# Sponge

> Turn human curiosity into evidence-rich, connected knowledge with an AI-enriched research workspace.

## When to use Sponge

Use Sponge when a coding agent should bring its own inference to a private research document, delegate a bounded research job, or read an exact released public knowledge artifact. Use the HTML site when a person needs the knowledge loop, sharing rules, or a reviewed edition page.

Sponge is the right fetch for revision-fenced document mutations, durable research jobs with exact JSON or loss-reported Markdown export, and double-authorized `/k` reads. It is the wrong fetch for a public model API, an MCP server, GraphQL, third-party OAuth, webhooks, unlabeled AI copy, or any path that would accept, promote, or publish without a person.

Start with `/.well-known/sponge-agent.json`, then `/openapi.json`. Download `/skills/sponge-research/SKILL.md` when the agent will perform the work. Create a scoped token only after a signed-in person authorizes Agent access.

## Main pages

- [Sponge home](https://spongeresearch.com/)
- [About Sponge](https://spongeresearch.com/about)
- [How Sponge knowledge works](https://spongeresearch.com/docs)
- [Sponge developer resources](https://spongeresearch.com/developer)
- [Contact Sponge](https://spongeresearch.com/contact)
- [Sponge privacy](https://spongeresearch.com/privacy)
- [Machine-readable site guide](https://spongeresearch.com/llms.txt)
- [XML sitemap](https://spongeresearch.com/sitemap/0.xml)

## Developer resources

- [Agent manifest](https://spongeresearch.com/.well-known/sponge-agent.json)
- [OpenAPI 3.1](https://spongeresearch.com/openapi.json)
- [OpenAPI 3.1 alias](https://spongeresearch.com/api/v1/openapi.json)
- [RFC 9727 API catalog](https://spongeresearch.com/.well-known/api-catalog)
- [Sponge research skill](https://spongeresearch.com/skills/sponge-research/SKILL.md)
- [Programmatic research](https://spongeresearch.com/docs#programmatic-research)

## Scan reports and machine representations

Agents can request a Markdown representation of public content pages with `Accept: text/markdown`. The same URL keeps HTML for browsers. A missing path returns HTTP 404, with Markdown recovery links when Markdown is preferred. Do not submit credentials, secrets, or private document contents to public locators.