Roadmap

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

v0.16.0 — Fusion Default ✓ Released 2026-08-29

• fusion is now the default recall strategy (#1123) — weighted RRF of the vector and hybrid rankings; LongMemEval 500Q R@5 0.946 vs 0.854 hybrid (+9.2 pts), zero config needed
• Public benchmark page: dual-metric view (recall_any + strict recall_all) with comparison chart (#1141)
• Full-release validation: 500Q pure-default run, binary built from the exact release SHA
• Post-release fixes: repair --reembed handles NULL-embedding rows (#1146), CLA check bot allowlist (#1132), repair report reflects final index state (#1149)

v0.15.0 — Trust Across Surfaces, Portable Store ✓ Released 2026-08-19

• Supersession workflow (#1069) — mark a stale decision as superseded by a newer one; recall flags superseded entries
• Structural export/import (#1068) — full-store round-trip (rooms, graph, edges, documents, timeline) without loss
• UUIDv7 IDs for new memories — time-ordered at the front
• MCP uteke_get + uteke_update (#1067), richer MCP outputs (#1066)
• uteke_dream runs dry-run first — destructive passes need explicit confirmation (#1065)
• /export keeps namespace attribution (#1064); recall cache score parity (#1063)

v0.14.0–v0.14.3 — Hybrid Default, Scene Extraction, Surface Parity ✓ Released 2026-08-14–15

• Hybrid (RRF) is now the default recall strategy (#1005) — recall@5 85.4% → 98.0%; revert via strategy = "vector" in [recall]
• Scene-segmented LLM extraction with priority scoring (#1009) — scene:<topic> tags, semantic types, importance values
• GET /lifecycle/deprecated + GET /guide endpoints (#1007, #1010)
• Source provenance for extracted memories (#1012, #1013)
• Chunker fixes via mutation testing: heading duplication + multibyte infinite loop (#1024) — mutation score 50% → 97%
• uteke-cli crates.io publish restored + release verify step (#1021, #1030)
• HTTP & MCP recall strategy parity — default hybrid, loud 400/JSON-RPC errors on invalid values (#1034, #1035, #1038)
• Cross-compilation for Android/iOS targets (#1014), batch-mode update check skip (#1006), embedding pre-truncation (#1002)
• Release notes heredoc hardening — template file, no shell expansion (#1041)

v0.13.0–v0.13.2 — Safe Lifecycle, Offline Extraction, Security Hardening ✓ Released 2026-08-08–11

• Safe memory lifecycle (#928–#937) — ACTIVE → DEPRECATED (30-day TTL, restorable) → PRUNED; soft-delete default, uteke lifecycle commands + HTTP endpoints
• Offline rule-based extractor as default (#890) — no LLM/API key needed for basic extraction
• Persistent embedding cache + lazy ONNX load (#896), repair --reembed/--rebuild (#919, #901)
• Update check on all surfaces (#990), NULL-embedding startup crash fix (#992, #993)
• Security batch (#966–#987): command injection fix, checksum abort, pinned-memory data-loss guard, UTF-8 panics, zero unsafe blocks, N+1 elimination, case-sensitive routes
• Branding refresh (#991) — "One memory. Every agent. Zero cloud."

v0.12.0 — Full-text Doc Search, Junction Tools, Windows Installer ✓ Released 2026-07-22

• FTS5 indexes full document content body — every word searchable, not just titles (#783)
• Room↔Document junction in CLI & MCP — uteke room doc add/remove/list, 4 new MCP tools (#789)
• Renamed MCP tool: uteke_room_document → uteke_room_summary_document (#735)
• Windows PowerShell installer (install.ps1) + .zip binary release (#781)
• POST /memory/feedback trust scoring endpoint (#718)
• Bug fixes: deprecated memory leak in vector search, doc search scope

v0.11.0 — Hardening & Governance ✓ Released 2026-07-10

• POST /doc/move #[serde(deny_unknown_fields)] — silent data loss fix (#833)
• Recall score reports cosine similarity instead of RRF rank (#831)
• MCP/CLI store mismatch — hardcoded ~/.uteke paths fixed (#830)
• POST /consolidate accepts string threshold via flex_f32 (#826)
• API route drift guard test — 46 handler paths verified (#829)
• Open-source governance docs — Code of Conduct, PR template, YAML issue forms (#834)

v0.10.0 — Rust Edition 2024 ✓ Released 2026-07-05

• Rust edition 2021 → 2024 — minimum Rust 1.85 (#755)
• POST /room/remember endpoint — store + link to room in one call (#762)
• DELETE /forget returns 404 for non-existent IDs (#762)
• Clippy lints after edition upgrade

v0.9.0 — Onboarding Wizard & API Versioning ✓ Released 2026-06-30

• uteke onboard — interactive onboarding wizard (#743)
• API URL versioning /api/v1 and /api/v2 (#741)
• Configurable dream pipeline thresholds (#742)
• SECURITY.md and PR template (#746)
• Deprecated memory leak in vector search fix (#748)
• Windows ERROR_LOCK_VIOLATION fix (#747)
• Embedding model streaming download (#740)
• Contributors: @webhop123, @gnoviawan

v0.8.0 — Trust Scoring & Cross-Entity Linking ✓ Released 2026-06-29

• PUT /memory — partial memory updates with embedding regen (#676)
• Room↔Document junction table — schema v15 (#689)
• Memory↔Document linking via [[doc-slug]] wikilinks (#691)
• Trust scoring — uteke feedback helpful/unhelpful (#718)
• Jaccard token similarity reranking (#719)
• Auto-contradiction scan as Dream pipeline Phase 4 (#720)
• Cross-entity enrichment in recall (--enrich) (#703)
• FTS5 indexes memory_type — schema v14 (#662)

v0.7.0 — HTTP API Expansion & Global Documents ✓ Released 2026-06-28

• Project-aware memory tagging (#616)
• OpenCode init support (#612)
• Maintenance HTTP endpoints — /prune, /consolidate, /aging (#607)
• Monitoring HTTP endpoints — /importance, /orphans, /rebuild-backlinks (#608)
• Extract/Import/Export HTTP endpoints (#604–#606)
• Document partial update CLI + HTTP (#589)
• MCP: pin/unpin, 6 room tools, tag management, doc update+move tools
• uteke upgrade command (renamed from uteke update) (#603)
• Documents now global — no namespace isolation, schema v13 (#614)

v0.5.0 — LLM Extraction & Hermes Integration ✓ Released 2026-06-27

• #46 LLM fact extraction on import ✓ Done
  • uteke import --extract distills noisy text into atomic facts
• Hermes memory-provider plugin ✓ Done
  • Automatic recall + extraction, no daemon required
• Configurable embedding endpoint path ✓ Done
• Default max_seq_length 256 → 2048 ✓ Done
• Public store() accessor for downstream crates ✓ Done
• rusqlite 0.31 → 0.40 upgrade ✓ Done

v0.6.0 — Batch Import & Embed Fallback ✓ Released 2026-06-30

• Batch directory import (--batch-dir) ✓ Done
  • Auto-detection: .md → document, .txt/.jsonl → memory extraction
  • --as-doc/--as-memory override, --recursive, --dry-run
• Cloud embedding fallback ✓ Done
  • Optional [embed_fallback] config for cloud API failover
  • Dimension validation at startup
• Mode C (shell hook) docs ✓ Done
• Migration upgrade regression test ✓ Done
• Schema migration fix (#492) ✓ Done

v0.4.x — Polish & Stability ✓ Released 2026-06-22–24

• Hierarchical documents — depth-10 tree engine ✓ Done
• Hybrid document search (semantic + FTS5 + RRF) ✓ Done
• MCP tools: uteke_context, uteke_dream ✓ Done
• Auto-dream (3-day cycle) + configurable maintenance daemon ✓ Done
• Dedup on insert (cosine ≥ 0.95) + pinned memory protection ✓ Done
• Binary version mismatch fix, release workflow fixes ✓ Done

v0.3.0 — Graph RAG ✓ Released 2026-06-21

• #401 Cosine auto-linking + dedup ✓ Done
  • similar_to (≥0.80) and possible_duplicate (≥0.92) edges
• #404 Configurable limits ✓ Done
  • Env vars + [limits] config section
• #405 Markdown/prose chunker ✓ Done
  • Heading-aware splitting, code block protection
• #406 Document engine ✓ Done
  • Wiki/knowledge base, schema v11, documents + document_chunks
• #407 Embed-aware chunking ✓ Done
  • Chunk size from embedder.max_seq_len()
• #408 /graph API endpoint ✓ Done
  • Nodes + edges JSON for visualization
• #409 View-only API key ✓ Done
  • Dual-role tokens (admin + read-only)
• #410 Hermes plugin room_remember ✓ Done
• #411 Document CLI commands ✓ Done
  • uteke doc create/get/list/delete/export

v0.2.1 — DX & Ecosystem ✓ Released 2026-06-21

• #337 OpenAI + Ollama embedding backends ✓ Done
  • reqwest-based HTTP backends, ONNX stays default
• #346 Typed auto-edges — self-wiring knowledge graph ✓ Done
  • Auto-wired memory edges on every remember()
• #347 Timeline event tracking ✓ Done
  • Per-memory chronological audit log
• #348 Citation & source attribution ✓ Done
  • source, source_type columns (schema v10)
• #349 Memory type formalization ✓ Done
  • Typed categories with auto-inference
• #350 Backlink auto-generation ✓ Done
  • Bidirectional memory edges
• #351 Orphan detection ✓ Done
  • Find disconnected memories in the graph
• #352 Salience + recency dual-axis recall ✓ Done
  • Boost by memory type and age
• #353 Dream cycle ✓ Done
  • Coordinated maintenance pipeline (lint → backlinks → dedup → orphans)
• #381 MCP Streamable HTTP transport ✓ Done
  • Protocol version 2025-06-18, POST /mcp endpoint
• #385 Hermes plugin auto-install ✓ Done
  • Direct install to ~/.hermes/plugins/uteke-tool/
• #393 uteke room create command ✓ Done
• #395 Room operations in Hermes plugin ✓ Done
• #402 Fix: plugin missing __init__.py ✓ Done
• #403 Fix: contradictory server log + DB size label ✓ Done

v0.1.0 — Rooms, Intelligence & Pluggability ✓ Done

• #292 Time-travel queries ✓ Done
  • Recall/list at specific point in time (--at flag)
  • Temporal validity filter: created_at, valid_from/valid_until, deprecated
• #249 Pluggable embedding models ✓ Done
  • Embedder trait abstraction for multiple backends
  • ONNX backend (default), future: OpenAI, Ollama
• #306 Room document view ✓ Done
  • Structured document output grouped by memory_type
• #305 Room summary ✓ Done
  • LLM-free room summary via tag clustering
• #304 Semantic room recall ✓ Done
  • recall_room_semantic() with --query flag
• #184 Normalize tags junction table ✓ Done
  • Schema v5, memory_tags table, O(log n) lookups
• #252 Configurable recall threshold ✓ Done
  • --min, --strict, [recall] min_score config
• #286 Room-based memory ✓ Done
  • Full room management with author attribution
• #181 Recall cache optimization ✓ Done
  • LRU cache with TTL, --context output format
• #246 Relationship graph ✓ Done
  • --related --depth N BFS traversal
• #247 Smart memory decay ✓ Done
  • Composite importance scoring, pin/unpin
• #49 Benchmark suite ✓ Done
  • uteke bench command + LongMemEval retrieval harness
• #316 LongMemEval harness ✓ Done
  • Retrieval accuracy evaluation (Recall@k, NDCG@k)

v0.0.15 — CLI Performance ✓ Done

• #185 Lazy ONNX model loading
  • CLI cold start: ~3s → ~20ms for non-embedding commands
  • Model loaded on first use (remember, recall, search)
• #131 Modular CLI refactor
  • CLI args extracted to cli.rs, logging to logging.rs
• Release workflow decoupled: parallel builds + crates.io publish

v0.0.14 — Security & Polish ✓ Done

• #134 Binary checksums & file permissions
  • SHA256 checksum verification for ONNX model files
  • Owner-only permissions (0700/0600) on database and model dirs
• #277 Indonesian README translation
• #100 TLS & Reverse Proxy docs
• Crates.io metadata in all Cargo.toml files

v0.0.13 — Search & Concurrency ✓ Done

• #250 FTS5 hybrid search with RRF
  • FTS5 full-text search as parallel retrieval channel
  • Reciprocal Rank Fusion (k=60) merges vector + FTS5 results
  • Schema migration v1→v2 (auto, zero data loss)
• #251 Metadata enrichment via CLI flags
  • --entity, --category, --meta key:value,... on remember
• #209 Concurrent reads via RwLock
  • Mutex<VectorIndex> → RwLock<VectorIndex> for read-heavy workload
• #139 Vector index consistency
  • Atomic save for .keys sidecar file (temp + rename)

v0.0.10 — Codebase Quality ✓ Done

• #187 Split commands.rs into per-command modules
• #186 Split store.rs into focused modules
• #178 Remove all Hermes branding
• #196 Address all Cora code review findings

v0.0.9 — Website Migration ✓ Done

• #194 Website migrated to VitePress

v0.0.8 — Stability & Architecture ✓ Done

• #130 Architecture: module split, #132 Input validation, #134 Binary checksums
• #138 Schema versioning, #144 Error handling rewrite
• Memory consolidation, import/export (JSONL)

v0.0.7 — Core Stability ✓ Done

• #120 Tag queries → json_each(), #127 Configurable tier thresholds

v0.0.5–v0.0.6 — Docker & Hardening ✓ Done

• #95 UTEKE_HOME, #97 Dockerfile, #99 GHCR

v0.0.4 — Server Mode & Intelligence ✓ Done

• #54 Daemon/server mode, #51 Temporal facts, #52 Consolidation

v0.0.2–v0.0.3 — Foundation ✓ Done

• #40 usearch persistent index, #39 Multi-agent namespaces
• #38 Tiered memory, #42 Tag management

---

v0.2.0 — Knowledge Graph & Scale ✓ Done

• #317 SQLite graph storage — optional knowledge graph mode
• #245 Code-aware embedding with AST chunking — entity extraction from code
• #293 Structured memory — nested JSON content
• #46 Import from external knowledge sources
• #55 Hermes plugin — uteke integration