{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/104cf5ebcfc144a09f49c62830755408\" frameborder=\"0\" width=\"2304\" height=\"1728\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1728,"width":2304,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1728,"thumbnail_width":2304,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/104cf5ebcfc144a09f49c62830755408-ca93fcf538f2885c.gif","duration":104.969,"title":"Autopatch AI for Safe Autonomous Code Fixes","description":"This Loom explains how Autopatch uses a zero-trust Docker-based sandbox to safely apply AI-generated code fixes. It runs patches in ephemeral containers with strict OS quotas such as 256 MB RAM and 128 PIDs and with network severing to prevent malicious payloads. To reduce overhead, it uses an AST-based parser to extract exact line-level diffs rather than sending whole files, cutting processing latency and payload size by up to 80%. In the front-end flow orchestrated by a LangGraph state machine, an initial unified diff patch fails but the execution arbitrator rolls back the file system and retries with a search and replace fallback, succeeding on attempt 3 while streaming telemetry to a React front-end without blocking the loop."}