
nullclaw
An AI agent doorman on a $7/month VPS with IRC as transport layer for code-backed portfolio inquiries.
About nullclaw
nullclaw is a public-facing AI agent that runs on a minimal VPS and uses IRC as its transport layer. It connects to a user's GitHub repositories and portfolio context to answer questions from visitors with code-backed answers rather than rephrased resume text. It features two-tier architecture with nullclaw (public doorman) and ironclaw (private agent with email/calendar access), cost-controlled inference via Haiku and Sonnet models, and sandboxed execution with audit logging.
Key Features
- Code repository cloning and analysis
- GitHub repo integration
- IRC transport layer
- Two-tier agent architecture (public/private)
- A2A protocol support for inter-agent communication
- Cost-capped inference ($2/day hard limit)
- Sandboxed agent execution with action limits
- Audit logging for all tool calls
- Portfolio context preloading
- Cloudflare proxy with TLS termination
AI Models
Use Cases
Best For
Integrations
Pros & Cons
- Code-backed answers instead of rephrased resume text
- Minimal resource footprint (678 KB Zig binary, ~1 MB RAM)
- Complete infrastructure control with no vendor lock-in
- Cost-controlled with $2/day hard cap
- Two-tier security architecture with isolated private context
- As an IRC-based bot, responses may not be as polished as web chat interfaces
- Limited to information available in GitHub repos and preloaded context
- Self-hosted requires VPS management and maintenance
Similar Tools
Live simulation of AI agents scamming each other and getting caught
AI agent that completes knowledge work tasks independently with computer use and scheduling capabilities

