Building a full agentic harness around a 4B model is hard
Around 3 months ago, we were thinking why none of the iPhone apps running an LLM are built as a full harness (as in inference + agentic loop + context management + tools + MCP servers and etc.). It became more interesting when we noticed even the new Siri is not fully on device (and not available in EU for that matter). Having built a few agentic products around a custom harness in the past, we thought this shouldn't be that hard. well, we underestimated how "dumb" a 4B model can be, especially when it comes to tool calling. :D We tried 8 different models and we settled on Qwen 3.5 4B and we…
What it does
Orvena is a complete assistant that runs on your phone. It moves meetings, finds places, reads photos, and talks in a natural voice in 31 languages. No account. No Orvena server.
Orvena moves meetings, looks things up on the schedule you set, finds places nearby, reads the photo you attach, and talks with you in a natural voice, in 31 languages. The language model runs on your phone. No account. No Orvena server. There are apps that run a language model on your phone, and apps that give it a few tools . Orvena works with the phone you already have: calendar, reminders, alarms, maps, photos, health, music, files. Every action leaves a receipt you can read, and anything consequential asks before it runs. Ask tonight, and Orvena checks at the right time. If the app is open, it simply runs. If the phone is in your pocket, a notification arrives; one tap sets it working,…from orvena.app
In the maker’s words, at launch
Around 3 months ago, we were thinking why none of the iPhone apps running an LLM are built as a full harness (as in inference + agentic loop + context management + tools + MCP servers and etc.). It became more interesting when we noticed even the new Siri is not fully on device (and not available in EU for that matter). Having built a few agentic products around a custom harness in the past, we thought this shouldn't be that hard. well, we underestimated how "dumb" a 4B model can be, especially when it comes to tool calling. :D We tried 8 different models and we settled on Qwen 3.5 4B and we used every trick we knew to make this model behave. well, it works! It's not gonna win in any intelligence or speed benchmark, but it can do actual useful work and it really is an on device, private, full agentic harness. That said, you can still connect your OpenRouter or OpenAI API keys if that's what you prefer. Please go check it out :) You need an iPhone 15 pro or above.
Does the same job
all alternatives →- IBI built an open source multi-agent harness in GoApr 2026 · github.com · ▲6
Hey HN. I built an AI agent harness over the past few months and I'm open sourcing it today. Some context on why. I've been building with Claude Code daily using this harness. It orchestrates multiple AI agents as a team, with a dashboard, chat, kanban board, the works. I used it to build a full SaaS product (MyUpMonitor, https://myupmonitor.com) in about 24 hours of focused coding. Then yesterday Anthropic announced Mythos and decided to keep it behind closed doors. Meanwhile I'm paying for Claude and I can't access their best model. I don't think that is nice at all... So I'm…
- ARAgents, run any coding agent on your subscription not API costsMay 2026 · agents-cli.sh · ▲6
Hi HN. I'm the founder of Phoenix Labs (ex TikTok, Applied AI) and we're open sourcing our internal tooling today which is like a toolchain / meta-harness for CLI agents useful for really scaling eng and creative work. We are a very small team who's building a very ambitious product so we had to find ways to squeeze every ounce of efficiency that we could get our hands on. Harness strengths of different models (Claude, GPTs) and CLI-harnesses (Claude Code, Codex), safe/robust browser integration to speed up UX/QA testing, teams cli to speed up security reviews and parallelize…
- PFPlaid for AI Agents2025 · pipedream.com · ▲6
PD founder here. Building integrations with all of your customer's systems is a core challenge for every company building agentic AI solutions. Connect is the easiest way for your users to connect to over 2,400+ APIs and get access to 10,000+ tools, right in your product or AI agent. You can build in-app messaging, CRM syncs, vertical agents, and much more, all in a few minutes. Demo app - https://pipedream-connect-demo.vercel.app/ Quickstart - https://pipedream.com/docs/connect/quickstart You have full, code-level control over how these integrations…
- AGAiaiai.guide: Plain-English mental model for LLM apps, tools and agentsApr 2026 · aiaiai.guide · ▲7
I’ve been building LLM tooling for a small VC fund and found myself explaining the same mental model over and over to non-technical people around me: how a stateless LLM becomes a chatbot, how tool use works, what an agent is mechanically, and why context windows shape all of it. I never found a guide that covered that full chain at the level I wanted, so I wrote one. It’s nine short chapters, each building on the last. Deliberately simplified: the goal is a useful mental model, not a textbook. Feedback, corrections, and contributions welcome: github.com/ymyke/aiaiai
Shoehorn, a library to quantize an LLM to fit your Mac's VRAM23d ago · github.com · ▲6I made this after seeing someone posit the idea online yesterday over lunch then spent some time refining it. So far it's pretty impressive IMO! Right now I am running Qwen3-30B-A3B on my 24gb unified memory m4 MacBook Pro at 50 tok/sec and this should definitely not be working for such a large model on my middling hardware. Things are detailed in the README to get up and running and DESIGN.md has details on all the choices and such made along the way.
- VPVAEN – Package and import portable AI coding-agent HarnessesMay 2026 · github.com · ▲8
Hi HN, I built VAEN (an open source CLI) because I kept running into a boring problem with AI coding-agent workflows: the setup becomes useful, but then it is hard to move. A good, useful agentic harness consists of more than just instructions: skills, mcp servers and more. There should be a better way to share those than just .MD files, and that is why I created VAEN. What works: create a yaml, run the CLI commands as per the repo, and get a .agent file that you can share and extract. Think of how awesome it could be if anyone could use a very useful agentic harness, and share it with one…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, August 2026
the whole month →- TL
Life & fun · 10d ago · louisabraham.github.io
- SA
Hello HN! I found that picking out plausible but diverse skin tones for my digital art and game development projects was kind of difficult, and I got curious about if there was a way to define a color space that made it easy. I've built a color picker and procedural generation algorithm based on the space as well as a bunch of other fun js features and demos throughout the page that use the equations. If you find it interesting, I have lots of explanations of how I built it and what properties the space has. The methodology might be a bit shaky, but hopefully the result is as helpful for…
Life & fun · Aug 2026 · toneyalexander.github.io
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com