frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: I built a graph store so my LLM can't hallucinate facts from my data

https://github.com/TyKolt/kremis
1•TyKolt•1h ago
The problem: you feed your own data to an LLM, ask a question, and get a confident answer — partly fabricated.

Kremis is an embedded graph store (Rust, Apache 2.0) built around one constraint: every answer must trace back to a real data point.

How it works: - You ingest your data as EAV signals (entity, attribute, value) - Kremis stores them in an append-only graph (redb underneath) - Every query returns one label: FACT — path exists in your graph INFERENCE — derived from graph traversal UNKNOWN — "not in your data, I won't guess"

The LLM never writes to the graph. Only your verified signals do. The graph can't hallucinate — it either has the path or it doesn't.

There's a runnable demo (Python stdlib, no pip): examples/demo_honesty.py It ingests 10 signals, asks 6 questions, shows which LLM answers are grounded and which are fabricated.

It also has an HTTP API and MCP support (Claude can query your graph directly via tool calls).

Honest limitation: ingestion is EAV format, so you need to reshape your data first. Working on making this easier.

Comments

TyKolt•1h ago
Happy to answer questions.

The EAV constraint came from a real frustration: I kept getting confident wrong answers from RAG pipelines and had no way to tell which parts were fabricated. I wanted something that would just say "not in your data" instead of guessing.

The biggest open question I'm wrestling with is ingestion friction — EAV is precise but you have to reshape your data first. Curious whether that feels like a dealbreaker to anyone who looks at it, or if it's acceptable given the tradeoff.

My weird ADHD to-do app, Do-dono

https://good.afternoonrobot.co.uk/my-weird-adhd-to-do-app-do-dono/
1•robotmay•32s ago•0 comments

Show HN: Independent monitoring of AI API reliability

https://ai-api-monitor-production.up.railway.app
1•techsearcher•36s ago•0 comments

Rig for Turbulence

https://davegriffith.substack.com/p/rig-for-turbulence
1•dxs•1m ago•0 comments

"Today may be the most important day of the 21st Century so far."

https://www.commentary.org/john-podhoretz/first-thoughts-on-the-new-iran-war/
1•SegfaultSeagull•2m ago•0 comments

BullshitBench v2

https://petergpt.github.io/bullshit-benchmark/viewer/index.v2.html
2•smusamashah•3m ago•0 comments

iPhone-17e

https://www.apple.com/ca/iphone-17e/
1•taesiri•3m ago•0 comments

Show HN: Qrvpn – deploy and run VPN server on any device and behind NAT/FW

https://qrvpn.com/
1•Serusk•3m ago•0 comments

Show HN: Valkey-powered semantic memory for Claude Code sessions

1•kaliades•4m ago•0 comments

Helldivers 2 Adds Millennials as Its Newest Enemy Faction

https://kotaku.com/helldivers-2-cyborg-update-automaton-faction-trailer-2000666908
1•PaulHoule•4m ago•0 comments

Comparing Python packages for A/B test analysis

https://e10v.me/python-packages-for-ab-test-analysis/
1•e10v_me•4m ago•1 comments

The last total lunar eclipse until New Year's Eve 2028-2029

https://www.space.com/stargazing/lunar-eclipses/a-total-lunar-eclipse-will-turn-the-full-moon-blo...
1•limelego•5m ago•0 comments

A Rock Star Philosopher

https://www.nytimes.com/2026/03/01/briefing/a-rock-star-philosopher.html
1•mitchbob•6m ago•1 comments

Type-Safe JDBC: Schema-First Native SQL in Java

https://github.com/manifold-systems/manifold/tree/master/manifold-deps-parent/manifold-sql
2•owlstuffing•6m ago•1 comments

Show HN: Punch card simulator and Fortran IV interpreter

https://punch.ehrlich.dev/
1•behrlich•6m ago•0 comments

Show HN: Workz – run 5 AI agents on parallel Git worktrees with one command

1•rohansx•7m ago•0 comments

We Audited 2,857 Agent Skills. 12% Were Malicious

https://grith.ai/blog/agent-skills-supply-chain
1•edf13•7m ago•0 comments

Bars close and hundreds lose jobs as US firm buys Brewdog in £33M deal

https://www.bbc.com/news/articles/c05v0p1d0peo
2•tartoran•7m ago•0 comments

Iranian strikes test the Gulf's trillion-dollar AI dream

https://restofworld.org/2026/amazon-uae-data-center-fire-iran-strike/
1•donohoe•8m ago•0 comments

Show HN: goilerplate and templUI Pro, two products for Go and templ developers

https://goilerplate.com/
1•axadrn•8m ago•0 comments

Workflows for OpenClaw

https://workflaw.ai
1•l-fy•8m ago•0 comments

From 1Password to Apple Passwords

https://pawelgrzybek.com/from-1password-to-apple-passwords/
2•speckx•8m ago•0 comments

I built a new Terraform agentic editor and auditor

https://grafos.ai
1•souvlakius•8m ago•1 comments

MiniMax M2.5 is beating Claude Opus 4.6 and MiniMax is 17x-20x cheaper

https://www.swebench.com/
3•thelinuxkid•9m ago•3 comments

AI is unhealthy in a variety of different ways

https://www.computerworld.com/article/4138046/people-are-getting-sick-of-ai-literally.html
3•dryadin•9m ago•0 comments

Show HN: Gipity – AI cloud computer in the browser

https://gipity.ai
1•siverson914•10m ago•0 comments

Felix86 26.03: AVX, AVX2, BMI1 and F16C Support

https://felix86.com/felix86-26-03/
1•camel-cdr•13m ago•0 comments

Three US fighter jets shot down in 'friendly fire' in Kuwait

https://metro.co.uk/2026/03/02/three-us-f-15-fighter-jets-shot-kuwait-friendly-fire-incident-2718...
2•harambae•14m ago•0 comments

The Pentagon strongarmed AI firms before Iran strikes

https://theconversation.com/the-pentagon-strongarmed-ai-firms-before-iran-strikes-in-dark-news-fo...
1•geox•14m ago•0 comments

Open-sourced an AI-native habit tracker where the LLM is the interface and coach

https://github.com/ericblue/habit-sprint
1•ericblue•14m ago•1 comments

bpfviewer: A tool for disassembling, debugging and visualizing BPF object files

https://github.com/tsint/bpfviewer
1•tanelpoder•15m ago•0 comments