An open catalogue of AV control protocols, built for the AI tooling era.
AI4AV publishes machine-readable control specifications for professional AV, lighting, projector, audio, and smart-home devices. Each spec captures the transport, actions, feedbacks, and safety interlocks in a compact format that both humans and LLMs can read.
The catalogue ships with an agent skill, ai4av-lookup, that installs into any AI tool that supports skills (Claude Code, Codex, and compatible runtimes). You ask the agent to control a device by name. The skill pulls the spec. The agent has the data it needs to scaffold a control script, a phone-app remote, or a scripted automation.
AV manufacturers ship long PDF manuals with control commands buried in appendices, formatted as half-readable tables. Hand one to an LLM and the output looks plausible until you try to run it. AV integrators have lived with this for decades. Translating documentation into proprietary control-system drivers is part of the price of doing the job.
LLMs change the economics. An LLM can read the PDF. It cannot read the PDF well enough to act on the result. AI4AV closes that gap by centralising the documentation work once, in a format an agent can use, and publishing it free.
Vesa Laasanen and contributors maintain AI4AV. An automated pipeline ingests manufacturer documentation, refines it into the open spec format, and cross-checks each command against the source document at the byte level before promotion. Specs that fail verification do not ship.
AI4AV is sponsorware. The data is open and free for any use, including commercial. Sponsorship from AV manufacturers, integration shops, and the individuals who rely on the dataset pays for ingestion, schema work, and verification. Become a sponsor on GitHub Sponsors.
Articles on AV control protocols and the open spec format live in Articles. Catalog updates ship in News.
Suggest a missing device, report a wrong spec, or open a pull request against the public mirror on GitHub. A maintainer triages each issue by hand. See How to contribute.
AI4AV