 0:00Intro and audience questions
- 5:00Common architecture questions that are hard to answer today
- 8:45How a single repo gets transformed into structured markdown docs
- 15:20Example: generating a report about monitoring tools across repos
- 21:40Why architecture history is as important as architecture snapshots
- 28:00Discovery, filtering, and repo selection
- 33:10Token and time costs per repository
- 34:00Future roadmap: Slack integration, RAG, MCP, cross-repo reasoning
- 35:15Open source repo and final thoughts
related talks
42mAgentes en el monorepo
25mDirecting a Swarm of Agents for Fun and Profit
24mAI Agent Swarms Are Amazing
01h 16mAutomating Large Scale Refactors with Parallel Agents
44mThe Swarm Next Door: Running Parallel AI Agents on a Real Android Codeba…
14mAgent-Driven Software Development
31mModern Architecture for AI-Augmented Development Teams
31mInternal Background Agent Systems: Architecture and Rollout