Roadmap

Demand-gated — we build what people actually use. Track progress on GitHub Issues.

v0.0.2 — Core Engine ✓ Done

• #40 Persistent vector index (usearch) ✓ Done
• #39 Multi-agent namespaces ✓ Done
• #38 Tiered memory (Hot/Warm/Cold) ✓ Done
• #37 Health checks (doctor, verify, repair) ✓ Done
• #53 Website (landing + docs) ✓ Done

v0.0.3 — Tag Management & Operations ✓ Done

• #42 Tag-based filtering and management ✓ Done
• #44 Memory aging with auto-cleanup ✓ Done
• #47 Shell hook for auto-context loading ✓ Done
• #48 Configuration file (uteke.toml) ✓ Done
• #31 Graceful shutdown (SIGINT handler) ✓ Done
• #32 File logging with daily rotation ✓ Done
• #41 Upgrade Node.js 20→24 in CI ✓ Done

v0.0.4 — Server Mode & Intelligence ✓ Done

• #43 Bulk memory operations (tag, cold, all) ✓ Done
• #45 Namespace switching & defaults ✓ Done
• #54 Daemon/server mode — warm recall <50ms ✓ Done
• #54 CLI auto-routes to server (21ms recall) ✓ Done
• #51 Auto-forget & temporal facts ✓ Done
• #52 Consolidation & deduplication ✓ Done
• #86 Cora AI code review in CI ✓ Done

v0.0.5 — Docker & Deployment ✓ Done

• #95 UTEKE_HOME env var for custom data directory ✓ Done
• #96 uteke-serve reads uteke.toml, default host 0.0.0.0 ✓ Done
• #97 Dockerfile (multi-stage build, model baked in) ✓ Done
• #99 Docker image build & push on release (GHCR) ✓ Done

v0.0.6 — Hardening & Fixes ✓ Done

• JSON output omits embedding vector (~3KB savings per memory) ✓ Done
• import() now persists vector index (fixes data loss on restart) ✓ Done
• Docker: non-root container with USER uteke ✓ Done
• Dependabot: automated dependency updates ✓ Done

v0.0.7 — Core Stability ✓ Done

• #120 Tag queries: LIKE → json_each() for exact matching ✓ Done
• #127 Configurable tier thresholds (hot_days, warm_days) ✓ Done
• #129 Test coverage: 34 → 94 tests ✓ Done
• Tag substring false positive fix ("rust" ≠ "rustacean") ✓ Done

v0.0.8 — Stability & Architecture ✓ Done

• #130 Architecture: module split (1471→352 lib.rs) ✓ Done
• #132 Input validation (max 10K chars, 20 tags, 1MB payload) ✓ Done
• #134 Binary checksums (SHA256) + verify-checksums subcommand ✓ Done
• #138 Schema versioning + migration framework ✓ Done
• #144 Error handling rewrite — sanitized user-friendly messages ✓ Done
• #137 Python wrapper: 7→21 methods with type hints ✓ Done
• #49 Memory benchmark binary (library-level timing) ✓ Done
• Memory consolidation command ✓ Done
• Import/Export (JSONL backup and restore) ✓ Done
• #174 README overhaul + landing page refresh (GTM Phase 1+2) ✓ Done

v0.0.9 — Website Migration ✓ Done

• #194 Website migrated to VitePress ✓ Done
  • SvelteKit (3,750 LOC, 10 deps) → VitePress (1,300 LOC markdown, 2 deps)
  • Built-in full-text search
  • Build time: ~15s → ~6s
  • Brand theme (amber/dark) preserved

v0.0.10 — Codebase Quality ✓ Done

• #187 Split commands.rs into per-command modules ✓ Done
• #186 Split store.rs into focused modules ✓ Done
• #178 Remove all Hermes branding ✓ Done
• #196 Address all Cora code review findings ✓ Done
• Safe slice for deprecated IDs ✓ Done
• Index lock before SQLite write ✓ Done
• HTTP status checking in server proxy ✓ Done
• Aging cleanup filter fix ✓ Done
• Schema migration transactions ✓ Done
• Batch bulk deletes ✓ Done
• SQLite-first dual-write ✓ Done
• Embedding docs corrected (768d) ✓ Done
• Shell hook idempotency guards ✓ Done

v0.0.12 — Search & Concurrency ✓ Done Current

• #250 FTS5 hybrid search with RRF ✓ Done
  • FTS5 full-text search as parallel retrieval channel
  • Reciprocal Rank Fusion (k=60) merges vector + FTS5 results
  • RecallStrategy enum: hybrid (default), vector, fts5
  • Schema migration v1→v2 (auto, zero data loss)
  • Phrase search + token-OR fallback
  • Deprecated memories excluded from FTS5
• #251 Metadata enrichment via CLI flags ✓ Done
  • --entity, --category, --meta key:value,... on remember
  • Post-filter on recall and list by entity/category
  • Auto type detection for meta values (string/number/bool)
• #209 Concurrent reads via RwLock ✓ Done
  • Mutex<VectorIndex> → RwLock<VectorIndex> for read-heavy workload
  • Multiple concurrent recalls share read lock
  • Embedder lock scope minimized
• #139 Vector index consistency ✓ Done
  • Atomic save for .keys sidecar file (temp + rename)
  • insert() and build() now return Result (error propagation)
  • HashMap already used for key mapping

Phase A — Growth (100+ stars) Planned

Better embeddings, richer integrations, broader reach.

• Better embeddings (larger model option) Planned
• #46 Import from external knowledge sources Planned
• Python SDK (PyO3 bindings) Planned
• Editor integrations (VS Code, JetBrains) Planned
• Node.js SDK (NAPI) Planned

Phase B — Collaboration (500+ stars) Future

Optional cloud sync, team features, and gateway integrations.

• Cloud sync (opt-in, end-to-end encrypted) Future
• Team collaboration & shared namespaces Future
• API gateway integration (LangChain, CrewAI, others) Future
• #100 TLS support & reverse proxy documentation Future
• #101 API key authentication for uteke-serve Future

Phase C — Ecosystem (1000+ stars) Vision

Plugin ecosystem, advanced consolidation, community extensions.

• Plugin ecosystem (custom embedding, storage backends) Vision
• Advanced consolidation (cross-namespace, summarization) Vision
• Community extensions marketplace Vision
• Managed cloud API (optional paid tier) Vision