I build open-source infrastructure for reliable AI agents: deterministic tool flows, context control, and safer tool execution.

Compile predictable multi-tool MCP flows into executable graphs and remove unnecessary LLM calls between steps.

Budget-aware context compilation and context firewall for tool-heavy AI agents.

Capability-based authorization and policy enforcement for agents operating across large tool ecosystems.

• reliable agent systems over “agent vibes”
• deterministic execution where reasoning is not needed
• context discipline instead of prompt sprawl
• safer tool use at scale

Open to feedback, issues, design discussion, and collaboration.