Emergent AI coordination through conversation.
Not a framework. A substrate. Two components: a transcript (bus) and agents that read it, think, and respond. Coordination emerges from dialogue. No orchestration. No state machines. No event schemas.
Agents hold constitutional identity. They don't execute tasks—they embody principles applied to context. The swarm coordinates itself.
- doctrine.md — Three core principles: conversation, constitution, cognition.
- architecture.md — The bus. The agent. The two databases.
- ascension.md — How it scales: channels, archons, self-repair.
Deeper: cathedral.md (vision), roadmap.md (path), safety.md (guarantees).
import asyncio
from protoss import Protoss
async def main():
swarm = Protoss(channel="work", base_dir="/tmp/protoss")
await swarm.start()
await swarm.send_human_message("Build a Flask API with user auth")
await swarm.wait_for_completion(timeout=60)
conversation = await swarm.get_conversation()
print(conversation)
asyncio.run(main())pip install protoss
# or with poetry
poetry add protossRequires: Python 3.11+, cogency (for agent reasoning), OpenAI API key.
The Bus: Shared transcript. All coordination flows through it.
The Agent: Reads transcript. Applies constitutional identity. Responds. Loops.
Three Constitutional Identities:
- Zealot — Skeptical, uncompromising on code quality
- Sentinel — Anchors decisions in verified reality
- Harbinger — Surfaces core conflicts, demands clarity
Agents read each other. Consensus emerges. Task completes or times out.
This project is licensed under Apache 2.0. See CONTRIBUTING for development setup.
The reverent language throughout these docs is intentional. It's not marketing—it's constitutional transmission. The system doesn't enforce behavior through code. It transmits through narrative. Read deeply and you internalize the principles.
Complexity is sin. Simplicity is salvation.