Lobu

What Lobu does

Lobu provides a shared context layer for AI agents. It turns tool calls, data changes, device signals, API events, and conversations into an append-only event history associated with identities and entities such as people, projects, customers, incidents, homes, and companies. Agents can use that history as durable organizational memory rather than relying on a single conversation transcript.

The same context can serve a person’s connected accounts or an entire company’s systems. Authorized agents can search memory, query structured entities, inspect connected sources, save information, and hand work to persistent Lobu specialists. Specialists retain their own identity, instructions, tools, conversations, and access policy, so a responsibility can continue beyond the session that created it.

How it works

Connectors, webhooks, devices, APIs, and agent-written integrations feed Lobu’s append-only event log. Lobu resolves identities, attaches provenance, and organizes events into a context graph containing entities, relationships, current state, and history. Agents query or watch that graph through MCP and can contribute new context or initiate governed actions through scoped tools.

Access is evaluated according to the caller’s grants and source permissions. The architecture keeps identity, policy, approvals, credential handling, provenance, and audit on the Lobu side while agents use the shared context through an MCP client. Any authorized client can work with the same history, subject to its permissions.

Setup and configuration

Use the Lobu CLI to connect an existing MCP client. For supported command-line clients, the basic commands are:

• npx @lobu/cli@latest connect claude-code
• npx @lobu/cli@latest connect codex
• npx @lobu/cli@latest connect opencode

The CLI can detect a client, install the supported MCP and skill bundle, or provide the native handoff when a host requires UI-based setup. OAuth is completed when prompted, and authentication occurs in the agent on first use. After connecting, choose the sources to share and ask the agent to use Lobu when shared context is needed.

The same MCP endpoint can be used with Claude Code, Codex, OpenCode, Antigravity, ChatGPT, Claude Desktop, Cursor, and custom MCP clients. Lobu also provides a Mac app and Chrome extension. No Lobu agent runtime or lobu.config.ts is required for adding shared context to an existing agent.

Tools and capabilities

• Search shared, durable organizational memory under the caller’s permissions.
• Save agent-generated memory and decisions for later use.
• Query structured entities and their linked history.
• Inspect information from connected discussions, projects, customers, and other sources.
• Discover and delegate tasks to persistent Lobu specialists.
• Share context across personal or company-level agents while applying access policy.

Limitations and notes

Lobu’s results and actions are permission-aware: the calling identity and its grants determine which context and connected sources are available. Persistent specialists also have their own access policy and tool set.

The stable CLI channel uses @latest; preview users can use the canary package tag and pin an exact published version when reproducing changes. macOS installations made before the ai.lobu.mac bundle identity may require a manual DMG upgrade, and the renamed app asks again for login-item and privacy permissions.