docs · reference

Catalog API

The catalog is a public, read-only, curated API — the thing the bmem CLI and the ecosystem pages are built on. No auth, no keys, no publish endpoint.

Base URL: https://api.browser-memory.com. The CLI can be pointed elsewhere with BMEM_SKILLS_API_BASE_URL.

Endpoints

Method	Path	Purpose
`GET`	`/v1/skills?q=&site=`	Search / list. Omit `q` to list everything. Returns `{ skills: [IndexEntry] }`
`GET`	`/v1/skills/{domain}/{task}/files`	File manifest for a skill: `{ skillId, files: [{ path, url }] }`
`GET`	storage url	The actual SKILL.md bytes, served from a public storage bucket

Index entries

Search results are deliberately small — under ~300 tokens each — so an agent can scan many before committing to one:

IndexEntry

{
  "name": "linkedin.com/search-people",
  "hostname": "linkedin.com",
  "task": "search-people",
  "slug": "linkedin-search-people",
  "side_effect": "read",
  "needs": ["navigate", "evaluate"],
  "intent": "Search people on LinkedIn, one or many queries…",
  "keywords": ["linkedin", "search", "people"],
  "verified": true
}

How a skill is fetched

bmem add never guesses file URLs — it asks for the manifest, then downloads the blobs it lists. Two hops, so multi-file skills work and storage stays decoupled from the API:

skill fetchmanifest → blobs

bmem addor any client

Catalog API/v1/skills/{domain}/{task}/filesreturns the file manifest

Public storageSKILL.md bytes at each files[].url

A direct, human-readable rendering of each skill also exists at /skills/{domain}/{task}.md on the website — that is the web and llms.txt surface, not what the CLI uses.

Zero-install discovery: llms.txt

The whole catalog is also published as a static, cacheable text file. Any agent — or any human — can read it with zero auth and zero install:

fetch the catalog index

curl https://browser-memory.com/llms.txt

llms.txt — one line per skill: name, link, intent, side effect.
llms-full.txt — the same list with every SKILL.md inlined: one fetch puts the entire catalog in context.

Publishing & curation

There is no public publish API — the catalog is curated. Skills land through the maintained catalog pipeline (validate → storage + index), which keeps every entry tested and keeps ids stable. If you want a site or skill added:

Request it from the ecosystem page — site requests and skill uploads both go to the team.
Or open an issue on GitHub.

Self-hosting

Self-hosting is free and unlimited. Run your own catalog that answers the endpoints above, then point the CLI at it — no config command needed, every command picks the variable up:

point bmem at your catalog

BMEM_SKILLS_API_BASE_URL=https://your-catalog.example.com bmem search linkedin