docs · reference

bmem CLI

bmem is a thin, consume-only client for the open skills catalog. It searches and registers skills — it never drives a browser and never sees your cookies. Your agent runs the skills with whatever browser it already has.

install

npm i -g bmem-cli

Package	Binary	Requires	License
`bmem-cli`	`bmem`	Node, plus `npx` for `add` / `install`	MIT

Commands

Command	What it does
`bmem search <query>`	Search the catalog by keywords
`bmem search --site <domain>`	Filter results to one site
`bmem list [--limit 50]`	List the whole catalog
`bmem show <name>`	Print a skill’s SKILL.md without installing it
`bmem add <name>`	Download a skill and register it natively in your agent
`bmem install`	Register the bundled meta-skill (one-time bootstrap)

Find a skill

search the catalog

bmem search linkedin people
bmem search --site linkedin.com
bmem list --limit 50

Results are tiny index entries (under ~300 tokens each), built to be read by an agent before committing to anything:

index entry (result shape)

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

Want the full recipe before committing? Read it without installing:

inspect

bmem show linkedin.com/search-people

Add a skill

add

bmem add linkedin.com/search-people

After this, the skill appears in your agent’s own skill list and is invoked natively — you don’t call bmem again to run it. Under the hood:

what bmem add doestwo hops

bmem add linkedin.com/search-people

1 · file manifest

GET /v1/skills/linkedin.com/search-people/files{ skillId, files: [{ path, url }] }

2 · download blobs

Local cache~/.config/bmem/skills/linkedin.com/search-people/temp dir + atomic rename

npx skills add <path>

Registered nativelythe skills installer detects your agent (Claude Code, Cursor, …) and writes to the right place

important

Never invent a skill id. Only add a name that came back verbatim from search or list — the catalog is curated and read-only, so a made-up id like linkedin.com/find-jobs just 404s.

The meta-skill (`bmem install`)

bmem install registers a bundled meta-skill — the piece that teaches your agent the loop: search before exploring, how to read a SKILL.md, and how to map each capability (navigate, evaluate, …) to its own browser tools. Run it once per agent; details of that mapping are on the skills page.

Configuration

Env var	Default	Purpose
`BMEM_SKILLS_API_BASE_URL`	`https://browser-memory.com`	Skills API base — point it at a self-hosted or staging catalog
`BMEM_HOME`	`~/.config/bmem`	Local cache of added skills

No config command needed — set the variable and every command picks it up. The API contract the base URL has to answer is on the catalog API page.