The archive · Developer & Business Tools · Product decision · 2023–2026
Reor bets local AI notes beat cloud; 411-pt Show HN, 8.5k stars, archived
Solo UK dev Sam L'Huillier builds a local-first AI note app that runs models on-device; 411-point Show HN and 8.5k stars, then the repo is archived in 2026.
Reor
What the business is
Reor is an open-source (AGPL-3.0) desktop note-taking app that runs LLMs and embeddings locally via Ollama, auto-links related notes by vector similarity, and answers questions over an Obsidian-style markdown vault.
How it started
Sam L'Huillier, an AI engineer at the UK Ministry of Justice, started Reor in late 2023 (first commit 2023-10-31) from the belief that AI tools for thought should run locally by default, targeting lawyers, doctors, and researchers who would never send private notes to an embedding API.
What happened
The Show HN on 2024-02-14 drew 411 points and 102 comments, largely debating whether AI-assisted organization weakens thinking; development continued through 2024–2025 with releases up to v0.2.31 and the repo peaking around 8.5k stars with 34 contributors, over 60% of commits from the solo maintainer.
How it ended up
The repository was archived by the owner on 2026-03-07 and is now read-only; the last commit was 2025-05-13, so the project wound down rather than being acquired or shut down, with no funding signal ever recorded.
Background
Reor was a bet that AI note-taking should run entirely on the user's machine. The open-source desktop app used Ollama for local LLMs and embeddings, indexed an Obsidian-style markdown vault into a local vector database, auto-linked related notes by semantic similarity, and answered questions over the corpus with RAG — no cloud service ever touched the notes.
The design treated human writing and LLM retrieval as two generators sharing one vector store: the editor showed related notes retrieved from the corpus while Q&A ran over the same index. It shipped with a strict AGPL-3.0 license, a deliberate no-cloud stance, and engineering choices like an Ollama three-tier fallback and LanceDB schema self-healing.
Traction was real but peaked quickly: the Show HN on 2024-02-14 drew 411 points and 102 comments, and the repo reached about 8.5k stars with 34 contributors. Development slowed through 2025 — last commit 2025-05-13 — and the owner archived the repository on 2026-03-07, leaving the local-first AI notes niche to AnythingLLM, Khoj, and Logseq plugins.
What has to be true
- Privacy was the sharpest wedge: lawyers, doctors, and researchers would not send private notes to embedding APIs, and Reor made local the default.
- The dual-generator concept — AI as an amplifier of writing, not a replacement — gave the project a clear, defensible story.
- Open-source under AGPL-3.0 made the privacy promise auditable and matched the audience's values.
- The 411-point Show HN and 8.5k stars confirmed real demand, but the solo maintainer's capacity capped the roadmap.
What can be applied
A solo project can prove a paradigm and still die: local-first AI notes were the right bet with real traction, but one maintainer could not ship the v1.0 roadmap alone.
Aftermath
As of 2026-09-02 Reor is archived and read-only on GitHub (owner action on 2026-03-07, last commit 2025-05-13); Tencent Cloud's June 2026 profile records 8,563 stars and 525 forks, notes zero commits in the prior 365 days, and recommends Khoj, Logseq AI plugins, or AnythingLLM Desktop as living alternatives. The domain reorproject.org has dropped its old docs and the project has no maintainer.
Sources
- Show HN: Reor – An AI note-taking app that runs models locally
- reorproject/reor — Private & local AI personal knowledge management app
- 18 个月、35k 行单兵干出 8.5K stars 的本地 AI 笔记：Reor 把 RAG 装进 Markdown 后归档了
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card