# Distillio

> The open source training data collective — a community platform where members post, discover, and mirror LLM training datasets.

Distillio is a community-run library of LLM training datasets (RLHF, instruction-tuning, math, code, conversation, and more), plus an index of fine-tuning models and free Colab notebooks. Datasets are community-supplied; licenses are user-declared and not independently verified — check before use. The platform is built around public content: browsing and downloading datasets requires no account; posting, mirroring, and starring do.

## Core pages

- [Home](https://distillio.co/): overview, stats, and recent community activity
- [Browse datasets](https://distillio.co/datasets): search and filter the dataset library
- [Model index](https://distillio.co/models): curated open-weight models for fine-tuning and local GGUF quants
- [Fine-tuning notebooks](https://distillio.co/notebooks): free Unsloth Colab notebooks, indexed and linked
- [Submit a dataset](https://distillio.co/datasets/new): post a dataset or host a file
- [Privacy Policy](https://distillio.co/privacy)
- [Terms of Service](https://distillio.co/terms)

## API

The full machine-readable contract lives at [openapi.yaml](https://distillio.co/openapi.yaml) (discoverable by agents via the [API Catalog](https://distillio.co/.well-known/api-catalog)).

Key endpoints (all under `https://distillio.co/api`):

- `GET /datasets?q=&tag=&sort=newest|popular|downloads|stars&limit=&offset=` — list/search datasets (public)
- `GET /datasets/:id` — dataset detail (public)
- `GET /datasets/:id/benchmarks` — community benchmark results (public)
- `GET /models/list` — curated fine-tunable models (public)
- `GET /models/gguf` — local GGUF model list (public)
- `GET /notebooks?q=&baseModel=&taskType=` — indexed fine-tuning notebooks (public)
- `GET /stats` — platform-wide statistics (public)
- `GET /activity?limit=` — recent public activity feed (public)
- `POST /mcp` — MCP server (Streamable HTTP, JSON-RPC) exposing the same public data as tools (search_datasets, get_dataset, get_platform_stats, get_activity)

Authenticated actions (star, mirror, post, report, benchmark submission) require a Clerk session; they are documented in the OpenAPI spec.

## Agent discovery

- [API Catalog](https://distillio.co/.well-known/api-catalog) — RFC 9727 linkset
- [ARD capability manifest](https://distillio.co/.well-known/ai-catalog.json) — agentic resource discovery
- [Agent skills index](https://distillio.co/.well-known/agent-skills/index.json) — includes a dataset-search skill (SKILL.md) with a verified sha256 digest
- [MCP server card](https://distillio.co/.well-known/mcp/server-card.json) — MCP endpoint `https://distillio.co/api/mcp`
- [OAuth authorization server metadata](https://distillio.co/.well-known/oauth-authorization-server) — Clerk issuer, endpoints, and agent_auth registration info
- [OAuth protected resource metadata](https://distillio.co/.well-known/oauth-protected-resource)
- [auth.md](https://distillio.co/auth.md) — how agents register and authenticate
- [security.txt](https://distillio.co/.well-known/security.txt) — security contact
- This file (`llms.txt`) and [llms-full.txt](https://distillio.co/llms-full.txt)

## Content notes for agents

- Mirrored resources on this site are not owned by Distillio; ownership and licensing remain with their original owners.
- Dataset license labels are user-supplied and should be treated as unverified.
- Hosted dataset files stream through `GET /datasets/:id/download`; external datasets link to their origin (often Hugging Face).