Metadata-Version: 2.4
Name: tomefile
Version: 0.1.0
Summary: Local RAG as a single file
Author-email: Simeon Emanuilov <simeon.emanuilov@gmail.com>
License-Expression: MIT
Project-URL: Homepage, https://tomefile.com
Project-URL: Repository, https://github.com/s-emanuilov/tomefile
Keywords: rag,embeddings,sqlite,local
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Operating System :: OS Independent
Classifier: Environment :: Console
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: typer>=0.12
Requires-Dist: fastapi>=0.115
Requires-Dist: uvicorn>=0.32
Requires-Dist: httpx>=0.27
Requires-Dist: sqlite-vec>=0.1.6
Requires-Dist: cryptography>=43
Provides-Extra: dev
Requires-Dist: pytest>=8; extra == "dev"
Dynamic: license-file

# Tomefile

A local RAG index you pack **into one file**.

Fine-tuning a small model on your knowledgebase will not make it remember them. Retrieval will. Most RAG still calls a remote store, an embeddings API, and a set of tools. A `.tome` is the other shape: chunks, vectors, and the embedding model in one archive you open on `127.0.0.1`.

Build the index once. Hand someone the file.

- **`.tome`** — documents already chunked and embedded, plus the embedder (GGUF inside, or a SHA-256 reference). Checksums on every member. Optional Ed25519 signature.
- **`.tbox`** — the same file, with a chat model packed in or pinned by hash.
- **`tomefile serve`** — OpenAI-compatible `/v1/chat/completions` on the loopback interface. Answers cite source files and character offsets.

Chunk text is plaintext SQLite; this is not a way to hide a corpus. Vector search uses the embedder named in the file. A different model with the same vector size will retrieve the wrong neighbours.

Site: [tomefile.com](https://tomefile.com).
