PromethealDocumentation

Autonomous AI agents in network-isolated sandboxes. Self-hosted, open-source.

Agent Runtime

Server-side LLM loop with sandbox tools, browser automation, persistent memory, MCP integrations, and streaming — up to 30 tool rounds per turn.

Multi-Agent Teams

Agents delegate tasks to other agents. A coordinator routes to specialists — each with its own sandbox, tools, and model. Secure by design.

Sandbox Isolation

Every agent runs in its own Docker + gVisor container (or E2B microVM) with persistent workspace, browser, and iptables blocking all outbound traffic except to Prometheal.

MCP Integrations

Connect agents to GitHub, Google Workspace, Slack, Notion, PostgreSQL, and more. Credentials never enter the sandbox.

External Channels

Let users interact with agents via Slack or Telegram. Webhook-verified, with automatic conversation threading.

Configuration

Environment variables, per-agent model selection, spending limits, role-based access, and domain whitelisting.

Deployment

Single-instance Docker Compose, multi-instance with Redis + nginx, reverse proxy setup, and health checks.

API Reference

40+ REST endpoints covering agents, chat, integrations, sandbox control, audit, and more.