Release v0.0.15

What's New in v0.0.15

Changed

• CLI cold start: ~3s → ~20ms for non-embedding commands (#185)
  ONNX embedding model is now loaded lazily on first use. Commands like
  list, get, stats, tags, forget, namespace, aging, export,
  doctor, and verify start instantly without waiting for model load.
  Commands that need embedding (remember, recall, search) still take
  ~3s on first use per process invocation.
• Refactor CLI into modular structure (#131)
  CLI argument definitions extracted to cli.rs, logging setup to logging.rs.
  main.rs reduced from 449 to ~100 lines for easier maintenance.
• Release workflow now decoupled: crates.io publish runs in parallel with
  builds, GitHub Release only waits for builds. Single platform failure
  no longer blocks release.
• Shell hook scripts inlined into uteke-cli crate for crates.io compatibility.
• Added .cora.yaml config and pre-commit hook (Cora v0.5.0).

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.15-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.15-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.15-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.15-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/codecoradev/uteke/blob/main/CHANGELOG.md

Release v0.0.14

What's New in v0.0.14

Security

• Set owner-only file permissions (0700/0600) on database and model directories (#134)
• Add SHA256 checksum verification for downloaded ONNX model files (#134)
• Pin expected model checksums to detect corrupted/tampered downloads

Added

• Indonesian README translation (README.id.md) with language switcher (#277)
• TLS & Reverse Proxy documentation page (Caddy, Nginx, Cloudflare Tunnel) (#100)
• Crates.io metadata in all Cargo.toml files (#136)

Changed

• Server now handles requests concurrently via thread-per-request (#233)
  Uses Arc<Mutex<Uteke>> for safe shared access across threads.
• Contradiction threshold is now a parameter instead of hardcoded 0.65 (#253)
• Rename euclidean_to_cosine to cosine_distance_to_similarity (#232)
• 9 code quality improvements from Cora scan (#232)

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.14-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.14-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.14-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.14-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/codecoradev/uteke/blob/main/CHANGELOG.md

Release v0.0.13

What's New in v0.0.13

Added

• FTS5 hybrid search with RRF — Full-text search (FTS5) as parallel retrieval channel merged with vector search via Reciprocal Rank Fusion (RRF, k=60). New RecallStrategy enum: hybrid (default), vector, fts5. FTS5 virtual table auto-created; existing DBs get schema migration v1→v2. Phrase search + token-OR fallback. Deprecated memories excluded from FTS5. 6 new tests (#250, PR #261)
• Metadata enrichment via CLI flags — --entity, --category, --meta key:value,... on remember. Post-filter on recall and list by --entity and --category. parse_meta_pairs() with auto type detection (string/number/bool). JSON output includes metadata when present (#251, PR #262)
• Concurrent reads via RwLock — Mutex<VectorIndex> → RwLock<VectorIndex> for read-heavy workload. Multiple concurrent recalls share read lock. Embedder remains Mutex (ONNX tokenizer requires &mut self) (#209, PR #260)

Fixed

• Vector index consistency — Atomic save for .keys sidecar file (temp + rename). insert() and build() now return Result for error propagation (#139, PR #263)
• FTS5 BM25 score conversion — Negative unbounded BM25 values were always clamped to 0.0. Fixed to proper sigmoid-based normalization (PR #264)
• RRF normalization — .min(1.0) → .clamp(0.0, 1.0) with clearer math (PR #264)
• memories.remove().unwrap() — Replaced with .expect() for meaningful panic message (PR #264)
• Server-mode metadata support — remember via HTTP API now includes entity, category, and meta in request body (PR #264)
• Clippy collapsible_else_if — 2 pre-existing warnings fixed (PR #260)

Changed

• Repository transferred — ajianaz/uteke → codecoradev/uteke. All references updated across 16 files
• Cora Review CI — switched from local Infisical OIDC action to codecoradev/cora-review-action@v1 with GitHub Secrets. Removed .cora.yaml project config
• README simplified — 400 → 97 lines. Detailed content moved to VitePress docs (docs/architecture.md, docs/cli-reference.md, etc.)
• Roadmap cleaned — consolidated old versions, removed speculative Phase B/C phases
• CONTRIBUTING.md — added Cora CLI integration docs, CI checks table, architecture updated to 3 crates, Key Design Decisions section
• AGENT.md — new file with persistent AI agent context: critical rules, architecture, lessons learned, proven workflow
• docs/architecture.md — new VitePress page with system overview, data flow diagrams, performance benchmarks, design decisions
• docs/roadmap.md — v0.0.12 section added, old versions consolidated, "What's Next" list
• Star History chart added to README (cora-cli + uteke)

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.13-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.13-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.13-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.13-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/codecoradev/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/codecoradev/uteke/blob/main/CHANGELOG.md

Release v0.0.12

What's New in v0.0.12

Fixed

• TOCTOU race in tag operations — rename_tag and delete_tag now start transaction before SELECT, preventing lost updates from concurrent writers (#235)
• TOCTOU race in aging/prune — aging_cleanup and prune now delete by specific IDs instead of re-querying by criteria, preventing vector index orphans (#235)
• bulk_forget_ lock order* — All 3 bulk delete methods now acquire index lock before SQLite delete, matching the pattern from forget() (#236)
• Server 500 leaks internals — 500 responses now return generic "Internal server error" to client; full error logged server-side (#237)
• Server JSON fallback — json_response fallback now uses serde_json::json! instead of format!, preventing broken JSON (#237)
• Atomic write tmp naming — Temp files now named filename.tmp instead of fragile extension swapping (#238)

Added

• Store::delete_by_ids() — New method for atomic batch deletion by specific IDs

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.12-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.12-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.12-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.12-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.11

What's New in v0.0.11

Fixed

• [CRITICAL] Timestamp format mismatch — Aging/pruning queries never matched because SQLite datetime('now') format differs from stored RFC3339 timestamps. Now computes cutoffs in Rust using chrono (#221)
• Namespace=None inconsistency — Tag operations (tags_with_counts, rename_tag, delete_tag, count_by_tag) treated None as "default" namespace instead of "all namespaces". Now consistent with unique_tags behavior (#222)
• Non-atomic model file write — Model downloads now use atomic write (.tmp + rename) to prevent corrupt files on crash. Cleans up leftover .tmp files on startup (#225)
• uteke_home() panic — Replaced .expect() with Result return type to prevent crashes in minimal Docker/CI environments (#226)
• Server path matching — DELETE /forget now uses exact path match, preventing false matches on /forgetful etc. (#228)
• Query param parsing — Use splitn(2, '=') to preserve values containing = (#228)
• Missing CLI arg value — --host/--port without value now prints error instead of silently ignoring (#228)
• 404 path reflection — Generic "Not found" message instead of echoing request path (#228)
• SQLite/index inconsistency — forget() now acquires index lock before SQLite delete to narrow the inconsistency window (#231)
• Memory type validation — remember_typed() now validates memory_type against known variants (#229)

Added

• Security scanning workflow — New security.yml CI workflow with cargo audit + Trivy filesystem scan. Runs on push, PRs, and daily schedule (#177, #220)
• quinn-proto update — Updated to v0.11.14 fixing CVE-2026-31812 (DoS via crafted QUIC packet)
• Error::Generic variant — New error type for general-purpose errors

Changed

• uteke_home() returns Result — All callers updated to handle potential failure

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.11-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.11-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.11-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.11-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.10

What's New in v0.0.10

Fixed

• Safe slice for deprecated IDs — dep_id.get(..8).unwrap_or(dep_id) prevents panic on short IDs (#192)
• Index lock before SQLite write — Acquire vector index lock before any SQLite writes so lock failures are detected early, preventing false errors (#191)
• HTTP status checking — Server proxy now validates response status codes, returning proper error messages instead of silently accepting failures (#193)
• Aging cleanup filter — cleanup_aged now includes deprecated = 0 filter matching find_aged criteria (#189)
• Schema migration transactions — Each migration step + version stamp wrapped in SQLite transaction (#188)
• Batch bulk deletes — Replace N individual DELETE statements with single batched query for better performance (#190)

Changed

• Store module split — store.rs (2,065 LOC) split into 8 focused modules: schema, crud, tags, aging, bulk, vector, types, store (#179)
• Commands module split — commands.rs (820 LOC) split into 9 per-command modules (#180)
• SQLite-first dual-write — remember() now writes to SQLite before vector index, matching forget() pattern (#182)
• Embedding docs corrected — All docs now correctly state 768d (not 256d) for EmbeddingGemma (#183)
• Shell hook guards — Bash PROMPT_COMMAND and Zsh chpwd_functions now have idempotency guards (#143)
• Hermes branding removed — All product-specific branding replaced with generic names; only --namespace examples remain (#178)

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.10-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.10-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.10-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.10-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.8

What's New in v0.0.8

Added

• Architecture: module split — lib.rs (1471→352) and main.rs (1538→422) broken into focused modules: operations, maintenance, consolidate, error, types, import_export, commands, init, output, bench
• Input validation — Max content 10K chars, max 20 tags, max server payload 1MB (#132)
• Binary checksums — SHA256 checksums in release artifacts + verify-checksums subcommand (#134)
• Schema versioning — schema_version table + migration framework for future DB upgrades (#138)
• Error handling rewrite — Error enum with sanitized user-friendly messages, ~90 call sites migrated from raw rusqlite/usearch/ONNX errors (#144)
• Python wrapper expansion — 7→21 methods covering all CLI commands, namespace support, type hints, Google-style docstrings (#137)
• Memory benchmark — memory-bench binary for library-level timing across dataset sizes (#49)
• Memory consolidation — consolidate command to find and merge near-duplicate memories
• Import/Export — JSONL-based memory backup and restore via import / export commands

Changed

• Contradiction detection — Now read-only during check; deprecation happens after new memory is safely persisted (prevents data loss on insert failure) (#139)
• README — v0.0.8 badge, Design Philosophy section, Performance benchmarks

Fixed

• Deadlock in check_contradiction — Mutex re-acquire pattern fixed by separating read-only check from mutation (#139)

Security

• Error sanitization — Internal error details (file paths, SQL, model names, ONNX internals) no longer exposed to users (#144)

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.8-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.8-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.8-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.8-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Quick install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Pin a specific version
UTEKE_VERSION=v curl -fsSL https://raw.githubusercontent.com/ajianaz/uteke/main/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.7

What's New in v0.0.7

Added

• Tag storage: json_each() queries — All 8 tag query methods refactored from LIKE '%\"tag\"%' to json_each() for exact matching and performance (#120)
• Config wiring: tier thresholds — TierConfig struct with configurable hot_days, warm_days, hot_boost; Uteke::open_with_tier() accepts custom config (#127)
• Test coverage: 34 → 94 tests — Comprehensive tests for store, lib, and config modules (#129)
• Config tests — 7 new tests for merge_from_file, expand_tilde, set_namespace_in_toml (#129)

Changed

• MemoryTier::from_last_accessed() — Now accepts hot_days and warm_days parameters (was hardcoded 7/30)
• tags_with_counts() — N+1 query pattern replaced with single GROUP BY via json_each()
• unique_tags() — SQL returns individual tag values directly (no in-Rust JSON parsing)
• tier_counts() and bulk_delete_cold() — Now accept configurable threshold parameters

Fixed

• Tag substring false positives — Tag "rust" no longer matches memory tagged "rustacean"
• README configuration docs — Fixed config search paths, removed non-existent --config flag, corrected TOML format (#128)

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.7-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.7-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.7-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.7-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Install
curl -fsSL https://uteke.ajianaz.dev/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.6

What's New in v0.0.6

Fixed

• JSON output omits embedding vector — Memory.embedding now uses #[serde(skip_serializing, default)]
  • Reduces JSON response size by ~3KB per memory
  • Embeddings are populated programmatically via ONNX, not from JSON
• import() now persists vector index — previously imported memories were lost on restart because the index was never saved
• CI: Node.js 24 enforcement — added FORCE_JAVASCRIPT_ACTIONS_TO_NODE24 to all workflows
• Docker: non-root container — added USER uteke directive (uid/gid 1000) with owned /data directory
• CI: removed unused musl-tools install — targets are glibc only

Added

• Dependabot — automated dependency updates for cargo, GitHub Actions, and Docker

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.6-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.6-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.6-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.6-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Install
curl -fsSL https://uteke.ajianaz.dev/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md

Release v0.0.5

What's New in v0.0.5

Added

• UTEKE_HOME environment variable — single env var to override all dirs::home_dir() paths
  • Affects: database path (uteke.db), vector index (uteke_index.usearch), model cache (models/)
  • Default: $HOME/.uteke when not set
  • Essential for Docker volume mounts and custom data directories
• Server reads uteke.toml config — uteke-serve now respects configuration file
  • Reads [server] section: host, port
  • Default host changed to 0.0.0.0 (was 127.0.0.1) for Docker/network compatibility
  • Config loaded at startup, printed to logs
• Smart server fallback — CLI auto-falls back to local mode for server-unsupported commands
  • Commands not yet available via HTTP API gracefully fall back to local execution
  • No more error when server.enabled = true and command lacks server endpoint
• API parity — expanded remember endpoint — POST /remember now accepts all CLI fields
  • memory_type, detect_contradiction, valid_from, valid_until parameters
  • Returns contradiction detection result when enabled
• GET /memory endpoint — retrieve single memory by ID via GET /memory?id=<id>
• DELETE /forget bulk operations — DELETE /forget?all=true&cold=true for mass deletion
• Multi-stage Dockerfile — production-ready Docker image for uteke-serve
  • Base: debian:bookworm-slim (glibc/ONNX compatible)
  • Model baked into image at build time (~208MB total)
  • Non-root user, health check endpoint, configurable via env vars
• Docker image CI — automatic build and push to GHCR on release
  • Multi-platform: linux/amd64 + linux/arm64
  • Buildx with cache, tags: latest + version tag
• Release notes from CHANGELOG.md — dynamic extraction via awk (no hardcoded notes)

Changed

• Server default host: 127.0.0.1 → 0.0.0.0 (Docker/network accessible)
• Cora review action: hardcoded version → latest (auto-updates)

Fixed

• Pre-existing format issue: .to_string_lossy().to_string() chain cleaned up

📦 Binaries

Each archive contains two binaries:

• uteke — CLI tool
• uteke-serve — HTTP server daemon

Platform	File
Linux (x86_64)	uteke-v0.0.5-x86_64-unknown-linux-gnu.tar.gz
Linux (ARM64)	uteke-v0.0.5-aarch64-unknown-linux-gnu.tar.gz
macOS (Apple Silicon)	uteke-v0.0.5-aarch64-apple-darwin.tar.gz
Windows (x86_64)	uteke-v0.0.5-x86_64-pc-windows-msvc.zip

🚀 Quick Start

# Install
curl -fsSL https://uteke.ajianaz.dev/install.sh | sh

# Store a memory
uteke remember "Important context" --tags project

# Recall by meaning
uteke recall "what was that context?"

# Start server for fast AI agent access
uteke-serve --port 8767

Full changelog: https://github.com/ajianaz/uteke/blob/main/CHANGELOG.md