I built CervellaSwarm because I was tired of AI assistants forgetting context and making unchecked mistakes. So I created a team of 16 specialized AI agents that work together - and more importantly, check each other's work.
- 1 Queen (orchestrator) routes tasks to the right specialist
- 12 Workers (frontend, backend, testing, security, etc.) execute
- 3 Guardians review everything before merge
The key insight: single AI assistants lack specialization and accountability.
CervellaSwarm gives you specialists AND a quality layer.
Honest limitations: This is v2.0.0-beta. Context windows are still limited. But when it works - and it does work - it feels like having a real dev team.
Built on Claude + Anthropic API. Apache-2.0 licensed. 287 sessions of development
(yes, CervellaSwarm built CervellaSwarm).
Would love feedback on:
- The multi-agent pattern (does it resonate?)
- Guardian review flow (too much overhead?)
Hi HN,
I built CervellaSwarm because I was tired of AI assistants forgetting context and making unchecked mistakes. So I created a team of 16 specialized AI agents that work together - and more importantly, check each other's work.
Installation: npm install -g cervellaswarm cervellaswarm init cervellaswarm task "add login page"
How it works:
- 1 Queen (orchestrator) routes tasks to the right specialist - 12 Workers (frontend, backend, testing, security, etc.) execute - 3 Guardians review everything before merge
The key insight: single AI assistants lack specialization and accountability. CervellaSwarm gives you specialists AND a quality layer.
Honest limitations: This is v2.0.0-beta. Context windows are still limited. But when it works - and it does work - it feels like having a real dev team.
Built on Claude + Anthropic API. Apache-2.0 licensed. 287 sessions of development (yes, CervellaSwarm built CervellaSwarm).
Would love feedback on: - The multi-agent pattern (does it resonate?) - Guardian review flow (too much overhead?)
Try it: npx cervellaswarm init