Independent developer · open-source systems

I build the quiet layer between agents and systems.

My work sits where AI agents meet Linux, automation, and the awkward edges of real software.

Systems / agents / tools — joined through small, inspectable interfaces.

A filesystem as an interface

CortexFS

Models, agents, tools, and sessions appear beneath /ctx. Runtime state becomes visible, scriptable, and governed by the operating system around it.

SourceDocumentation
/ctx
modelsinference
agentsruntime
toolscapability
sessionsdurable state
agentrequestjudgmentresume

A human is not an exception path

human-mcp

An agent can ask for judgment, wait without losing context, and continue once a person responds. Human input becomes part of the protocol—not a broken automation.

SourceOpen human-mcp

Selected systems and experiments

A working index

Different scales, one recurring question: where should the boundary live?

01CortexFSA Linux filesystem ABI for models, agents, tools, and durable sessions.02human-mcpA human-in-the-loop bridge for MCP workflows.03MagicNetNetwork tooling shaped around real deployment constraints.04LightFlowLightweight, composable automation for repeatable work.05AstrBotContributing to an extensible open-source agent ecosystem.06OniModsMods and utilities made for Oxygen Not Included.

Working practice

Infrastructure should disappear when it works.

Clear boundaries, small interfaces, and enough observability to recover when it doesn’t.

  1. 01Make the boundary explicit.

    Interfaces should reveal responsibility.

  2. 02Prefer primitives to platforms.

    Small pieces remain useful longer.

  3. 03Keep the system inspectable.

    State should be possible to understand.

  4. 04Ship what survives reality.

    The useful test begins after the demo.

Code, issues, and ongoing work

The work stays open.

Found a rough edge? Open an issue. Found an idea? Start a conversation where the code lives.

Visit GitHub