Getting Started
Five steps take you from a new account to a working knowledge base your tools can use.
From zero to working
Set up access, bring in what matters, and give your tools a way through the same front door.
1. Create a token
In the web app, open Access Tokens, create a token — full access if your tools will write, read only if you only want to serve context — and copy it when it appears. It will not be shown again.
2. Install and connect the CLI
Install the recuerd0 CLI with Homebrew, then add your account using the token you just created.
brew install maquina-app/tap/recuerd0
recuerd0 account add personal --api-url https://your-server.com --token YOUR_TOKEN3. Install the skill
recuerd0 skills install copies the bundled skill to ./.claude/skills by default. Use --global for ~/.claude/skills, --target <directory> for any directory, and --force to overwrite an existing copy. Claude Code loads it automatically — restart an open session to pick it up. Other agents need the file added to their instructions.
recuerd0 skills install4. Import what you already know
Import is propose → review → commit. Run recuerd0 workspace list to find the workspace ID.
Propose, read the plan, then commit.
recuerd0 workspace list
recuerd0 import propose ./vault --workspace 12 --prettyPropose writes import.plan.yaml in your current directory. Open it — the manifest lists one entry per file with the title, category and tags it will use.
recuerd0 import commit import.plan.yaml --prettyAsk your agent: "I just imported notes into recuerd0 workspace <id>. Do the after-import pass." It will cluster the memories, fix weak titles, and propose hubs for review.
5. Connect over MCP
Claude Desktop, Claude.ai, and other MCP clients can use the remote MCP server with browser-based authorization, so there is no token to copy. MCP can list, search, read, and write memories; have your agent read Map — how this workspace is kept first so it picks up the workspace's shape before it writes. MCP clients do not include the CLI installer, so download the MCP skill and add it to your client. Add the connector under Settings → Connectors, and upload the skill file to your project's settings.
Three doors, one knowledge base
Choose the surface that fits the work. Every door reaches the same workspaces, memories, versions, and search.
Web
Write, edit, pin, organize, and review memories in the browser.
CLI + API
Use the CLI for terminal workflows and the API for scripts and automation.
MCP
Use MCP when an AI client should search, read, and write memories during the conversation.
What you start with
Every new account includes My Workspace with four memories in a deliberate reading order:
- Map — how this workspace is kept
- Continuation Brief
- Index — decisions
- D001 — the first decision
Map — how this workspace is kept
The front door and getting-started checklist. Read it first, then keep every memory reachable within two hops.
Continuation Brief
The second read: current state, open threads, and the first thing to pick up next session.
Index — decisions
One line per locked decision, each pointing to the memory that records what you chose and why.
D001 — the first decision
A real decision in the full format — what was chosen, why, what was rejected — and the reason this workspace starts flat.
When it grows
A hub is a routing-table memory for crowded territory: one line of judgment per entry, each pointing at the memory with the detail. It earns its place around twenty memories, when clusters start crowding the map — create the hub, move the cluster's map lines into it, and leave a single line on the map pointing there. The seeded D001 is the worked example of making, and recording, the call not to have one yet:
Today the map is flat:
- Payments retry logic — #12
- Payments webhook quirks — #15
- Payments provider limits — #18
Past ~20 memories, those lines collapse into one:
- Payments — routing in Hub — Payments (#22)
And the hub is just a routing table — one line of judgment per entry:
- Why retries drop large jobs — #12
- Which webhook events lie — #15
- Provider limits that bit us — #18None of this is enforced — rename it, trim it, or delete any of it; a blank workspace is a fine workspace.
Read the full reasoning in the workspace blueprint.
