Quick Start
# 1. Install AgentVisor CLI
curl -fsSL https://raw.githubusercontent.com/manetu/agentvisor/main/install.sh | sh
# 2. Create a new agent project
agentvisor template create langgraph/chatbot-agent
cd chatbot-agent
# 3. Start development services (Temporal, Ollama)
docker compose up -d
# 4. Run the agent
agentvisor serve .
See the Getting Started guide for prerequisites.
Six Pillars of Secure Execution
Sandbox Isolation
Agents run in hardened sandboxes with complete host isolation — no network access, no filesystem access, and 14 independent security layers for defense-in-depth. gVisor adds syscall-level interception for maximum isolation on Linux.
Credential Brokering
Real API keys never enter the agent’s memory. Symbolic tokens are substituted at the host boundary, ensuring credentials remain protected while agents use standard HTTP libraries unchanged.
Policy-Based Governance
Every HTTP request, tool call, MCP operation, and A2A call is evaluated inline by Manetu PolicyEngine. Policy management is unified; enforcement is distributed at every runtime boundary.
Durable, Observable Execution
Temporal workflows provide crash recovery and horizontal scaling. OpenTelemetry tracing, Prometheus metrics, structured logging, and syscall-level runtime tracing give visibility from agent to upstream API.
Multi-Protocol Interoperability
Expose agents via REST, MCP, and A2A simultaneously. Connect agents to external MCP servers and A2A agents through policy-enforced gateway connections. Both directions governed.
Transparent Compatibility
Standard HTTP libraries work without modification. LangGraph applications drop in unchanged. Exec mode runs interactive AI tools — Claude Code, Cursor, Aider — inside the same governed sandbox.