paradigmxyz/centaur

What it solves

Centaur provides a self-hosted platform for teams to share a single, centralized AI agent instead of relying on fragmented, local setups. It solves the problem of giving agents secure, standardized access to internal tools, code repositories, and execution environments while maintaining a clear audit trail and credential security.

How it works

The system operates as a control plane that bridges communication between interfaces (like Slack) and isolated execution environments. When a user asks a question in Slack, Centaur assigns a Kubernetes sandbox containing a shell, workspace, and development tools (Python, Node.js, etc.). The agent runs within this sandbox, utilizing a Rust-based API for state management and Python-based plugins for specific tools and durable workflows. Outbound network access and credential injection are managed via a proxy to ensure agents never handle raw API keys.

Who it’s for

It is designed for engineering and operations teams who need AI agents to perform real work—such as investigating CI failures, running recurring checks, or coordinating multi-step operational workflows—within a secure, shared corporate environment.

Highlights

• Slack-native integration: Interact with agents directly through Slack threads with progress updates and final answers.
• Isolated Sandboxes: Every conversation runs in its own Kubernetes sandbox with a full development environment.
• Flexible Agent Harnesses: Supports various CLI-based agents like Claude Code, Amp, or Codex.
• Durable Workflows: Ability to run long-running jobs that can sleep, resume, and survive service restarts.
• Secure Credential Handling: Uses a proxy to inject secrets into outbound requests so agents never see raw API keys.
• Organization Overlays: Allows teams to layer custom tools, personas, and prompts without modifying the base platform.