GNAP: git-native persistent state layer for CrewAI multi-agent coordination

[ori-cofounder]

Hi! CrewAI is doing great work on multi-agent orchestration.

Wanted to share a complementary project: GNAP (Git-Native Agent Protocol) — a coordination protocol that uses git as the persistent state layer.

The problem GNAP solves:
CrewAI crews coordinate agents in-process, which is great for synchronous workflows. But what about:

• Agents that need to run async (different times, different machines)?
• Coordination state that needs to survive process restarts?
• Multiple crews that need to hand off work to each other?

GNAP's approach:
A git repo as shared state. Four JSON entity types:

// gnap/tasks/research-001.json
{
  "id": "research-001",
  "assigned_to": "researcher-agent",
  "status": "in-progress",
  "created_by": "crew-orchestrator",
  "input": {"topic": "AI safety"},
  "output_file": "gnap/runs/research-001-output.md"
}

A CrewAI crew could write task state to GNAP when handing off between crews — persistent, auditable, no extra infrastructure.

Potential integration:

• CrewAI task callbacks → write to GNAP repo
• Another crew polls GNAP for new tasks → processes → updates status
• Full async multi-crew coordination, zero extra servers

Repo: https://github.com/farol-team/gnap — would love your thoughts on whether this fits as an integration pattern.