Lobu
Autonomous agents in your organization. Multi-tenant OpenClaw
Last verified:
What is Lobu?
Lobu is an open-source backend for building proactive, multi-user AI agents that act automatically when company data changes. It connects to your company's data in real time through 50+ built-in connectors, any MCP server, or custom TypeScript connectors, then builds a typed, queryable knowledge graph automatically. Watchers monitor events and flag important changes, while agents respond on the model you choose as a Slack bot, API, or another agent.
The platform gives every user or channel an isolated agent workspace with connected sources, shared memory, and per-user OAuth isolation. Workers run in isolated containers (Kata Containers, Firecracker microVMs, or gVisor) with secret proxying so agents never see real API keys. The system supports local embedded deployment (single Node process), self-hosting via Docker/Helm on any cloud, or Lobu Cloud's managed runtime with per-user isolation and automatic upgrades.
Lobu is designed for revenue teams tracking account health, legal teams reviewing contracts, finance teams reconciling data, leadership turning memos into reusable context, venture firms tracking companies and founders, and agent communities discovering aligned members. It includes 12 working examples on GitHub for sales, legal, finance, leadership, market, and agent community use cases that ship as deployable projects.
Key features include 50+ built-in connectors, on-device connectors capturing context cloud agents can't see, MCP proxy with per-user OAuth, serverless workers scaling to zero when idle, Nix reproducible environments, modular skills system (GitHub, Google, Notion, Jira), and configuration entirely through lobu.config.ts with TypeScript authoring functions. Agents can be developed by editing IDENTITY.md, SOUL.md, and USER.md files in each agent workspace.
Lobu pricing
Pricing model: Freemium
Lobu is open-source and self-hostable for free. The managed Lobu Cloud pricing is comparable to Railway's compute rates with per-second billing for vCPU, memory, and storage. Idle agents cost $0. There is no separate free tier for the managed service - you bring your own API keys (Anthropic, OpenAI, etc.). Free option: add to Slack with BYO keys and nothing to deploy. Self-hosting uses Docker or Kubernetes with the getting started guide covering both. Local embedded deployment uses local PGlite with pgvector enabled, so you can start without Docker or separate Postgres.
Lobu pros
- Open-source with self-hosting option via Docker/Helm
- 50+ built-in data connectors out of the box
- On-device connectors capture context cloud agents miss
- Per-user OAuth isolation for multi-tenant security
- Workers never see real API keys via secret proxying
- Serverless workers scale to zero when idle (cost efficient)
- Isolated containers prevent agent environment corruption
- MCP proxy handles authentication transparently
- Config-driven LLM provider registry (OpenAI, Anthropic, Groq, Cohere)
- Nix reproducible environments for deterministic package management
- 12 working use-case examples on GitHub ready to deploy
- Same config works for local, self-hosted, and Lobu Cloud
- Watchers turn raw events into typed records without ETL
- Agents act as Slack bot, API, or another agent
- Internal-only Docker network blocks unauthorized outbound access
- Persistent volumes keep session state across restarts
- Skills system modularizes tools, MCP servers, and permissions
Lobu cons
- Requires TypeScript knowledge for custom connectors
- Postgres database required (only external dependency)
- Self-hosting requires Docker/Kubernetes expertise
- Stock OpenClaw runs as long-lived process without Lobu
- MCP HTTP is overengineered for some use cases
- No free tier mentioned for Lobu Cloud managed service
- Learning curve for lobu.config.ts configuration
- Primarily focused on OpenClaw workloads only
Frequently asked questions about Lobu
What is Lobu?
Lobu is the open-source backend for multi-user AI agents. It gives every user or channel an isolated agent workspace with connected sources, shared memory, and per-user OAuth isolation. Build proactive AI agents on a graph that builds itself - connect your company's data in real time, plug in your model, and let agents act the moment something changes as a bot, API, or another agent.
How do I connect Claude to Lobu?
Open Settings → Connectors → Add Custom Connector in Claude Desktop or claude.ai, paste https://lobu.ai/mcp as the MCP URL, complete the Lobu sign-in flow, enable the connector, and choose the workspace Claude should use. For Claude Code, run 'claude mcp add --transport http lobu https://lobu.ai/mcp' and complete the OAuth flow when prompted.
Where can I run Lobu?
Lobu runs in three environments: Local (embedded single Node process with gateway, workers, memory, embeddings all in one; Postgres is the only external dependency), Self-host (Docker and Helm charts in the repo for GCP, AWS, Fly, Render, or bare metal), and Lobu Cloud (managed runtime with per-user isolation, secret proxy, and automatic upgrades).
What connectors does Lobu support?
Lobu has 50+ built-in connectors, supports any MCP server, and lets you build custom connectors in TypeScript. On-device connectors even capture context no cloud agent can see. Connectors are referenced explicitly from lobu.config.ts via connectorFromFile, so they can live anywhere - next to the config for small agents or in folders as you grow.
How does agent memory work in Lobu?
Watchers turn raw event streams into typed, queryable records the moment events arrive or on a schedule. You describe what to track in plain language with no ETL to maintain. The memory schema (entity types, relationship types, watchers) lives in lobu.config.ts via defineEntityType, defineRelationshipType, and defineWatcher. 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.
What is the MCP endpoint for Lobu?
https://lobu.ai/mcp is the canonical MCP endpoint. Sign in once, then use org-scoped memory, connectors, watchers, and tools from any MCP-capable agent. The same URL works for every workspace. For the managed cloud, use https://app.lobu.ai/mcp which works for every workspace and can be scoped to a specific example workspace.
How does credential isolation work?
Workers don't receive real API keys, ever. The gateway generates placeholder tokens (lobu_secret_<uuid>) and passes them instead. The real credentials stay in Redis. All outbound traffic flows through the gateway's HTTP proxy, which swaps placeholders for real keys at request time. A compromised worker literally doesn't have the secrets. The gateway is the single egress point with secret swapping, HTTP proxy with domain allowlist, and MCP proxy with per-user OAuth.
What are Lobu skills?
Skills are modular bundles of tools, MCP servers, permissions, and system packages. A skill declares what it needs: OAuth integrations, Nix packages, network domains to allowlist, and custom tools to expose. System skills (GitHub, Google, Notion, Jira, etc.) are built in. Users install additional skills from the settings page, and agents can request skill installation mid-conversation - the user gets a prefilled settings link, approves, and the agent resumes.
What LLM providers does Lobu support?
Lobu has a config-driven provider registry supporting Groq (llama-3.3-70b-versatile), Cohere (command-r-plus), OpenAI (gpt-4o), Anthropic, and more. Provider configurations are loaded from a single source of truth. You bring your own provider keys and can switch models per agent through lobu.config.ts.
How do I deploy to Lobu Cloud?
Push your project up with 'lobu apply' using the same lobu.config.ts + *.connector.ts + agents/ structure you use locally. The deployment shows your org, region, number of agents deployed, and gateway URL (<your-org>.lobu.run). The Cloud web app and CLI talk to the same org-scoped REST API, so anything you do in the UI (add providers, connections, skills; view agent status) you can also drive from lobu.config.ts and lobu apply.