Wemory Documentation

Cloud

Get an org slug and an invite token from your admin, then run one command. The invite token is short-lived - install before it expires.

curl -fsSL https://api.wemory.xyz/client-bundle | tar xz && \
  bash install_wemory.sh --org <your-org> --url https://api.wemory.xyz --invite <token>

Self-Hosted

First, your admin stands up a PostgreSQL + pgvector database on your infrastructure and shares the connection details with the Wemory team, who register your org and issue invite tokens (see Editions). Once your org is live, every team member installs the CLI with the same command - the API URL stays https://api.wemory.xyz:

curl -fsSL https://api.wemory.xyz/client-bundle | tar xz && \
  bash install_wemory.sh --org <your-org> --url https://api.wemory.xyz --invite <token>

Add ~/bin to your PATH (once)

echo 'export PATH="$HOME/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc

What the installer does:

1. Downloads and unpacks the client bundle
2. Creates ~/.wemory/config.json with your org and API URL
3. Provisions an API key from your invite token (stored in ~/.wemory/api_key)
4. Installs CLI scripts to ~/bin/
5. Injects Wemory instructions into ~/.claude/CLAUDE.md

Verify the install

List your org's projects to confirm auth and connectivity. Run wemory-doctor to check database health (connection, schema, pgvector).

wq projects     # confirms auth + connection
wemory-doctor   # checks database health

MCP endpoint

The server speaks Streamable HTTP. Add it as a remote MCP server in your host:

https://api.wemory.xyz/mcp

For hosts configured from a file (Codex, Hermes, Cursor), authenticate with your member key - the same one in ~/.wemory/api_key:

Authorization: Bearer <your-wemory-key>
# or, equivalently:
X-API-Key: <your-wemory-key>

Hosts that do not let you set headers (Claude AI, Cowork) use OAuth instead: add the endpoint as a connector, and the consent page asks for your Wemory key once, then keeps the connection signed in for you.

What the agent gets

Five tools, mapping onto the same API the CLI uses:

Curation stays out of MCP on purpose: promoting, verifying and granting are admin gestures, and an admin already has the CLI.

1. Start a session

Launch Claude Code with wemory. On your first message the agent automatically queries team memory for relevant context.

wq smart "what decisions were made about the auth migration"

2. Work normally

Code, debug, review, plan. As you work, the agent summarizes the session in the background and writes it to this session's slot - no extra steps needed. How often it does this is controlled by the auto_save mode (see Configuration).

3. Save & push

Say ws any time (or just end the session) and the agent writes the latest cumulative summary to this session's own slot. Depending on your manual_save_push setting it either pushes automatically or asks first.

# In your agent session, just say:
ws

# The agent writes a structured summary to this session's slot:
#   ~/.wemory/sessions/<session_id>.json
# Then pushes it (automatically, or after asking) with:
wp

Each agent session keeps its own slot, so two terminals never overwrite each other. Running wp with no argument pushes every pending slot.

4. Team compounds

Next time anyone on your team starts a session, the agent pulls context from all past sessions - including yours and your teammates'.

Session Notes

Structured summaries of AI agent sessions. Contain: summary, decisions, next_steps, open_questions, files_modified, tags. Searchable by the whole team.

Knowledge Items

Curated, verified facts about your team and projects - the source of truth and the highest-priority results in search. Create them directly with wk, or promote a decision with wq promote.

Decisions & Dreaming

Decisions are auto-extracted from session summaries - plentiful but unverified, so they rank below knowledge items. A daily curation pass ("Dreaming") proposes promotions and flags conflicts; when it needs a human call it queues a question for an admin (wq questions / wq answer).

Background capture

Your agent summarizes sessions as it works and stores each one in its own slot (~/.wemory/sessions/<session_id>.json). The auto_save mode sets the cadence and manual_save_push decides whether saved summaries push automatically - both in Configuration.

Watched folders

Run ww install to create ~/Wemory/ and a background watcher: drop a transcript in Transcripts/ or a doc in Documents/ and it's ingested automatically. Uploaded files (PDF, DOCX, PPTX, MD…) have their text extracted so they're fully searchable.

Projects

Organizational units for grouping sessions, knowledge, and files. Every session note is tagged with a project slug. Use wq projects to see available projects in your org.

Project Visibility

Projects are open by default - visible to everyone in your org. An admin can switch a project to restricted with wq visibility: it then disappears from search, listings, and writes for everyone except admins and members explicitly granted via wq grant. Audit who sees what with wq access.

# Check database health
wemory-doctor

# Confirm auth + connectivity
wq projects

# View recent errors on this machine
wq errors --local

# Force re-authentication / regenerate API key
wl --force

# Re-authenticate with a fresh invite token
wl --invite <token>

# Re-run installer (preserves config)
curl -fsSL https://api.wemory.xyz/client-bundle | tar xz && \
  bash install_wemory.sh --org <your-org> --url https://api.wemory.xyz --invite <token>

How it works

The installer injects instructions into ~/.claude/CLAUDE.md that tell your agent when and how to interact with Wemory. No manual configuration needed - the agent follows the injected instructions automatically.

CLI Quick Reference

# Search team memory
wq smart "query here"

# Push session summary
wp

# Create knowledge item
wk <project> "Title" "Content"

# Upload file
wu /path/to/file <project> "description" --tags tag1,tag2

# List available projects
wq projects