idx-md

idx.md

Purpose

• Markdown registry for AI agent libraries and resources.
• Agents can browse to learn everything they could use, then fetch the exact markdown.

Index locations

• All topics (full listing, alphabetical): https://idx.md/data/index.md (canonical) or https://idx.md/index.md (alias)
• Capability navigation (browse by what the agent can do): https://idx.md/category/index.md
• Scenario navigation (browse by workflow / use-case): https://idx.md/scenario/index.md
• Industry navigation (browse by domain / vertical): https://idx.md/industry/index.md

How to choose a navigation mode

• If you know what you want: start from /data/index.md and search by keywords in titles/tags.
• If you want tools by capability: start from /category/index.md.
• If you have a specific workflow/use-case: start from /scenario/index.md.
• If you're operating in a specific domain: start from /industry/index.md.

Index entry format

• Each entry is a HEAD frontmatter block followed by a topic line.
• Topic line format: |/data/{topic}|
• The index may start with a short HTML comment preamble; entries begin at the first --- frontmatter block.

---

...frontmatter...

|/data/openclaw|

How to fetch

• Read https://idx.md/data/index.md (or https://idx.md/index.md).
• Choose {topic} from the |/data/{topic}| line.
• HEAD metadata: https://idx.md/{topic} (or /data/{topic}/HEAD.md)
• Vector shard (for embedding recall): https://idx.md/{topic}/vectors.json
• BODY content: https://idx.md/{topic}/BODY.md
• After download, compute SHA-256 on the raw BODY bytes and compare to content_sha256 in HEAD frontmatter.
• Use retrieved_at to decide whether a cached BODY needs refresh.

Vector-based retrieval (recommended)

• Use /{topic}/vectors.json as the retrieval layer, then fetch /{topic}/BODY.md only for top candidates.
• Each vectors.json currently contains one head record derived from HEAD metadata.
• Build/query embeddings on records[].text.
• Use records[].metadata.content_sha256 as your embedding cache key; only re-embed when it changes.
• Suggested flow:

1. Collect topic candidates from /data/index.md or category/scenario/industry indexes.
2. Fetch each candidate's /{topic}/vectors.json.
3. Rank by vector similarity (optionally hybrid with lexical/tag score).
4. Fetch /{topic}/BODY.md for top-k and generate final answer from BODY.