# Stakpak

> Open-source Rust agent that runs on your machines 24/7 to monitor, heal, and operate your apps, pinging a human only when needed.

Stakpak is an open-source DevOps agent from Stakpak, Inc., written in Rust and licensed under Apache 2.0. It runs on your own machines as a single binary, either as an interactive terminal UI or as an autonomous background process. The homepage carries a banner stating that Stakpak is joining Vercel.

## What It Is

Stakpak is an autonomous operations agent for production apps. After install, `stakpak init` discovers your stack and generates an APPS.md file. `stakpak up` then starts Autopilot, a long-running process that combines scheduling, messaging channels, and a local API, and installs as a system service that persists across reboots. You can also work interactively with `stakpak`, or check schedules, channels, and recent activity with `stakpak autopilot status`.

## How Autopilot Works

Schedules run cron-based prompts such as health checks, and channels (Slack, Telegram, Discord) start sessions from inbound messages and deliver notifications. Behavior is controlled by profiles in `~/.stakpak/config.toml` (model, allowed tools, auto-approval, system prompt, max turns, provider credentials), while `~/.stakpak/autopilot.toml` holds schedules, channels, and notification routes. The homepage illustrates use cases like auto-healing after a latency alert, a cost watchdog for idle cloud resources, and renewing expiring certificates. Rulebooks (markdown SOPs and playbooks) customize agent behavior.

## Security Model

The project positions itself for production work rather than code writing. Network requests pass through a Warden proxy with Cedar policies. Secret substitution swaps detected secrets with placeholders before the LLM sees them and restores real values only at execution time. Sessions are audit-logged, file changes are backed up before modification, and MCP traffic uses mTLS by default. A privacy mode redacts items like IP addresses and AWS account IDs.

## Models, MCP, and Editor Integration

You can use a Stakpak API key, your own Anthropic, OpenAI, or Gemini keys, or a custom OpenAI-compatible endpoint such as Ollama or LM Studio. Stakpak can run as an MCP server (local, remote, or combined tool modes) and as an MCP proxy multiplexing upstream MCP servers. It also supports the Agent Client Protocol for editors such as Zed. Infrastructure code (Terraform, Kubernetes, Dockerfile, GitHub Actions) is indexed locally for semantic search.

## Setup Path

Install via the install script, Homebrew, a binary release, or a Docker image that bundles common DevOps CLIs. For remote VMs, Docker access and 2GB+ RAM are recommended, and `stakpak autopilot doctor` checks deployment readiness.

## Features
- Autonomous 24/7 Autopilot runtime
- Interactive terminal UI
- Cron schedules and Slack/Telegram/Discord channels
- Warden network sandbox with Cedar policies
- Secret substitution before LLM sees credentials
- Session audit logs and reversible file operations
- Infrastructure code indexing and semantic search
- Rulebooks for SOPs and playbooks
- MCP server and MCP proxy
- Agent Client Protocol (ACP) support for Zed
- Bring-your-own keys or local OpenAI-compatible LLMs
- Subagents for research and sandboxed analysis

## Integrations
Slack, Telegram, Discord, Anthropic, OpenAI, Gemini, Ollama, LM Studio, Zed, Docker, Homebrew, Terraform, Kubernetes, GitHub Actions, MCP

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, CLI

## Pricing
Open Source, Free tier available

## Version
v0.3.88

## Links
- Website: https://stakpak.dev
- Documentation: https://stakpak.gitbook.io/
- Repository: https://github.com/stakpak/agent
- EveryDev.ai: https://www.everydev.ai/tools/stakpak