# Semantix

> Semantix is a self-evolving agent kernel that sits between any agent harness and its resources — extracting reusable semantic slices from past sessions, retrieving them with BM25, and injecting stable reuse blocks into later sessions. Open source (MIT), implemented in Go.
>
> Semantix 是一个自进化的 Agent Kernel 层：架设在 agent harness 与资源之间，从历史会话提取可复用的语义切片，用 BM25 检索，并向后续会话注入稳定的复用块。开源（MIT），Go 实现。

## Capabilities（能力清单）

能力按交付状态分层；路线图能力不视为已交付功能。

### Implemented（已实现）

- **Semantic slice extraction（语义切片提取）** — turns session JSONL into typed P/T/R slices with persistent stores. Source: [`kernel/slice`](https://github.com/Gnosil/semantix/tree/main/kernel/slice)（含测试）.
- **BM25 retrieval（BM25 检索）** — k1=1.2 / b=0.75, CJK-aware tokenization, `bm25` / `vector` / `hybrid` retrievers. Source: [`kernel/bm25`](https://github.com/Gnosil/semantix/tree/main/kernel/bm25)（含测试）.
- **CLI** — `semantix extract` / `search` / `lookup` / `inject` / `verify` / `usage` / `eval`；`verify` 提供离线回放命中率验证。快速开始：https://semantix.ensureok.ai/docs/quickstart（`docs/QUICKSTART.md`）.
- **Stable slice injection（基础跨会话注入）** — canonical-order injection of matched slices for provider prefix-cache stability. Source: [`kernel/inject`](https://github.com/Gnosil/semantix/tree/main/kernel/inject)（含测试）.

### Partial / experimental（部分实现 / 实验性）

- **L3 verified result reuse（L3 验证与复用）** — dependency and fingerprint verification pipeline in [`kernel/judge`](https://github.com/Gnosil/semantix/tree/main/kernel/judge)（含测试）；端到端基准结果尚未发布，不作为已达成性能声明.

### Roadmap（规划中）

- **Kernel scheduler（内核调度器）** — interface frozen, MVP stub. [`kernel/sched`](https://github.com/Gnosil/semantix/tree/main/kernel/sched).
- **Speculative prefetch（投机预取）** — interface frozen, MVP stub. [`kernel/prefetch`](https://github.com/Gnosil/semantix/tree/main/kernel/prefetch).
- **Evolution loop（自进化闭环）** — parameter-tuning component implemented（[`kernel/evolve`](https://github.com/Gnosil/semantix/tree/main/kernel/evolve)，含测试）；闭环集成待完成.

## Project

- [Semantix website](https://semantix.ensureok.ai/): Official homepage — positioning, features, components, roadmap, and install instructions.
- [About Semantix](https://semantix.ensureok.ai/about): Operator identity, official contact channels, and legal entity behind the project.
- [Contact Semantix](https://semantix.ensureok.ai/contact): Official project and website contact channel.
- [Semantix repository](https://github.com/Gnosil/semantix): Source code, tests, design documents, and issue tracker.

## Documentation

- [Documentation index](https://semantix.ensureok.ai/docs/): Entry point for all project documentation.
- [Install and first run](https://semantix.ensureok.ai/docs/quickstart): Release and source installation, followed by the smallest extract/search/inject loop.
- [Coding-agent integration](https://semantix.ensureok.ai/docs/agent-integration): Reasonix, Claude Code, and custom harness integration paths.
- [Configuration and scopes](https://semantix.ensureok.ai/docs/configuration): TOML precedence, stores, scopes, retrieval, and injection settings.
- [Extract and search](https://semantix.ensureok.ai/docs/extract-search): Session ingestion and BM25, vector, or hybrid retrieval.
- [Lookup and injection](https://semantix.ensureok.ai/docs/lookup-inject): Structured tool results and budgeted L2 context blocks.
- [Verification and observability](https://semantix.ensureok.ai/docs/verify-observe): Replay verification, usage accounting, dashboard, and doctor.
- [CLI reference](https://semantix.ensureok.ai/docs/cli-reference): Task-based command index, JSON envelope, and exit codes.
- [Slices and cache levels](https://semantix.ensureok.ai/docs/slices-and-cache): Semantic slices and the boundaries between L1, L2, and L3 reuse.
- [Retrieval and reuse safety](https://semantix.ensureok.ai/docs/retrieval-safety): Fusion, zones, fingerprints, judge gates, and fail-open behavior.
- [Scheduling and evolution](https://semantix.ensureok.ai/docs/scheduling-evolution): Round plans, conservative prefetch, feedback, and bounded adaptation.
- [Gateway deployment](https://semantix.ensureok.ai/docs/gateway): OpenAI-compatible proxy deployment and cache observation.
- [Storage and maintenance](https://semantix.ensureok.ai/docs/storage-maintenance): Backup, GC, archives, permissions, and sensitive-data boundaries.
- [Evidence and implementation status](https://semantix.ensureok.ai/docs/evidence-and-status): How to distinguish designs, tests, synthetic replay, and production evidence.
- [Semantix Project Overview](https://semantix.ensureok.ai/docs/profile): Concise Chinese overview — positioning, terminology, and current progress (M0).
- [Semantix Project Overview (EN)](https://semantix.ensureok.ai/docs/profile-en): Concise English overview of the project's purpose, terminology, and progress.
- [Understanding Semantix from Scratch](https://semantix.ensureok.ai/docs/guide): Chinese deep dive from first principles — ecosystem, architecture, and the four components plus evolution engine.
- [Understanding Semantix from Scratch (EN)](https://semantix.ensureok.ai/docs/guide-en): English deep dive covering the same architecture and mechanisms.
- [Semantix FAQ](https://semantix.ensureok.ai/docs/faq): Project positioning, semantic cache levels, evolution mechanisms, and contribution requirements.
- [Benchmarks and evidence](https://semantix.ensureok.ai/benchmarks): Reproducible commands, repository test boundaries, and the small labeled retrieval fixture.
- [Evidence methodology](https://semantix.ensureok.ai/evidence/methodology): E0–E3 labels explaining what each result does and does not establish.

## Blog

- [Semantix Blog](https://semantix.ensureok.ai/blog/): Research and practical guides about persistent memory, semantic retrieval, and coding-agent infrastructure.
- [Persistent Memory Open-Source Evaluation Guide](https://semantix.ensureok.ai/blog/persistent-memory-open-source-evaluation-guide): Criteria for evaluating persistent memory layers for coding agents.
- [Open-Source Semantic Memory Shortlist](https://semantix.ensureok.ai/blog/open-source-semantic-memory-shortlist): A practical shortlist of open-source semantic memory options.
- [Cross-Session Reuse Guide](https://semantix.ensureok.ai/blog/cross-session-reuse-guide): How reusable context moves between coding-agent sessions.
- [Semantic Memory Comparison Guide](https://semantix.ensureok.ai/blog/open-source-semantic-memory-comparison-guide): A comparison framework for open-source semantic memory architectures.

## Site

- [Terms of Service](https://semantix.ensureok.ai/terms): Terms governing use of the Semantix website and project.
- [Privacy Policy](https://semantix.ensureok.ai/privacy): How the site handles visitor data.

## Operator（运营主体）

- Legal operator: 确石人工智能科技（上海）有限公司（Queshi Intelligence）— https://www.ensureok.ai/
- Public contact: junhaihuang@aiqueshi.com（via https://semantix.ensureok.ai/contact）

## License（许可证）

- Current license: **MIT**.
- Full license text: https://github.com/Gnosil/semantix/blob/main/LICENSE

## Last updated

- 2026-08-22