Rayline routes Claude Code subagents to on-device and cheaper models
Hi HN, I’m one of the builders of Rayline. Rayline is a Claude Code compatible LLM gateway. It intercepts and overrides claude code’s internal routing and lets you route subagent calls to different models instead. For example, you can run the main agent on Opus, some subagents on cloud-hosted open models, and other subagents on-device. We’ve seen others implement routing for claude code as tools the agent can invoke. In our experience, that doesn’t work well because it requires the main agent to use tokens to think about + call the tools, and LLMs are generally a very inefficient way to make routing decisions. By implementing Rayline as a gateway, we let users deterministically configure routing decisions, and you can optionally use our ML model to make routing decisions. We built it after noticing that Claude Code sessions contain a lot of subagent calls that don’t all need the same model. Other routers exist, but we built Rayline to let us continue using claude code (no separate harness), route tasks at a subagent level, and route across cloud and on-device. The main agent often benefits from Opus. But many delegated calls have narrow scope: search the repo, summarize context, inspect an error, poll for CI updates, etc. The thing we’re exploring is subagent-level routing. The main cost lever in coding agents is usually cached vs non-cached input. Subagent delegations are a natural point to make routing decisions because you avoid busting cache. We look at the message-thread context for a delegated call and choose a model for that call. At a task level, Sonnet and Haiku are almost always less capability-per-dollar than open models, so the main advantage is better + (much) cheaper subagents (60-90% in our private beta). The whole world seems to have started talking about model routing in the past two weeks, so apparently others agree it’s a relevant product area. We’d love to get feedback from the HN community!
AI Analysis
Analysis coming soon.
Similar Products
X402vps
Show HN: X402vps – Docker containers for AI agents, paid per hour with USDC
What's AI's go-to, public or private healthcare?
Show HN: What's AI's go-to, public or private healthcare?
Bento
Over the past few months, our team has been building more and more slidedecks using web frontend technologies with coding harnesses like Claude Code, but a common complaint is to make even small edits we need to edit the code either manually or via the harness. To avoid this loop, I ended up creating Bento, a single HTML file with everything you need in a slide tool including animations and shared editing. There's no install or cloud login, everything works offline. The default deck is around 560 KB and it doesn't need to fetch anything once you got it. Open it in a browser and then you can edit, present, print and save. Share it via email or via Airdrop and all they need is a browser to edit, present and also do live collab on the slides. Drop it in to Claude or ChatGPT to transform existing pptx files into Bento slides. There is no cloud involved, only an encrypted blind relay to allow for shared editing. The relay doesn't see any of the data. Check it out at https://bento.page/slides/ which takes you straight to the editor. Go to https://bento.page/guestbook/ to try out the live guestbook to experience share editing / collab. There is also a gallery with some sample decks on the website - https://bento.page/ All the code is MIT licensed and you can find it here - https://github.com/nyblnet/bento . I used reveal.js with several other libraries (including some homegrown ones), and Claude Code.
threadfork
Hi HN, I built threadfork. It's an AI notetaker where the whole pipeline runs locally on Apple Silicon. Your microphone and system audio gets captured without a meeting bot and most importantly your meeting data doesn’t get sent or saved anywhere. I've been building in the privacy + local LLM space for the past several years (I started DataFog, an OSS PII detection library with 300K+ downloads); keeping all the processing and data on-device was a day 0 condition for me when building this out. Granola's updated privacy policy, effective later this week, keeps non-enterprise users opted into model training on "de-identified" data by default. Having built PII detection for a living, I'll say plainly: de-identifying meeting transcripts is really hard. Names are the easy part; deal terms, client details, and health disclosures are the transcript. For the lawyers, consultants, and therapists I've been talking to, the only version of this that holds up is the model never seeing the data at all. It's macOS 14+ on Apple M-Series computers (for now), 14-day trial, no credit card. I have a user running this on an M1 8GB so older machines welcome. $40/month after that. I use this product daily and you can expect regular updates. Welcome any/all feedback.
Newrepublic
FBI Forced to Reveal New Details on How It Redacted Epstein Files