Easy Memory product spine
Local-first semantic memory + structured task orchestration as one product. Memory engine indexes Obsidian; task-graph runs DAG workflows over MCP; launcher unifies both.
Artifact 16.07.26 · Scale: 2 engines · ~107 source files · money priority 1
How to make / save money
- Ship local agent infra (open-core memory + tasks) without cloud lock-in
- MCP-native: sell into Claude Code / Devin / multi-agent stacks
- Optional Fileverse hybrid for paid coordination later
- Cost tracking in task-graph is a B2B accounting hook
Reusable learnings
- Two engines, one launcher: product feels unified without merge
- Local-first embeddings + SQLite beat cloud search for private vaults
- Phases/gates/claiming prevent multi-agent thrash
- REPO_HANDOFF_MINIMUM is the cleanup contract
Workstreams / landmarks
Memory engine (memory/)
Semantic + BM25 + graph/facts; local Xenova embeddings; vault watcher
Task graph engine (task-graph/)
DAG tasks, workflows, gates, atomic claim, cost/time accounting
Launcher + bridge + vault
init/build/search/task-status; remember_task_outcome; starter vault templates
Read order: root README → REPO_HANDOFF_MINIMUM → memory README → task-graph README.