nanocoai/nanoclaw

Project overview

The project uses OS-level container isolation to run agents with controlled mounts, and emphasizes a small, understandable one-process codebase.

Project type

AI Agent · AI Coding

Use cases

Coding & Development

Deployment

Refer to project documentation

License

MIT

Best for

• Users and developers who want to run automated agents with OS-level isolation and are comfortable customizing behavior by modifying code or adding skills.
• Environments where specific functionality is requested on demand, such as adding only selected channel adapters to a fork.

Key capabilities

• Runs agents in separate Linux or Docker containers with filesystem isolation and explicitly controlled mounts.
• Routes messages between channel adapters, agent groups, sessions, and containerized agents; channel adapters are installed separately through skills.
• Supports recurring jobs executed by an agent, with optional script gates to avoid waking the agent when there is no work.
• Provides each agent group with separate instructions, memory, container, and explicitly allowed mounts.
• Creates ready-to-run agents from reusable bundles containing instructions, MCP tools, and skills without secrets.
• Extends a fork by copying requested channel, provider, or other modules into standard paths through skills.

Limitations and risks

• Changing behavior is documented as code modification rather than configuration-file editing.
• The trunk repository ships channel registry and infrastructure, while specific channel adapters are installed on demand through skills.
• Users should review what they are running and which directories are explicitly mounted to manage isolation boundaries.

Getting started

• The setup difficulty is medium. A scripted installer bootstraps Node, pnpm, and Docker, registers credentials, builds the agent container, and pairs an initial channel.
• Users clone the repository, run the bash installer script, pair a first channel, and message the named agent.