Lab
Tools built against real problems from the field. Each one exists because something I needed didn't.
-
onvif-mcp
AI agents can't talk to security cameras. ONVIF is the industry standard, and it's a swamp of SOAP, profiles, and vendor quirks. onvif-mcp is an MCP server that gives an agent a working camera interface: discovery, streams, PTZ, events.
For anyone wiring agents into camera infrastructure.
GitHub · Launched on Show HN, Aug 2026
-
Agent Action Receipts
When an AI agent acts on your systems, there's no evidence trail you could hand to an auditor. AAR is a spec and tooling for signed, replayable receipts of agent actions. Chain-of-custody thinking, applied to agents.
For teams running agents anywhere "trust the logs" isn't good enough.
-
EdgeProof
Video evidence is only as good as its provenance. EdgeProof signs video at the camera edge using public PKI, so footage can prove it hasn't been altered between the lens and the courtroom.
For agencies and integrators who need footage that survives challenge.
Built on the MIT-licensed signed video framework.
-
camspec
Camera installs get signed off on "it shows a picture." camspec generates pass/fail commissioning reports from what the camera actually delivers: resolution, frame rate, stream health, measured against spec.
For installers and the people who inherit their work.
Ships inside onvif-mcp. PDF report renderer live.
-
Commissioning Walk
Wireless camera links fail in ways a laptop speed test doesn't catch. This experiment measures what a camera actually needs to hold a stream, day and night, and turns it into installer pass/fail thresholds.
For installers certifying wireless camera links.
Thresholds v0.1 measured and frozen, including night and IR conditions.
Building something adjacent?
Open-source collaboration and standards work are the reasons this page exists.