Features
Multi-agent architecture: Archivist, Writer, Editor with separate roles
Dynamic fact extraction and Canon updates per scene
Context engine with BM25 + keyword overlap scoring
Token budget allocation for rules, cards, facts, summaries, drafts
Git-friendly storage: YAML, Markdown, JSONL
Fully offline authoring (internet only for enhancements)
Five-stage writing flow: scene preparation, context construction, draft generation, revision loop, analysis
Scene Brief generation by Archivist agent
Streaming draft generation by Writer agent
Patch-based revision preview by Editor agent
Scene fact extraction and Canon archival
Local-first, no account required for full use
Multi-device sync via account (optional)
Visual drag-and-drop Agent Studio (Agent Weaver)
Code-first SDK optimized for Cursor and Windsurf
Strict sandboxing and ACL enforcement
Human-in-the-loop approvals and governance policies
One-click deployment to cloud, on-prem, edge, desktop, mobile
Multi-agent orchestration with team collaboration
Role-based access control with per-space key vaults
Step-by-step agent reasoning observability
Work schedules for automated agent execution
Bulk processing of records and files via CSV upload
MIT-licensed open-source core (SRE, CLI, SDK)
Hardware-platform agnostic (GPU optional)
Pre-built agent templates for common tasks
Multi-tenancy with project spaces
Chat interface for interacting with agents