Building in the open
The Myco roadmap.
Myco Brain ships a deliberately small, rock-solid core and grows in the open. This is where things are headed. Dates are intentionally omitted — this is direction, not a contract.
Canonical version lives in the repo: ROADMAP.md · Issues and PRs that move it forward are welcome
Now
Shipping today
Live in v1.0 — clone the repo and run it.
Self-hosted MCP memory server
11 brain_* tools over MCP, backed by your own Postgres 16 + pgvector. Boots with one docker compose up — zero configuration, seeded workspace included.
v1.0.0 →Keyless full-text search & ingestion
BM25 search and document ingestion work with zero API keys. Add keys later only for vector search or the most accurate graph extraction.
Content-hash deduplication
Identical content is rejected on write, so re-ingesting a folder never multiplies your memory.
Provenance on every fact
Every accepted fact links back to its source — ask brain_why and get the evidence chain, with how it was extracted and how many sources support it.
Bulk ingest
Point mycobrain-ingest at a local folder or a GitHub repo and your text files become searchable, source-traceable memory (binaries and noise directories are skipped).
Knowledge graph — local or hosted
Entity extraction, entity resolution (duplicate names collapse into one node), and entity-to-entity relationships. Build it fully locally with Ollama — no API key — or with Anthropic for the most accurate graph.
Next
Near-term
Actively being worked on. This is where our weeks go.
Local vector search
Semantic search with no API key, via local embeddings. Today, vector search uses an OpenAI key — full-text is already keyless.
Published, reproducible benchmark
An open memory-quality benchmark you can run yourself, so retrieval quality is measured, not asserted.
Richer relationship extraction
Better predicate accuracy and direction, and a recommended local model profile for higher-quality graphs.
Compounding confidence (phase 1)
Surface how much independent evidence supports each fact — “seen in N sources” as a visible signal.
Later
The bigger bets
Committed direction, sequenced behind the near-term work.
Dynamic schema (full)
Phase 1 ships now: the extraction worker proposes new entity kinds and relationship types it observes in your data, surfaced by brain_stats — promotion is manual. The full version auto-promotes proposals under confidence rules with an audit trail, so the schema evolves with your domain.
Compounding confidence (full)
A fact's confidence rises as independent evidence accumulates and falls when contradicted. Older facts are superseded, not silently overwritten — memory that gets more reliable the more it sees.
More ingestion sources
First-class connectors beyond files and repos.
Managed cloud
Hosted, multi-tenant Myco for teams that would rather not operate Postgres themselves. Currently waitlist-only — not generally available.
Join the waitlist →Principles that won’t change.
The roadmap flexes with what people build on Myco. These don’t.
- ✓Postgres is the source of truth; the LLM is an advisor, not the database.
- ✓Writes are deterministic, deduplicated, and traceable.
- ✓Your data stays in plain Postgres tables you can inspect, export, and own.
- ✓The core stays self-hostable and open source (Apache-2.0).
This roadmap is shaped by what people build.
Have a use case or a feature you need? Open an issue — and if you’d rather not run Postgres yourself, the managed cloud waitlist is open.