New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Pangram: The Most Reliable AI Detector Has a Human Problem(bloomberg.com ↗)
    discuss
  2. Meta to report child abuse material directly to Indian authorities(bbc.com ↗)
    discuss
  3. In Germany's east, old ties to Moscow fuel AfD rise(ft.com ↗)
    discuss
  4. Insights on AI from the Reformation(amytis.io ↗)
    discuss
  5. Stop Scanning Blindly: The Beginner's Guide to Stealthy Network Recon(medium.com/meetcyber ↗)
    discuss
  6. Don't Look Back at the Reference Too Often(justinmath.com ↗)
    discuss
  7. Misconceptions(arcgis.com ↗)
    discuss
  8. Should We Reverse Corporate Tax Cuts?(nominalnews.com ↗)
    discuss
  9. Show HN: Social Reps – Earn Your Scroll(socialreps.app ↗)
    discuss
  10. The Bannon-Bernie Bromance over AI and Oligarchs(theatlantic.com ↗)
    discuss
  11. Knowledge, Understanding, and Proofs(finiteape.com ↗)
    discuss
  12. Annie Dillard, Meditative Author of 'Pilgrim at Tinker Creek,' Dies at 81(nytimes.com ↗)
    1comments
  13. Microsoft says AI rival Anthropic could have 'disastrous impact' on humanity(bbc.co.uk ↗)
    4comments
  14. The DeepMind Institute(deepmind.com ↗)
    discuss
  15. Donations at Work: One Year of Sponsored Servo Development(servo.org ↗)
    discuss
  16. Delta – A Multiplayer Environment for Coding with Agents(delta.dev ↗)
    discuss
  17. Show HN: ManyBot – Framework to build WhatsApp bots, without the boring part(manybot.org ↗)
    discuss
  18. System One Lite – typed decisions from a local LLM, with no generated tokens(github.com/snellingio ↗)
    discuss
  19. SnagFlat – Rent-stabilization badges on every NYC StreetEasy listing(snagflat.com ↗)
    discuss
  20. (Re)Introducing Developer Story(stackoverflow.blog ↗)
    discuss
  21. Space Force gives SpaceX latitude on design of $2.3B satellite network(spacenews.com ↗)
    discuss
  22. PS5 Linux lead quits: "a bunch of noobs using LLMs" that "they don't understand"(frvr.com ↗)
    1comments
  23. European Union invites Canada to become first ever 'associate member'(theguardian.com ↗)
    discuss
  24. Anatomy of a Texture(agentlien.github.io ↗)
    1comments
  25. Backing up Hetzner snapshots locally(enjoys.it ↗)
    discuss
  26. A warning about 'model welfare'(mustafa-suleyman.ai ↗)
    discuss
  27. Show HN: OmnisBench, a re-gradable, open LLM routing benchmark on fresh tasks(github.com/fortitude-group ↗)
    discuss
  28. Fourteen studies and every walk-forward fold validated a hindsight stop(honestbook.io ↗)
    discuss
  29. Better Icon and Label Alignment(ishadeed.com ↗)
    discuss
  30. Show HN: Aplexica – resume a Claude Code conversation in Codex, and back(github.com/aplexica ↗)
    discuss

Show HN: A Stateful LLM API that works across providers

1 pointsby 52m agotwigg.ai
0 comments
For the last year we've been building AI workspaces. With every new product, we found ourselves rebuilding the same context layer and infrastructure. We tried existing solutions, but we'd either have to lock into one provider or have to manage the storage infrastructure.

What we wanted was an API that's stateful and isn't tied to a specific provider. So that we didn't have to think about compaction, differing schemas, or running the infrastructure ourselves.

Since we'd already built most of the constituent parts, we’ve decided to release Twigg.

The API is deliberately simple. You create a chat and get back an ID. When you have a new prompt or tool result, you send that plus the ID. There's no need to store or manage context yourself. It appends the new prompt and assembles the history for you. To switch models, you change one field in the next request.

The dashboard covers most of what you'd want to configure: tool schemas, system prompts, context window limits and retention settings.

We'd like to hear from anyone who has run into the same problems. How are you hosting your context right now, and what would make you not want to hand it to a third party?

A quiet thread, for now.Start the conversation on HN ↗