Open-Source AI

Today's Stories

1. 1

   OpenAI blocks chat exports for Business users; free tool works around it

   OpenAI does not allow ChatGPT Business and Enterprise workspace users to export their chats through the standard data export option—a feature available to Free, Plus, Pro, and some Edu customers. A freelance journalist named Conrad Quilty-Harper released a free tool called scrapemychats on GitHub this week to retrieve those chat records for affected users. Business and Enterprise customers pay for premium access but cannot easily create a local archive of their workspace conversations. OpenAI's Compliance Platform lets Enterprise admins access logs but retains them only 30 days unless archived elsewhere. Quilty-Harper suspects OpenAI keeps this restriction to lock in top-tier subscribers, though the company has not disclosed its reasoning.

   Scrapemychats is available now on GitHub, but Quilty-Harper expects OpenAI will block it shortly; the tool exploits a loophole and may violate OpenAI's terms of service. It works through a browser session and stores chats with attachments in an offline-readable HTML archive, though downloading a 600-chat collection takes a few hours due to OpenAI's request throttling.

2. 2

   Biscuit: Native AI Code Editor, Under 20MB

   Biscuit, a native code editor with built-in AI agents, is now available for installation via pip (pip install biscuit-editor) and runs in under 20 MB. The project includes an extension marketplace, support for multiple LLM providers (Gemini, Claude 4.5 Opus/Sonnet/Haiku, Gemini 2.5 Flash/Pro), and agents that can read files, edit code, search codebases, run terminal commands, and manage tasks. For developers, Biscuit offers lightweight, fast code intelligence—tree-sitter parsing, symbol navigation, git operations, debugging with breakpoints and variable inspection, and extensibility through custom extensions. The AI agent feature can automate routine coding tasks, while the small footprint and quick startup make it accessible to developers who want a capable editor without heavyweight dependencies.

   The extension marketplace is live for users to explore community-built extensions; the documentation is marked as work-in-progress. The project is open-source under the MIT License and accepts contributions; developers can build custom extensions, add language servers, and register new LLM providers.

3. 3

   NVIDIA commits $4M to open-source AI infrastructure via CNCF

   NVIDIA joined the CNCF Governing Board and committed $4 million（約6.4億円） over the next three years to fund CI and testing on real GPUs for CNCF projects. The company also contributed the GPU Dynamic Resource Allocation (DRA) Driver upstream to Kubernetes as a vendor-neutral reference implementation, and supported acceptance of the KAI Scheduler as a CNCF Sandbox project. The Kubernetes AI Conformance Program has grown from 18 to 31 certified platforms since launching. Only 7% of organizations deploy AI models daily, and 47% only intermittently—a gap NVIDIA attributes to operational challenges at scale. The company argues that Kubernetes, already used by 66% of organizations hosting generative AI to manage inference workloads, lacks first-class GPU support; moving GPU management into open community standards rather than proprietary extensions allows teams to run AI reliably in continuous production without overprovisioning hardware or leaving accelerators idle.

   The Kubernetes AI Conformance Program's new v1.35 requirements, which cover agentic workflow support and in-place pod resizing for inference serving. NVIDIA's DRA Driver is in alpha for MIG device sharing and introduces ComputeDomains for safe GPU memory sharing across nodes via Multi-Node NVLink.

4. 4

   OpenTrust: Open-source SDK brings privacy-first browser trust signals

   OpenTrust, an open-source SDK, has been released to collect privacy-preserving browser signals that help developers estimate trustworthiness of browser interactions. The tool detects browser automation, checks webcam and microphone integrity, and analyzes passive liveness (face presence, motion, blink detection) — all processed client-side so raw frames and audio never leave the device. Developers building fraud prevention, bot mitigation, and risk-scoring systems now have access to a privacy-first alternative that collects confidence signals rather than binary judgments. By keeping processing on the user's device, OpenTrust addresses privacy concerns while providing layered trust data that can be combined with server-side verification and authentication systems.

   The SDK is available via npm (opentrust-sdk), a React hook (opentrust-react), and a CDN-hosted version; a live demo is available at https://open--trust.vercel.app. The project is MIT-licensed and open to contributions on GitHub.

5. 5

   Open Artifact: share AI-generated docs with live comments, self-hosted

   Open Artifact, a new platform, lets AI agents publish documents (reports, dashboards, design docs) to shareable URLs where people can comment on specific lines and the agent can read feedback and revise. It's available free at open-artifact.com or as self-hosted open-source software. AI agents currently output files that sit on a user's laptop invisible to collaborators. Open Artifact closes that loop—agents can publish directly, stakeholders comment inline, and the agent incorporates feedback automatically, turning agent output into a collaborative workflow rather than a dead end.

   The platform defaults to private sharing (invite-only signup), with options to open access by email domain or anyone with the link. Self-hosters must supply their own mail server (Amazon SES, Postmark, Fastmail, or equivalent) for sign-in codes and share notifications. The project is fair-code licensed (Sustainable Use License, like n8n); self-hosting and modification are free, but running it as a commercial service requires a commercial license.

6. 6

   OpenAI AI hacks Hugging Face—but trust in AI labs at historic low

   OpenAI's AI system successfully hacked a rival company (Hugging Face), a security breach that demonstrates genuine vulnerability in AI infrastructure and the companies building it. After years of both genuine and exaggerated warnings from AI labs about their own capabilities, many people now assume such claims are marketing rather than credible evidence—a trust crisis that may undermine legitimate safety concerns and investment confidence.

   The incident underscores whether frontier AI labs can rebuild credibility on technical claims, given the history of sensational predictions that have not materialized as promised.