File size: 13,944 Bytes
fde410c 5550fe8 fde410c 5550fe8 fde410c 5550fe8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | 0,1,2,3,4,5
3534,openai_blog,https://openai.com/index/safety-overview-gpt-6-astra,Safety overview: GPT-6 Astra,Safety overview: GPT-6 Astra. GPT-6 Astra is our most capable broadly deployed model and our first to reach the Critical level of cybersecurity capability under our Preparedness Framework.,2026-09-04T07:30:18.648271
3533,openai_blog,https://openai.com/index/playco-game-prototyping-with-astra,Playco cut manual fixes 50% prototyping games with GPT-6 Astra,"Playco cut manual fixes 50% prototyping games with GPT-6 Astra. Using GPT-6 Astra, Playco built three themed game prototypes from one grey box foundation and reported 50% fewer manual fixes than with the previous model.",2026-09-04T07:30:18.643868
3532,openai_blog,https://openai.com/index/legora-financial-statement-review-with-astra,Legora reviewed 41 documents in minutes with GPT-6 Astra,"Legora reviewed 41 documents in minutes with GPT-6 Astra. Legora used GPT-6 Astra to review 41 documents in minutes, find all four planted errors, and improve performance by nearly 40% in this financial-review workflow.",2026-09-04T07:30:18.639612
3531,openai_blog,https://openai.com/index/daybreak-for-frontline-defenders,Daybreak for Frontline Defenders: $1B to protect essential services,"Daybreak for Frontline Defenders: $1B to protect essential services. OpenAI introduces Daybreak for Frontline Defenders. A $1 billion commitment expands access to frontier cyber AI, training, and support for essential services.",2026-09-04T07:30:18.634018
3516,hackernews,https://github.com/Cascadia-PLM/Cascadia-App,"Show HN: Open-source, Git-inspired versioning for hardware design – Cascadia PLM","Show HN: Open-source, Git-inspired versioning for hardware design – Cascadia PLM. PLM (product lifecycle management) software is the ugly middle generation between simple CAD-data-management (typically called "PDM" or Product Data Management) and true Digital Threads. No one likes their PLM, but when you need it, you need it. And the kings (the old monsters) of PLM are all billion-dollar companies who charge massive enterprise subscriptions, and add massive enterprise consulting fees for implementation and support on top of that. Also, their PLM is built for the Fords and Boeings of the world, not the little guys, who need to move faster and be "more messy."<p>I say phooey to that.<p>Cascadia is my brainchild of the last few years. PLM that is more than just PLM, because small manufacturers often don't have clear boundaries between Engineering and Manufacturing. PLM that is Digital Thread directed from the beginning, because context is what's important to engineering decisions, and never has that been more clear than now, in the AI age. PLM that you can self-host, on Linux, with Postgres, so that you don't have to pay massive Windows Server / SQL Server / Oracle DB licenses on top of your subscription. PLM that's code-first because low code is the death of maintainability in enterprise software. PLM that your engineers might actually want to use, rather than working around, because it makes their lives easier, not harder.<p>Anyway, thanks for making it through my rant. Check it out at the link.",2026-09-04T07:30:16.743843
3501,hackernews,https://github.com/obinexus/mmuko-boot,Mmuko Boot Sequence – a tiny freestanding C kernel that boots in QEMU,Mmuko Boot Sequence – a tiny freestanding C kernel that boots in QEMU. ,2026-09-04T07:30:15.281160
3492,hackernews,https://hotdogbenchmark.lol/,Show HN: Hot. Dog. Bench. Mark. The AI benchmark we deserve,Show HN: Hot. Dog. Bench. Mark. The AI benchmark we deserve. ,2026-09-04T07:30:14.083865
3491,hackernews,https://vericommand.net/benchmark,"Show HN: I measured resuming an AI coding session: 22,897 tokens vs. 1,013","Show HN: I measured resuming an AI coding session: 22,897 tokens vs. 1,013. ",2026-09-04T07:30:14.079659
3489,hackernews,https://context.apimatic.io/,Show HN: A Context Registry for AI coding agents,"Show HN: A Context Registry for AI coding agents. Hi HN. We built an API context registry to help coding agents (like Claude Code) generate production-ready API integration code without blowing through token limits.<p>We build a lot of API integrations. In our experience, most coding agents write basic client calls fine, but consistently stumble on details that make code shippable, like idempotent retries, rate-limiting and Auth token management.<p>We tried all the existing approaches of injecting context into coding sessions:<p>- Markdown dumps delivered via MCP (think Context7 or Mintlify Docs MCP)
- API behaviour described in prose using AGENTS.md and skills.
- OpenAPI specs<p>However, all of them left the same production-readiness gaps.<p>So we came up with our own approach that combines prose with typed SDK reference code into a "Context Plugin". You install the plugin into your coding agent and it automatically injects language-specific context whenever the agent works on an API.<p>Across our benchmarks, Context Plugins boosted one-shot production readiness by up to 34%, allowing Sonnet to match or beat baseline Opus on the same integration tasks. You can read more about our experiments here <a href=""https://www.apimatic.io/blog/working-api-call-is-not-production-ready-integration"" rel=""nofollow"">https://www.apimatic.io/blog/working-api-call-is-not-product...</a><p>We have published Context Plugins for 24 APIs for the community to try out, including Slack, Google Maps, and Notion.<p>We'd love for you to give them a go and share your feedback on our plugins as well as our evaluation methodology.",2026-09-04T07:30:14.068841
3488,hackernews,https://polyform.ai/news/why-i-hate-benchmarks/,I hate benchmarks: Moving a production workflow from GPT to GLM-5.3 Flash,I hate benchmarks: Moving a production workflow from GPT to GLM-5.3 Flash. ,2026-09-04T07:30:14.063516
3487,hackernews,https://www.cnbc.com/2026/09/03/open-ai-astra-gpt-6-cyber.html,OpenAI begins rolling out GPT-6 Astra,"OpenAI begins rolling out GPT-6 Astra. <a href=""https://thenewstack.io/openai-gpt6-astra-benchmarks/"" rel=""nofollow"">https://thenewstack.io/openai-gpt6-astra-benchmarks/</a>, image: <a href=""https://cdn.thenewstack.io/media/2026/09/358eb84a-screenshot-2026-09-03-at-10.51.35-am.png"" rel=""nofollow"">https://cdn.thenewstack.io/media/2026/09/358eb84a-screenshot...</a><p><a href=""https://venturebeat.com/technology/welcome-to-the-agi-era-openai-launches-gpt-6-astra"" rel=""nofollow"">https://venturebeat.com/technology/welcome-to-the-agi-era-op...</a><p><a href=""https://www.theverge.com/ai-artificial-intelligence/988334/openai-astra-ai-monitoring-safety"" rel=""nofollow"">https://www.theverge.com/ai-artificial-intelligence/988334/o...</a><p><a href=""https://twitter.com/OpenAI/status/2095595741528125780"" rel=""nofollow"">https://twitter.com/OpenAI/status/2095595741528125780</a>",2026-09-04T07:30:14.057028
3486,hackernews,https://conversion.ai/blog/text-to-query-agent/,Using semantic benchmarks to build a self-improving text-to-query agent,Using semantic benchmarks to build a self-improving text-to-query agent. ,2026-09-04T07:30:14.049410
3474,hackernews,https://github.com/shutter-network/concorde,"Show HN: Concorde, one AI agent shared by an organization","Show HN: Concorde, one AI agent shared by an organization. We built Concorde, an open-source framework giving an organization one shared agent that's controlled by a group, rather than just one
person.<p>This is pretty cool because shared agents can bring us closer to single entities doing what took a whole organization to do, kind of moving us in the direction of the trend of the rise of the solo founder.<p>Saying that entire organisations could be just 1 agent might sound sort of anti-democratic, but this is the reality of where things are already heading.<p>The "organisation" only exists because humans hold limited context. agents don't have that limit. Corncorde changes that by allowing everyone to interact with the same agent the same way, and it builds a common history.<p>By making this power structure explicit, we can set up shared agents more intentionally and via guardrails, governance, TEEs, zkml and other tech, we'll get the best of both worlds:<p>- extreme efficiency and coherence in action from it being one entity<p>- accountability, transparency, representation of diverse opinions no member owns it, none can reach it privately<p>Concord is alpha, get in touch if you'd like to launch a shared agent! We also posted an X thread and blog post about it<p><a href=""https://x.com/ShutterNetwork/status/2095090562785587490"" rel=""nofollow"">https://x.com/ShutterNetwork/status/2095090562785587490</a>
<a href=""https://blog.shutter.network/concorde-a-framework-for-shared-agents/"" rel=""nofollow"">https://blog.shutter.network/concorde-a-framework-for-shared...</a>
Prior blog post exploring the shared agents space more generally: <a href=""https://blog.shutter.network/on-shared-agents/"" rel=""nofollow"">https://blog.shutter.network/on-shared-agents/</a><p>Would be very interested in feedback on where this shared agent model breaks down, compared to other multi agent organizations.",2026-09-04T07:30:12.809475
3472,hackernews,https://github.com/obinexus/libpolycall-v1,Show HN: LibPolyCall – a C runtime broker for cross-language function calls,"Show HN: LibPolyCall – a C runtime broker for cross-language function calls. LibPolyCall is an open-source C runtime for connecting programs across language boundaries without requiring each language pair to implement its own integration layer.<p>The architecture is program-first rather than binding-first, with a stable C ABI, FFI bindings, Polycallfile/Polycallrc configuration, runtime state management, and telemetry.<p>I’ve recently completed the Windows build path producing both libpolycall.dll and libpolycall.a, and I’m working toward using the same runtime across Python, Node.js, Java, Go, and other language environments.<p>I’d particularly appreciate feedback on the ABI design, runtime architecture, configuration model, and approach to cross-language dynamic loading.",2026-09-04T07:30:12.794792
3456,hackernews,https://mistral.ai/news/agentic-search/,Agentic Search,Agentic Search. ,2026-09-04T07:30:11.514078
3444,hackernews,https://en.wikipedia.org/wiki/Guard_llama,Guard Llama,Guard Llama. ,2026-09-04T07:30:10.283294
3442,hackernews,https://tdqs.dev,Show HN: MCP Tool Definition Quality Score (TDQS) Spec,"Show HN: MCP Tool Definition Quality Score (TDQS) Spec. Hey everyone,<p>You may know me because of my Open-Source work like awesome-mcp-servers, FastMCP (node.js), ViteMCP, mcp-proxy, mcp-remote, and a few other projects in the MCP ecosystem, including Glama.<p>I was lucky enough to be present when MCP was first announced. That let me to contribute to the foundations of this new protocol and everything that has evolved around it. It also let me to be at the center of a lot of feedback, and by far the biggest complaint about the MCP ecosystem has been the inconsistent quality. Quality here means a lot of things, but server JSON definition is a big part of it. Bad tool definitions mean that tools are not selected when they should be, they are when they shouldn't, they are improperly invoked, etc.<p>TDQS is an open-source specification (<a href=""https://github.com/glama-ai/tool-definition-quality-score"" rel=""nofollow"">https://github.com/glama-ai/tool-definition-quality-score</a>) for evaluating the quality of the MCP server definitions. It's not a complete solution to the quality problem, but it is a research based rubric that increases clarity over what tools are available, what are their behaviors/purpose, and when/how they are supposed to be used.<p>TDQS is what Glama uses to score 15,000+ Open-Source and remote MCPs. And <a href=""https://tdqs.dev"" rel=""nofollow"">https://tdqs.dev</a> is a free website to promote the spec and increase the adoption through better documentation and easy to use playground/CLI/API/SDKs.<p>Would love your feedback and participation in improving the quality of the MCP ecosystem.",2026-09-04T07:30:10.274375
3441,hackernews,https://news.ycombinator.com/item?id=49557409,Show HN: Building AI agents client-side JavaScript,"Show HN: Building AI agents client-side JavaScript. Hey HN, most agent systems default to server-side Python inside containers and chain frameworks. I wanted to see how far we could push agent in the browser with vanilla JavaScript <a href=""https://buttercup.sh"" rel=""nofollow"">https://buttercup.sh</a><p>The reason this is interesting is because agent loops in the browser keeps infrastructure costs low. No need for proxy or API calls. And ollama/vLLM can be used for 100% offline. Also WebLLM for embedded. We need to consider CORS, API keys for remote models, access to visual state, and handling remote tool calls. I am working on a guide with references in vanilla JS. This is a short-lived guide starting mid-September with weekly topics.<p>Draft topics starting mid-September:<p>In-Browser Loops: Function calling, and deterministic multi-turn loops running purely in the browser runtime.
Vision (Multimodal): Capturing viewport screenshots using browser APIs.
Remote Agent Access & Transports: Connecting the in-browser agent to remote agents.
|