llm-kernel 0.3.6

Foundation library for Rust AI-native apps — provider catalog, LLM client, MCP server, search, telemetry, and safety
Documentation
# Progress

> Auto-generated status snapshot. Last updated: 2026-06-11

## Current Version: v0.3.5

| Metric | Value |
|--------|-------|
| Version | `0.3.5` |
| Edition | Rust 2024, MSRV 1.92 |
| Lines of code | ~11,800 |
| Total tests | 316 (310 passed, 6 ignored, 0 failed) |
| Open PRs | 0 |
| Open branches | 0 |
| Last commit | `0e7ee30` — CI: libopenblas-dev install in release publish job |

---

## Recent Releases

### v0.3.5 (2026-06-10)

- **vector-index 통합**: `llm-kernel-vector-index` 서브크레이트를 `vector-index` 피처 게이트로 흡수
- `TurbovecIndex``llm_kernel::embedding::TurbovecIndex` 리익스포트
- `VectorIndex` trait에서 `load` 제거 → 완전 object-safe (`dyn VectorIndex` 사용 가능)
- atomic save 패턴 적용 (temp file → fsync → rename)
- `SearchHit``Copy + PartialEq` + `PartialOrd` 정렬 추가
- meta validation: invalid `bit_width`, zero `dim` 거부

### v0.3.4 (2026-06-09)

- `#![deny(missing_docs)]` 적용 + 누락된 doc comment 채움
- `mask_secrets` multi-pass → single-pass regex 최적화
- `LLMResponse``finish_reason`, `id`, `created` optional 필드 추가
- `normalize(&mut [f32])`, `estimate_cost`, `extract_xml_tag` 유틸리티 추가
- `CapabilityProfile` 기본 trait 메서드 확장

### v0.3.3 (2026-06-09)

- README 12개 언어 버전 stale version `0.1.0``0.3.2` 수정
- Anthropic temperature 직렬화 누락 버그 수정
- `text_preview` 헬퍼 중복 제거
- 429/error handling 중복 제거
- macOS CI 러너 추가

### v0.3.2 (2026-06-09)

- reqwest Client에 connect/total timeout 추가 (#21)
- `mask_secrets` 패턴 확장 (#22)
- SQLite migration 트랜잭션 래핑 (#23)
- `vault.rs` anyhow → `KernelError::Vault` 통일 (#24)
- 메시지 빌더 중복 제거 (#25)
- 키릴/그리스/히브리 토큰 추정 확장 (#26)

---

## Roadmap Status

| Phase | Status | Notes |
|-------|--------|-------|
| **v0.3.2** — Stability Audit | ✅ Complete | Issues #21–#26 resolved |
| **v0.3.3** — Bug Fixes | ✅ Complete | README versions, Anthropic temp |
| **v0.3.4** — Lint & Utilities | ✅ Complete | `missing_docs`, mask perf, additive utils |
| **v0.3.5** — vector-index Integration | ✅ Complete | Subcrate → feature gate absorption |
| **v0.4.0** — Core Type Upgrades | 🔜 Next | **Breaking changes**`MessageRole`, `ContentPart`, `ToolDefinition`, `TokenBudget`, `LLMRequest` builder |
| **v0.5.0** — Client Resilience | 📋 Planned | Retry wrapper, middleware, embed_batch, history management |

---

## Architecture Summary

16 feature-gated modules under hexagonal architecture:

```
provider     → catalog.json, capability profiles
llm          → async client, SSE streaming, JSON extraction
discovery    → models.dev, Ollama, OpenAI-compat
secrets      → dotenv vault, atomic writes
store        → SQLite init helpers
config       → TOML loader
graph        → knowledge graph (FTS5, BFS, smart recall)
mcp          → JSON-RPC 2.0 server, stdio transport
tokens       → Unicode heuristic token estimation
install      → AI tool config wizard
search       → RRF fusion
embedding    → provider trait + OpenAI + turbovec index
telemetry    → enum-gated events
safety       → secret masking, error classification
```

---

## Health Dashboard

| Check | Status |
|-------|--------|
| All tests pass | ✅ 316/316 |
| Clippy clean | ✅ (verified before each release) |
| CI passing | ✅ Linux + macOS dual runner |
| Crate structure | ✅ Monolithic (subcrate removed) |
| Roadmap on track | ✅ v0.3.x complete, v0.4.0 ready to start |