Implement the Agent Graph system - a DAG-based conversation model that unifies navigation, forking, joining, and multi-agent coordination.
See: Omni/Agent/Graph/DESIGN.md for full design.
Key Features
- DAG of conversation nodes (not linear sessions)
- Humans and agents have identical capabilities via same API
- Fork/join for parallel agent work
- Dynamic context from graph position + memory
- Auto-summarization when paths get long
Non-Goals (v1)
- Pruning/garbage collection
- Distributed graphs
- Real-time collaboration
FRAMING UPDATE: One agent, many thought-lines (not multiple agents). Fork = pursue parallel line of thought. Join = synthesize thoughts. All threads share identity + memory.