Studio / actualis
actualis 0.5 demo video
A 1 minute 43 second demo of actualis 0.5: what your coding agents ran, which credentials reached them, and the new offline host-reputation check.
About this video
The presenter is an AI-generated, fictional character; the terminal output is real actualis 0.5.0 output from an invented demo fleet. The video runs 1:43 and has English captions.
A local, read-only tool that reads existing coding-agent logs and reports command activity, credential exposure, and cost.
Install actualis
$ pip install actualis
$ brew install digital-foundry/tap/actualisTranscript
Your coding agents run shell commands all day. They read env files, install packages and push to GitHub, often with nobody approving each step. So who reviews what they actually did?
Every session leaves a transcript on disk. But it's long, every vendor writes it differently, and nobody reads it. So the questions that matter go unanswered. Did an agent see a production key? What ran without asking?
Actualis answers in one command. It reads Claude Code, Codex, Copilot, Gemini and Qwen sessions on your own machine. In this demo fleet, it found two credentials that reached an agent's context, a Stripe key to rotate first, and five high-risk shell commands.
Put it in your pipeline. With --fail-on high, a leaked key fails the build, and the log tells you why.
The network view lists every download your agents made, and which ones nobody approved. New in version 0.5, it checks those hosts against Google Safe Browsing lists, offline. The only time anything goes online is when you refresh the lists, and it tells you exactly what it sends.
And you don't have to trust us. Self-check proves, on your own machine, that Actualis imports no networking code, and changes no file it reads.
Actualis is free and open source. Install it with pip or Homebrew. If your team runs coding agents at scale, become a design partner at actualis.app.