X2f X2f

Evidence

• Show HN: peerd – AI agent harness that runs entirely in your browser — Hey HN. <a href="http://peerd.ai" rel="nofollow">http://peerd.ai</a> is an AI agent harness that lives entirely in your browser as a web extension. You don’t have to install a separate “AI browser”. You don’t have to bolt on or run some external process or manage a clunky mcp integration. It’s just a fully contained web extension, written in no build vanilla JS with minimal non-browser dependencies, using your own provider keys, and Apache 2.<p>This isn’t just a fun hack. While it has largely been a solo side project, I genuinely believe the browser and the web could be the most natural platform for AI agents to operate safely, autonomously, and most importantly without A2A middlemen (more on that in a sec). To demonstrate that point peerd doesn’t just drive browser automation. It spins up isolated sandboxes using tabs and worker instances to support various real workload types. Those include headless JS computational work, visual JS notebooks, personal client side apps, and real Linux VMs on top of wasm with full http networking.<p>The industry discourse over the last several months has been dominated by “which substrate is the best for ai agent sandboxes” with many competing answers focused on different models and use cases. Cloudflare is one of the most prominent examples, positioning its v8 isolate based workers as the best in class solution thanks to faster than container startup times and strong isolation guarantees. The v8 isolate is of course the product of chromium, which runs on billions of browsers around the world for free. The browser as a whole is perhaps the most battle tested sandbox system in the entire software industry. It’s been built on 3 decades of learning from hostile content, hostile code, and hostile users. Native and cloud agents are necessarily rebuilding all or most of this posture from scratch. peerd doesn’t. It leverages everything the browser has to offer and pushes it to its functional limits, while inheriting its security baseline and isolation from the host system.<p>Robust sandboxing isn’t the only thing the browser offers and peerd uses. It comes with extremely powerful and underrated primitives, from webCrypto, webRTC, webAuthn, webGPU, and ~soon WebNN. Direct web access, with your real live sessions, and api calls with fetch present an alternative model to MCP integrations. The agent can write and spawn web apps right there in a tab, no hosted service necessary. Then there’s the A2A piece: peerd already has a rudimentary p2p (peerd-to-peerd?) network in place using webRTC. Today you can connect with peers on the network, add them as contacts, and share signed apps you’ve created. I’m working on extending these apps to be able to leverage the same p2p network to support decentralized web apps (dwapps), as well as facilitate true p2p A2A with no platform or middlemen.<p>Given this is an early part time project, this is an extremely experimental build and in a v0.x preview state. I’ve taken care to attempt to address the lethal trifecta: the main agent loops/sessions never ingest untrusted DOM code or possess low level navigation tools. It delegates those tasks to dedicated web runners with no wider tooling or secrets access that return summarized results. Both the DOM and the summarized results are bracketed as untrusted, meaning two stacked prompt injection escapes are needed. All egress goes through a central module that has a customizable deny list, and only models calls to designated allowed endpoints are possible. See more in the docs, site, and the code itself. Ultimately, use at your own risk.<p>Today anthropic, open router, local ollama, and even an experimental WebGPU instance of Gemma are supported.<p>Honest limitations: Chrome store and AMO are still pending until it can get more eyeballs and live usage. Just loading unpacked from GitHub is the easiest way to go, and as a bonus makes it easy to audit thanks to no build. Linux on wasm depends on the Cheerpx engine, which is not open source and has restrictions for commercial use. That may be a good reason to reassess it compared to alternatives, but it’s also the most performant and looks closest to implementing 64bit support.<p>Poke around, use it, critique it, and have fun.

  HACKER_NEWS

• Show HN: Active Source of Truth for Your Coding Agents — Howdy! Happy Saturday everyone!<p>As a solo founder, I have always tried to maximize my speed by letting coding agents build as much as possible in parallel. However, as an engineer, I don't trust that AI will always make the right decisions and work with the right context. In the past, I always needed to click through my sessions to glance at the AI's output, try to understand what it was doing, and hopefully steer it or stop it in time.<p>As a result, the maximum number of concurrent sessions I could manage at once was only 4. I didn't want to be the bottleneck, so I built Meetless Agent (MLA). It basically does what I had to do manually before:<p>- Monitors the coding agent's tasks and actions to supply it with the correct, up-to-date context.<p>- Continuously reconciles running information (such as provided/tagged documentation, the agent's output, and the agent's decisions) to actively maintain the source of truth at all times.<p>- Keeps track of the current rules for the repo, and if an action triggers a registered rule, the rule is injected into the agent context.<p>My benchmarks show that running coding agents with the help of an active monitor improves quality and accuracy, consumes fewer tokens, and finishes faster: <a href="https://research.meetless.ai/stale-context/" rel="nofollow">https://research.meetless.ai/stale-context/</a><p>Of course, the agent alone can't decide the source of truth; it requires human review and decisions for contradictions, etc. But for the most part, it can safely build a consistent ontology of the current source of truth.<p>From this, I want to build an AI layer to maintain the source of truth across the business, so I will release more connectors for Slack, Jira, Confluence, etc., soon. The goal is for this AI to assist in every part of the business. Eventually, the same coordination layer will understand that a decision made in Slack affects a Jira task, a document, an email conversation, and what a coding agent should do next without every tool becoming another isolated memory silo.<p>The coding agent connector is open source at: <a href="https://github.com/Meetless/mla" rel="nofollow">https://github.com/Meetless/mla</a><p>I am looking forward to your feedback!

  HACKER_NEWS

• Show HN: Groundtrack – Continual learning for coding agents — Hi HN! I built Groundtrack to help coding agents continuously learn and improve from their work across a company.<p>With coding agents handling most of code implementation, a lot of things like review corrections, failed approaches, and discovered constraints stay trapped in individual agent sessions, so future agents can repeat the same failures. Sometimes a broken or agent-unfriendly tool or workflow can cause those issues too; if multiple agents independently find workarounds, those failures don’t get connected and aren’t used to improve the system around the agents.<p>Groundtrack turns those learnings into scoped and shared engineering experiences that can help future agents across teams working with Codex, Claude Code, Cursor, and OpenCode. It retrieves relevant lessons during later work, records whether they helped, and updates or supersedes knowledge as it learns over time.<p>It also identifies recurring friction that shouldn’t just become another stored experience. If agents repeatedly struggle with the same tool or workflow, Groundtrack creates an evidence-backed draft initiative so the team can fix the underlying environment and prevent a certain class of mistake systemically.<p>For example, on the landing page, I follow a database migration incident through the entire loop. Groundtrack turned the original workaround into a scoped experience, surfaced it when a later agent encountered similar migration drift, and recorded that it was successfully applied. After connecting evidence from seven sessions, it also proposed making the safe migration workflow a permanent part of the engineering environment and prevent the issue altogether.<p>There have been many challenges I’ve already run into to make this helpful (and still many unsolved things!), one of which has been generalizing and determining the applicability of one lesson to another session. I wrote about some of my retrieval experiments here: <a href="https://groundtrack.dev/blog/retrieving-knowledge-that-applies/" rel="nofollow">https://groundtrack.dev/blog/retrieving-knowledge-that-appli...</a><p>It’s an early product and I’d really love any and all feedback I can get. If you or your team regularly use coding agents, please try it on an active project and see if it helps your agents learn and improve over time. It’s free for solo use.

  HACKER_NEWS