Overview
Letting an AI agent run code from a random bug report is a security problem. The report is untrusted, the code the agent writes is untrusted, and so is the agent's claim that it fixed anything.
Airlock handles that. You paste a real issue for a supported library. The agent reproduces the bug, writes a fix and hands back a patch you can download. All of it runs in throwaway Kata sandboxes on a Vultr VX1 host, with no network, no secrets and hard limits on CPU, memory and time. The agent can edit its own fix. It can't touch the tests that judge it, and it can't decide that it passed. A separate comparator reruns the frozen checks against a sealed copy of the patch and gives the verdict. A fake "all tests passed" log from the agent gets a fail.
It also runs general tasks: web research in a sandboxed Chromium behind an allowlisted proxy, offline data analysis, and form submissions that need a human to approve them first. The model is glm-5.3 on Vultr Serverless Inference.
We ran it on the real python-tabulate #365 bug and it fixed it in 3 of 3 fresh live runs. Paste `rm -rf /` into the hostile input panel and it kills one sandbox. The control plane, the other tasks and the host keep running.
Note: the demo URL is substituted with the GitHub repository link.
Tech stack
No comments yet
Be the first to comment on this project.
sentinel
Full-Stack Self-Healing DevOps & Security Copilot The Concept: An AI operations engineer that doesn't just suggest code fixes—it monitors cloud infra, detects uptime/security drops, automatically provisions fixes, and generates audit reports for enterprise compliance.
Agency Compiler
Agentic creative project compiler. Most creatives, agencies and designers dont know how to use ai, none the less agents. We created a platform that allows a creative agency or studio owner to instantly spin up agents based on the project ask, resources and scale.
ProofRun
ProofRun helps developers answer a critical question before shipping: will this update break behavior users rely on? An AI agent inspects two Git revisions, designs targeted experiments, and proposes repairs. An independent Docker-based verifier runs those experiments against both versions, preserves the original tests, and checks whether the repair actually works. Developers receive an update, skip, or postpone recommendation with recorded evidence and a reviewable patch. In a live synthetic…