Collaboration Module
Multi-agent teamwork
Neurova supports multi-agent teamwork, letting agents with different specialisations divide the work and complete complex tasks together.
Four collaboration modes
| Mode | How it works | Typical use |
|---|---|---|
| Sequential | Agents process one after another in a pipeline | Draft → review → publish |
| Parallel | Several agents handle different subtasks at once | Analysing many dimensions simultaneously |
| Master-slave | One lead agent directs several subordinate agents | Project management: assigning and consolidating work |
| Consensus | Multiple agents judge independently and then vote | Risky decisions, multi-perspective verification |
The NeurFlow workflow engine
NeurFlow is Neurova's visual workflow engine, raising workflows to the level of a "visual IDE":
- Workflow canvas — drag-and-drop node orchestration (start / end / task / condition / parallel / subflow)
- Debugging walker — a mock engine with step and skip execution
- Version snapshots and rollback — content fingerprints plus rollbacks recorded in history
- Trigger system — cron scheduling and inbound webhooks (HMAC-SHA256 with a token bucket)
- Sub-workflows — nesting depth ≤ 5 with cycle detection
- Agent compilation — compiles a workflow into an AgentManifest, bridged to chat dispatch
Task board
- Draggable cards, custom columns, live updates
- Task state tracking: unassigned / in progress / done
API
The collaboration module is served through /api/v1/collaboration and /api/v1/neurflow:
- CRUD for projects, teams and tasks
- Workflow definitions and execution
- Group chat (real-time over SSE)