Lobu

Executive Summary

Lobu is multi-tenant, sandboxed agent orchestration for teams. It runs Claude Code or OpenClaw behind a hardened gateway with MCP proxy, multi-provider auth, and per-context isolation. Originally launched as peerbot.ai in July 2025, it evolved to support both Claude SDK and OpenClaw runtimes while providing enterprise-grade security that OpenClaw lacks out of the box.

Attribute	Value
Repository	github.com/lobu-ai/lobu
Language	TypeScript
Stars	36 ★
License	BUSL-1.1
Created	July 2025
Creator	@bu7emba

---

Product Overview

Lobu's core value: run powerful AI agents (Claude Code or OpenClaw) without exposing your infrastructure. All worker traffic routes through a gateway proxy — workers have no direct network access. Secrets stay isolated, MCP tools are proxied with scoped tokens, and network egress is domain-filtered.

Key Capabilities

Capability	Description
Dual Runtime	Claude Code SDK (CLI subprocess) or OpenClaw Pi Agent SDK per-agent
Multi-tenant	Single bot serves Slack, Telegram, WhatsApp, REST API with per-channel isolation
MCP Proxy	Gateway handles OAuth, injects scoped tokens — workers never see client secrets
Scale to Zero	Workers scale down when idle
Network Filtering	Whitelist domains workers can access
OpenClaw Compatible	Supports skills, IDENTITY.md, SOUL.md, USER.md

Interfaces

Interface	Description
REST API	Programmatic agent creation
Slack	Multi-channel/DM agents
Telegram	@lobuaibot
WhatsApp	Baileys-based integration

---

Technical Architecture

Slack/Telegram/WhatsApp/API → Gateway → Orchestrator → Worker (Claude/OpenClaw)
                                ↑                           ↓
                              Redis                    HTTP/MCP Proxy
                                                           ↓
                                                    Domain-filtered Internet

Key Technical Details

Aspect	Detail
Deployment	Kubernetes, Docker Compose, Local
Runtimes	Claude Code SDK, OpenClaw Pi Agent SDK
Isolation	gVisor/Kata on K8s, NetworkPolicies, RBAC
MCP	Proxied with OAuth, env var substitution
Environments	Nix-based reproducible per-session tooling
Open Source	Yes (BUSL-1.1)

---

Lobu vs OpenClaw

Aspect	Lobu	OpenClaw
Scale	Workers scale to zero	Always-on
Multi-tenant	Per-channel/DM isolation	One instance per setup
Platforms	Slack, Telegram, WhatsApp, API	15+ chat platforms
Runtimes	Claude SDK + OpenClaw	OpenClaw only
MCP Access	Proxied, secrets isolated	Direct from agent
Network	Sandboxed, domain-filtered	No built-in isolation
Deployment	K8s, Docker, Local	Single node

---

Strengths

• True isolation — Workers have no direct network access. All traffic routes through gateway with domain filtering.
• Secrets never exposed — MCP OAuth, provider credentials, and env vars resolved at gateway level. Workers never see client secrets.
• Dual runtime flexibility — Choose Claude Code SDK or OpenClaw Pi Agent SDK per-agent based on use case.
• OpenClaw compatibility — Skills from ClawHub, IDENTITY.md, SOUL.md, USER.md all work. Migration path from vanilla OpenClaw.
• Enterprise deployment — Kubernetes with NetworkPolicies, RBAC, gVisor/Kata. Production-ready.

---

Cautions

• Early stage — 36 stars, recently published. Limited community validation.
• BUSL-1.1 license — Not truly open source. Commercial use restrictions apply.
• Fewer platforms — Slack, Telegram, WhatsApp, API only. OpenClaw has 15+ integrations.
• Complexity — Gateway proxy architecture adds operational overhead vs simple single-node OpenClaw.
• Solo maintainer — Single developer (@bu7emba). Bus factor concern for enterprise adoption.

---

Pricing & Licensing

Tier	Price	Includes
Self-hosted	Free	Full features, BUSL-1.1 license

Licensing model: Business Source License 1.1 (BUSL-1.1)

Hidden costs: Infrastructure (K8s cluster or Docker host), LLM API costs

---

Competitive Positioning

Direct Competitors

Competitor	Differentiation
Tensol	Lobu is self-hosted; Tensol is managed
NanoClaw	Lobu has dual runtime (Claude + OpenClaw); NanoClaw is OpenClaw-only
LaunchClaw	Lobu is open source; LaunchClaw is hosted
OpenClaw	Lobu adds isolation, multi-tenant, scale-to-zero

When to Choose Lobu Over Alternatives

• Choose Lobu when: You need multi-tenant isolation with Claude Code or OpenClaw
• Choose Tensol when: You want managed hosting without infrastructure
• Choose OpenClaw when: You need maximum platform integrations (15+)
• Choose NanoClaw when: You want simpler container isolation

---

Ideal Customer Profile

Best fit:

• Teams deploying AI agents to Slack/Telegram with security requirements
• Organizations needing per-channel/user isolation
• DevOps teams comfortable with Kubernetes
• Companies wanting Claude Code + OpenClaw flexibility

Poor fit:

• Individual users wanting simple setup
• Those needing 15+ chat platform integrations
• Organizations uncomfortable with BUSL-1.1 licensing
• Teams without Kubernetes expertise

---

Viability Assessment

Factor	Assessment
Financial Health	Unknown (open source project)
Market Position	Niche
Innovation Pace	Rapid (active development)
Community/Ecosystem	Limited (36 stars)
Long-term Outlook	Uncertain — depends on adoption

Early-stage project with compelling architecture. The security-first approach differentiates from vanilla OpenClaw. Success depends on community growth and potential commercial backing.

---

Bottom Line

Lobu is "OpenClaw for teams" — adding the multi-tenant isolation, network sandboxing, and MCP proxy that enterprises need. The dual runtime (Claude Code + OpenClaw) provides flexibility, and the architecture is sound. But at 36 stars with a solo maintainer, it's a bet on early-stage software.

Recommended for: Teams deploying agents to Slack/Telegram who need isolation and are comfortable with Kubernetes

Not recommended for: Individual users, those needing many chat platforms, or organizations requiring enterprise support

Outlook: If the project gains traction and adds more maintainers, Lobu could become the de facto "enterprise OpenClaw" layer. The security architecture is what OpenClaw should have built.

---

Research by Ry Walker Research • methodology