Add AI agents tutorial

[prrao87]

Summary

Adds a tutorial for building with AI agents and LanceDB. It shows how to install and update the LanceDB skill, explains the reviewable skill lockfile, and walks through a multimodal Camelot ingestion pipeline.

The tutorial asks readers to inspect the code an agent creates for strict Pydantic validation, batched writes, and the correct OSS maintenance step. It also demonstrates how to use branches for isolated embedding experiments and broadens that pattern to parser, retrieval, curation, and migration work.

Why this matters

Before this change, the documentation did not give coding-agent users a concrete LanceDB workflow or a way to evaluate agent-generated data changes safely. Readers could install a skill without a practical guide to validating the generated pipeline, selecting OSS or Enterprise, or keeping an experiment out of main.

The new guide makes those decisions explicit. It positions OSS for local and single-machine work where the application owns maintenance, and Enterprise for shared production infrastructure with distributed capacity and managed operations. It also frames a branch as an evaluation boundary: state a hypothesis, use a fixed test set, review the results, and promote only an approved change.

Validation

• Generated documentation snippets
• Ran the targeted Python ingestion test
• Checked Mintlify links
• Ran the staged-diff whitespace check

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
lancedb-bcbb4faf	🟢 Ready	View Preview	Jul 23, 2026, 7:39 PM

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
lancedb-bcbb4faf	🟡 Building	–	Jul 23, 2026, 7:38 PM

[dantasse]

(copying from slack)
it's a weird thing to write a doc about, right? like the thing we want users to really do is: type in "/lancedb do what I want"
... maybe we should? e.g. this is kind of another "lancedb end to end project", like I'm not sure how this particular example is about "using agents"
ooh ok, I have an idea: make this into two docs:

1. "build with agents" - which is basically the top section of this doc; the part before "Tutorial: Use the LanceDB skill"
2. "use branching and merging" - focuses on branching and merging

But, approving in case you disagree, I don't want to hold you up too much! like it wouldn't hurt anything to merge as is :)

[dantasse]

missing punctuation at the end of the line

[dantasse]

so I think on claude I would type:

/lancedb ingest the dataset in `data/` into a LanceDB OSS table

and on codex I would type:

$lancedb ingest the dataset in `data/` into a LanceDB OSS table

Should we recommend that kind of usage? It's more concise

[prrao87]

Split the tutorial into two docs based on the review feedback, grouped under a collapsible sidebar folder.

What changed

• build-with-ai-agents.mdx — now scoped to building with the skill: install + lockfile, the Camelot multimodal ingestion tutorial, and the OSS vs. Enterprise takeaways. Ends with a pointer to the new branching page.
• agent-branch-experiments.mdx (new) — the branch-experiment workflow: isolate experiments on branches, evaluate on a fixed test set, apply the winner to main, plus the "more experiments you can run" table.
• docs.json — both pages live in a collapsible folder that's expanded by default ("expanded": true).

Notes

• Split at the branching boundary so the ingestion tutorial stays whole rather than cutting before it.
• The branching page links to /tables/branching for merge/diff mechanics instead of re-documenting them, to avoid overlap with that reference.