Open-source backend formulti-user AI agents
Give every user an isolated agent workspace with OAuth, connected sources, shared memory, watchers, and secrets agents never see.
Devices
Files, Screen Time, browser history — sources that only live on your machine.
Cookies and tokens stay on-device. Lobu's servers never see them.
Chat events, watcher triggers, and tool calls in your menu bar.
Pin sensitive workloads to your device. Everything else runs serverless.
Connections
50+ built-in connectors. OAuth, API key, or browser session.
Point Lobu at any MCP endpoint. Tools wire into memory automatically.
Lobu runs agent-authored TypeScript connectors serverlessly — no hosting.
API key, OAuth, browser session, or none. Credentials stay where you choose.
Source-backed context, kept fresh.
Connect a source once. Lobu types it into memory agents can search and cite. Watchers refresh it on schedule.
Read the memory guideOne agent, every surface.
Bundle tools, packages, and network access into a skill. The same agent runs in Slack, Telegram, REST, MCP, and ChatGPT — no per-platform plumbing.
Explore skillsGive agents issue intake, owner routing, and safe GitHub actions in one reusable bundle.
One gateway, isolated workers, no leaked secrets.
Lobu receives messages at the gateway, starts sandboxed worker processes on demand, and proxies tools, memory, and credentials so agents stay isolated from your real keys.
- Link users across platforms with single sign-on
- Approval flows, rich cards, buttons, and more
- Workers never see secrets
- HTTP proxy with domain allowlist
- MCP proxy with per-user OAuth
- BYO provider keys (Anthropic etc.)
- One sandbox per user and channel
- Subprocess isolation with just-bash virtual filesystems
- systemd-run hardening on Linux production hosts
- No direct internet access (gateway proxy only)
- Nix reproducible environments
- OpenTelemetry for observability
Build your first
multi-user agent.
Latest blog posts
Filesystem vs Database for Agent Memory
Agents need a workspace to think in and a warehouse to remember in. The filesystem is for ephemeral work. The memory layer is for durable organizational knowledge.
MCP Is Overengineered, Skills Are Too Primitive
MCP HTTP is great for external services. MCP stdio is redundant. And most skill systems are just prompt text with no reproducibility. Here's what we built instead.
Introducing Lobu
From a Slack bot to multi-tenant OpenClaw infrastructure — the story behind Lobu.