Roadmap

Complete

Storage & Data Model

• Entry data model with structured metadata and type-specific extensions
• DistilleryStore protocol — async storage abstraction enabling backend migration
• DuckDB backend with VSS extension and HNSW index (cosine similarity)
• Configurable embedding providers (Jina v3 default, OpenAI adapter)
• Embedding model lock via _meta table — prevents mixed-model corruption
• MCP server with 22 tools over stdio and streamable-HTTP
• distillery.yaml config system with validation

Core Skills

• /distill — session knowledge capture with duplicate detection
• /recall — semantic search with provenance display
• /pour — multi-pass retrieval + structured synthesis with citations
• /bookmark — URL fetch, auto-summarize, store with dedup check
• /minutes — meeting notes with --update (append) and --list modes
• Shared CONVENTIONS.md — author/project identification, error handling patterns

Classification Pipeline

• ClassificationEngine — LLM prompt-based type assignment with confidence scoring
• DeduplicationChecker — skip/merge/link/create at configurable thresholds
• /classify skill — classify by ID, batch inbox, review queue triage

Quality & Observability

• Implicit retrieval feedback + distillery_quality tool
• Stale entry detection — distillery_stale tool
• Conflict detection — distillery_check_conflicts tool
• Usage metrics dashboard — distillery_metrics tool

Infrastructure

• FastMCP 2.x/3.x with @server.tool decorators
• Hierarchical tag namespace with validation and distillery_tag_tree tool
• 12 entry types including person, project, digest, github, feed
• distillery_type_schemas MCP tool for schema discovery

Team Access

• HTTP transport — distillery-mcp --transport http
• GitHub OAuth — team authentication via FastMCP GitHubProvider
• Prefect Horizon deployment (MotherDuck)
• Fly.io deployment with persistent DuckDB on volume
• Namespace taxonomy — hierarchical, validated tag system

Ambient Intelligence

• /radar — ambient feed digest with AI source suggestions
• /watch — add/remove/list monitored feed sources
• /tune — adjust relevance thresholds and trust weights
• Feed polling architecture — FeedPoller with configurable intervals
• Source adapters — GitHub events (REST API) and RSS/Atom
• Relevance scoring pipeline — embedding-based cosine similarity
• Interest extractor — mines entries for tags, domains, repos, expertise

Onboarding

• /setup skill — MCP connectivity wizard, auto-poll configuration

---

Planned

New Skills

• /whois — evidence-backed expertise map
• /investigate — deep domain context builder
• /digest — team activity summaries
• /briefing — team knowledge dashboard
• /process — batch classify + digest + stale detection pipeline
• /gh-sync — GitHub issue/PR knowledge tracking

Infrastructure

• Elasticsearch migration — hybrid search (BM25 + kNN + RRF)
• Access control — team/private visibility flag

---

Deferred

• LangGraph evaluation for complex skill orchestration
• CODE pipeline formalization for team workflows
• Web UI or REST API
• Multi-team support and cross-team knowledge sharing
• Re-embedding migration tooling

---

Technology Stack

Layer	Current	Planned
Interface	Claude Code skills	Same
Transport	stdio + streamable-HTTP	Same
Auth	GitHub OAuth (FastMCP)	+ multi-team RBAC
Storage	DuckDB + VSS / MotherDuck	+ Elasticsearch
Embeddings	Jina v3 / OpenAI	+ ES native
Language	Python 3.11+	Same
Hosting	Local / Fly.io / Prefect Horizon	Same