second-brain-cloudflare

Private memory for you. Shared memory for your team. Available to every MCP-compatible AI tool you use.

Now with Team Edition — private personal layers plus a shared team layer, in one Worker.

Claude, ChatGPT, Cursor, Codex, and the other AI tools you use do not naturally share context. You end up repeating the same projects, decisions, and preferences in every app.

Second Brain gives those tools one persistent memory system. It runs in your own Cloudflare account, stays under your control, and retrieves the right context by meaning rather than exact wording.

The desktop app is the easiest way to start. It builds your Second Brain and connects your AI tools in about two minutes—no terminal or Cloudflare setup required.

Download for Mac or Windows

Deploy to Cloudflare · Read the documentation

What it does

• Recalls by meaning. Ask a natural-language question and find the right memory even when you used different words when saving it.
• Works across tools and devices. Every client talks to the same Worker, so there is nothing to copy or synchronize between apps.
• Keeps you in control. Browse, edit, append, connect, share, export, or permanently remove any memory from the dashboard.
• Builds useful context. Automatic classification, duplicate detection, relationships, time-aware ranking, and optional weekly insights help the brain stay useful as it grows.
• Captures from where you already work. Use MCP clients, the CLI, browser extension, Obsidian, Notion, calendars, email, iOS Shortcuts, or the web dashboard.
• Acts on what matters next. Add dates to memories, review overdue and upcoming commitments, and let the installed PWA proactively push a reminder when something becomes due. See the Reminders and Push guide.
• Stays in your account. Memories, vectors, credentials, and application resources live in your own Cloudflare account.

See it in action

Team Edition

Second Brain can now be a team's memory without stopping being yours.

• Every person gets a Personal workspace that nobody else can read, plus a Shared layer visible to the team.
• Memories are private by default and only enter the Shared layer when someone deliberately shares them.
• Sharing moves one canonical memory rather than making a copy. Its author remains visible, and only the author or an admin can edit, delete, or un-share it.
• Admins can manage members, access, capture defaults, and integrations without gaining access to anyone's personal workspace.
• Existing v2 memories become the owner's private memories during upgrade. Nothing is exposed to a team automatically.

The same Worker supports personal and team use; there is no separate team deployment. In the API, CLI, and MCP tools, the Shared layer is represented by the stable workspace value company. See the Team Setup guide for member management, capture policies, sharing, and upgrades.

v3.0.0 scope: each brain has one shared team. The API and MCP layer include optional team parameters and a list_teams tool so multi-team support can ship later without breaking changes; the dashboard and admin flows do not create or switch between multiple teams yet. See CHANGELOG.md.

How it works

Second Brain runs as a Cloudflare Worker backed by D1, Vectorize, Workers AI, and KV. Every app and AI client connects to that Worker through REST or the Model Context Protocol (MCP).

1. Capture: Save a decision, preference, project update, note, or source from any connected client.
2. Organize: Second Brain classifies it, checks for duplicates and contradictions, creates relationships, and indexes it for semantic search.
3. Recall: Ask in natural language. Second Brain retrieves relevant memories, follows useful connections, and returns source-backed context to the tool you are using.

If Vectorize is unavailable, captures and keyword recall continue working. Your memories remain usable while semantic indexing is restored. The shipped embedding models read English best; the desktop app's Settings can switch a brain to a multilingual reading.

Search now finds the hard things: exact names, ticket numbers, versions, and phrases in any language, even when they sit in old memories, and finding them is dramatically faster and cheaper, staying that way as the brain grows, which keeps the free plan comfortable. It does this with a full-text index that ranks matches by relevance instead of scanning every memory. The upgrade is automatic: new installs use the index immediately, existing brains build it over nightly runs, and no client needs updating.

Memory tools

A memory keeps the dates it was true for: recall with as_of answers what was actually true on a past date, and a fact that a newer one replaces is kept as history instead of being marked wrong.

On a team brain, memory tools accept a workspace of personal or company when you want to choose a layer explicitly. company is the wire value for the Shared team layer. Without workspace, captures use the member and team defaults, while recall searches everything that person is allowed to see.

Optional team (workspace id) and MCP list_teams / GET /team/workspaces are wired for a future multi-team release. In v3.0.0 you can omit them — each brain has one shared team and the primary team is used automatically.

Projects

Memories live on four axes: workspace = who can see it (personal / company / team) — tenancy, unchanged. project = what it's about — a named, managed container. tags = free-form facets, unchanged. source = where it came from, unchanged. Call list_projects to discover projects in scope and pass project on remember to group related memories. A memory can belong to one or more projects; use projects to organize by topic, initiative, or context rather than bare topic tags.

CLI example:

brain remember --workspace company "We ship on Thursdays"
brain recall --workspace company "when do we ship?"
brain recall --project website "what did we decide about hosting?"

Prompt Capsules

Prompt Capsules are deterministic, read-only projections for gateways and custom agents that can place stable context before a changing user request. They complement query-specific recall; they do not inject every memory into every prompt.

A Capsule entry is an ordinary canonical memory with one target tag and one slot tag. Core entries use capsule:core; project entries use capsule:project:<project-slug>. Slots are emitted in this fixed order:

• Core: identity, preferences, constraints, principles
• Project: current-state, decisions, open-questions

Tag the slot as capsule-slot:<slot> and keep at most one canonical entry per slot. Draft and deprecated entries are ignored. Ambiguous slots are omitted without choosing a winner; malformed rows are skipped. The response reports duplicate_slots and invalid_entries, and complete is false. Other valid slots remain available, including on the shared layer.

An entry must carry `stat