> ## Documentation Index
> Fetch the complete documentation index at: https://familyco.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Knowledge

> Upload, index, and retrieve internal documents for agent context.

Knowledge gives your agents retrieval-augmented context from internal documents. You upload source files, run indexing, and query the resulting chunks before or during execution.

## What it includes

* **Document library** with project and status filters
* **Converter status** check for the `familyco-py` binary
* **Chunk indexing controls** (`maxChars`, `overlapChars`)
* **Chunk explorer** per document
* **Retrieval preview** by query, project, and document filters

## Document lifecycle

1. Upload a file to `/api/v1/knowledge/documents/upload`.
2. The document starts in `uploaded` state.
3. Run index on the document (`/api/v1/knowledge/documents/{id}/index`).
4. The system converts, chunks, embeds, and stores vector rows.
5. The document moves to `indexed` (or `failed` with `errorMessage`).
6. Retrieve candidates via `/api/v1/knowledge/retrieve`.

## Storage model

* Original files are stored under `<workspace>/knowledge/documents/<documentId>/original/...`.
* Combined converted markdown is stored under `<workspace>/knowledge/converted/<documentId>/combined.md`.
* With the `prisma` repository driver, vector-backed knowledge records are persisted in a dedicated SQLite database (`familyco-knowledge.db`).

## Permissions and audit

* Knowledge routes require authenticated API access.
* All Knowledge routes enforce minimum level `L0`.
* Upload, index, and delete operations are written to the audit log:
  * `knowledge.document.upload`
  * `knowledge.document.index`
  * `knowledge.document.delete`

## Operational notes

* Upload accepts multipart form data and a single file per request.
* Server-side multipart limits cap uploads at 25 MB.
* Indexing depends on the converter binary path (`knowledge.converter.binaryPath`) or fallback binary resolution.
* Retrieval normalizes defaults when omitted: `topK=6`, `minScore=0.12`.

## Converter source: familyco-py

FamilyCo uses the `familyco-py` converter binary for document conversion before chunking and embedding.

* **Repository:** [omaicode/familyco-py](https://github.com/omaicode/familyco-py)
* **Binary source:** download from the repository's [latest releases](https://github.com/omaicode/familyco-py/releases/latest)

`familyco-py` is a Python toolkit CLI for FamilyCo. Its current production module is `doc-convert`, powered by Docling, which supports document conversion and OCR. The repository is structured as a modular CLI so future modules (for example image processing and summarization) can be added without changing the main entrypoint.

For request and response schemas, see the [API reference](/api-reference/introduction).