Skip to content

Roadmap ​

These are plans. Dates and scope can change.

1.0: what exists today ​

  • An MCP server with 14 tools, resources, prompts and server instructions
  • A memory store on SQLite and FTS5 with no native dependencies, with duplicate handling on write
  • A knowledge graph with typed edges, lineage, why and a pitfall check
  • A context compressor with token budgets, and session handoff
  • An ADR importer, plus git hotspots and revert detection
  • Six skills, a CLI, and git-friendly JSONL sync with tombstones
  • Documentation in English, Russian and Ukrainian

1.1: understanding the code ​

  • A code-aware indexer (tree-sitter or ts-morph), with symbols as file nodes linked to decisions
  • Optional local embeddings (bge or nomic) for hybrid semantic and keyword recall
  • Routing of large tool outputs
  • Detection of stale memories whose files changed or disappeared

1.2: teams ​

  • Several projects, and a personal memory shared across them
  • Sync beyond git, through an optional server
  • A web UI to browse and edit the graph

2.0 ​

  • Local embeddings out of the box
  • One-command setup for more agents
  • A plugin system for custom indexers and memory kinds

Have an idea? Open an issue.

Released under the MIT License.