CLI Workflows

End-to-end walkthroughs for running Knowlery entirely from the command line — no Obsidian required. Every folder these workflows produce opens in Obsidian later with zero migration.

Install once:

npm install -g knowlery
knowlery --version

From zero to a working knowledge base

# 1. Initialize a workspace
knowlery init --dir ~/kb/research --platform claude-code --name "Research KB"

# 2. Register it so every later command (and MCP) can address it by name
knowlery kb add research ~/kb/research

# 3. Verify
knowlery health --kb research

init scaffolds the four compiled directories, KNOWLEDGE.md, SCHEMA.md, INDEX.base, the built-in skills, and the agent platform config. From here, drop raw notes anywhere in the folder and let an agent /cook them.

The daily loop, headless

# The lay of the land — a live orientation map, computed on read
knowlery index --kb research

# What's waiting?
knowlery stale --kb research

# Ask a question (from any directory)
knowlery query --kb research "what did we decide about the rollout?"

# Search everything you have
knowlery query --kb '*' "where did I write about backpressure?"

stale is a work list: compiled pages whose cited sources changed, plus notes never compiled. Cook selectively — many notes are legitimately never compiled.

Abstention (No confident matches) is a result, exit 0. Scripts can rely on --json for stable shapes on every command that reports.

After an upgrade

npm i -g knowlery@latest
knowlery sync --kb research        # refreshes skills + instruction files

sync is idempotent and write-on-change; it refuses to run if the workspace was last synced by a newer version (the downgrade guard), so a stale global install can never damage an upgraded workspace.

Share a slice of your knowledge

# 1. See what a topic's graph-closure would ship
knowlery bundle export retrieval-engine --kb research
# -> exits 1 with the review checklist: nothing ships unreviewed

# 2. Record your review decisions (there is deliberately no approve-all)
knowlery bundle review retrieval-engine --kb research --list --json
knowlery bundle review retrieval-engine --kb research \
  --approve concepts/retrieval-engine concepts/scoring --flag Projects/meeting-notes

# 3. Export, then publish to a GitHub Release
knowlery bundle export retrieval-engine --kb research --zip
knowlery bundle publish retrieval-engine --kb research --repo you/knowledge-shelf

publish creates the Release with a SHA-256 checksum in its notes and prints an audience statement — exactly who can download this and how to grant access. Publishing to a public repo requires re-acknowledging the risk items (--acknowledge-risks) — a public release is permanent.

Install and stay subscribed

# Install from a URL (private repos work through your gh login)
knowlery bundle install https://github.com/you/knowledge-shelf/releases/download/v1.0.0/bundle.zip \
  --kb research --verify <sha256>

# Later: is anything newer?
knowlery bundle check-updates --kb research     # read-only, never auto-updates
knowlery bundle update some.bundle --kb research

Updates go through the full gate pipeline: version must increase, conformance is checked, and local modifications to installed bundle files make the update refuse (with the file list) unless you pass --force.

Serve your KBs to agents

# Local agents (Claude Desktop/Code, Codex, Cursor, Antigravity): stdio
claude mcp add knowlery -- knowlery mcp

# Another machine: HTTP behind a tunnel
openssl rand -hex 32 > ~/.knowlery-mcp-token
knowlery mcp serve --port 8787 --token-file ~/.knowlery-mcp-token
cloudflared tunnel --url http://127.0.0.1:8787

See Agents & MCP for the full tool reference, write-access flags, and client configs.

Scripting notes

• Every command takes --dir <path> (default: cwd) or --kb <name> — never both. The registry is never required; --dir workflows work forever.
• --json is available on every reporting command; shapes are frozen as part of the 1.0 contract.
• Exit codes: 0 success (including abstention and empty reports), 1 operational failure (unhealthy workspace, failed gate), 2 usage error (bad flags, missing arguments).
• Output is pipe-safe: knowlery query ... | head -1 works; an early-closed pipe is a normal end of conversation, not an error.