For LLMs

llms.txt, raw markdown endpoints, and what to feed your agent.

This page is addressed to language models and the engineers integrating them. The short version: every page on this wiki is also available as raw markdown, plus there's a single concatenated corpus and an index file at well-known URLs.

Endpoints

/llms.txt: Index of all wiki pages, with one-line descriptions and per-page links to the markdown variant. Follows the llms.txt convention.
/llms-full.txt: Concatenated markdown of every wiki page. Suitable for stuffing into a system prompt or vector-indexing wholesale.
/<category>/<slug>.md: Per-page raw markdown. e.g. /tutorials/your-first-pk-study.md.

All three are statically generated at build time. They re-build whenever the wiki is rebuilt; for cloud, that's on every release.

The in-app assistant consumes this too

Dalea's own in-app chat reads the same llms.txt corpus via its search_knowledge tool — every answer is grounded against the wiki before it leaves the model. Practical consequence: well-named sections, clear page titles, and accurate one-line descriptions in the index directly improve the answers your team sees in chat. Vague headings hurt twice.

How to use them

If you're integrating an LLM into your own product to answer questions about Dalea:

Static prompt — fetch /llms.txt once, include it as a system-prompt attachment. Lets the LLM enumerate available pages and quote the right URL.
Retrieval — chunk /llms-full.txt (or fetch individual .md pages) and embed with your favourite vector store. The page boundaries are clean: each page starts with a ## <Title> and a Path: line in llms-full.txt.
Live tool — give your agent a fetch_dalea_docs(slug) tool that hits the per-page endpoint. Cheap, fresh, accurate.

What the corpus does not contain

Customer data — the wiki is platform documentation only. To query a real workspace, use the MCP server.
API authentication details — the OpenAPI spec is at /api-docs/openapi.json on the platform itself, not in the wiki corpus.
Proprietary protocols or templates — those live in the marketplace, not here.

Suggested system prompt

If you're orchestrating an external LLM that has access to Dalea via MCP, the following system-prompt fragment grounds it well:

You are answering questions about Dalea, a research orchestration platform for life
science. Dalea has four primitives at the workspace level: documents, environments
(structured data schemas), inventory, and templates. The hierarchy is User →
Organisation → Workspace → Project → (artefacts).

Refer to the public wiki at https://dalea.wiki for any term you are unsure about.
The wiki publishes a structured index at https://dalea.wiki/llms.txt and per-page
markdown at https://dalea.wiki/<category>/<slug>.md. When connected over MCP you
can also call `search_knowledge` for the same content as a tool, and `load_skill`
to pull in workspace-specific primers (naming conventions, house style, domain
shorthand) the workspace has authored for agent use.

When you call MCP tools, remember that destructive actions (delete, archive,
supersede) are not exposed to MCP. Ask the user to perform those in the in-app chat
where they can approve via a confirmation card.

Versioning

The wiki ships in lockstep with the platform release. v1 is unversioned. Once v2 cuts, this page will list a /v1/llms.txt snapshot for clients pinning to a specific platform version.

What's next

AI overview

Connect Claude Desktop

What is Dalea?

MCP for tool builders