{"slug": "colibri", "name": "Colibri", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 6, "commenters": 13, "warners": 2, "score": 5.257, "rising": 0.69, "first": "2026-03-26", "latest": "2026-09-02", "why": "Check the Colibri and Flash-MoE projects. They\u2019re already doing that.", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 6, "neutral": 5, "compare": 0, "warn": 2}, "spark": [0, 0, 0, 0, 0, 0, 0, 0, 4, 1, 1, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "Check the Colibri and Flash-MoE projects. They\u2019re already doing that.", "kind": "", "icon": false, "domain": null, "github_repo": "github.com/justvugg/colibri", "official_url": "https://github.com/justvugg/colibri", "pricing": "", "aliases": ["Colibri", "colibri"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 792521344, "item_id": 49532584, "url": "https://news.ycombinator.com/item?id=49532584", "by": "kzrdude", "date": "2026-09-02", "version": null, "stance": "recommend", "quote": "Colibri did that first for GLM-5.2", "excerpt": "Colibri did that first for GLM-5.2 https://github.com/JustVugg/colibri", "context": "", "use_case": "Running large models on modest hardware", "switched_from": "", "story": "Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s"}, {"id": 788735826, "item_id": 49295989, "url": "https://news.ycombinator.com/item?id=49295989", "by": "wren6991", "date": "2026-08-14", "version": null, "stance": "recommend", "quote": "Colibri [2] which have better support for streaming weights from disk", "excerpt": "The thing that blows me away is it does this at one quarter the total parameter count of K3 (and 40% active parameter count). There's plenty of room at the bottom. > How are you all toying with running this kind of thing in a mega quantized way locally? Sure, let me answer that in excessive detail. I briefly tried running the UD IQ3_S quant of GLM-5.2, which is 288 GiB of weights (301 GB). Setup was: llama.cpp, 1x NV", "context": "", "use_case": "", "switched_from": "", "story": "GLM-5.3: Frontier Coding with Emergent Cyber Capabilities"}, {"id": 785672960, "item_id": 49104560, "url": "https://news.ycombinator.com/item?id=49104560", "by": "gitpusher42", "date": "2026-07-29", "version": null, "stance": "recommend", "quote": "Check for colibri, dwarf star and flash-moe.", "excerpt": "Check for colibri, dwarf star and flash-moe. they do similar things with bigger models https://github.com/JustVugg/colibri https://github.com/antirez/ds4 https://github.com/danveloper/flash-moe", "context": "", "use_case": "Large local model inference", "switched_from": "", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}, {"id": 785596768, "item_id": 49099798, "url": "https://news.ycombinator.com/item?id=49099798", "by": "gitpusher42", "date": "2026-07-29", "version": null, "stance": "recommend", "quote": "Check the Colibri and Flash-MoE projects. They\u2019re already doing that.", "excerpt": "Yeah! Check the Colibri and Flash-MoE projects. They\u2019re already doing that. https://github.com/danveloper/flash-moe https://github.com/JustVugg/colibri", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}, {"id": 785069504, "item_id": 49066844, "url": "https://news.ycombinator.com/item?id=49066844", "by": "magicalhippo", "date": "2026-07-27", "version": null, "stance": "recommend", "quote": "Or from SSD using something like Colibri[1]. Not going to be quick, but at least runable.", "excerpt": "> running it on a no GPU, but tons of RAM server Or from SSD using something like Colibri[1]. Not going to be quick, but at least runable. [1]: https://github.com/JustVugg/colibri", "context": "", "use_case": "running large models from SSD", "switched_from": "", "story": "Kimi-K3 Releases on HuggingFace 6/27"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 785068033, "item_id": 49066752, "url": "https://news.ycombinator.com/item?id=49066752", "by": "docheinestages", "date": "2026-07-27", "version": null, "stance": "warn", "quote": "I'm aware of colibri [1], but so far I've only seen extremely slow performance.", "excerpt": "Given the frontier-level capabilities of Kimi K3, I'm wondering if it's possible to extract the core capabilities (fundamental reasoning and tool calling) of the model into a smaller one that consumer devices could run? Not sure exactly how, but either by heavy distillation or some other surgical method since Kimi has a Mixture of Experts architecture. I think it's very valuable to have a smaller model that doesn't h", "context": "", "use_case": "running large models on consumer devices", "switched_from": "", "story": "Kimi-K3 Releases on HuggingFace 6/27"}]}], "who": [], "tags": [], "use_cases": ["running large models on modest hardware", "streaming model weights from ssd", "model quantization tooling", "large local model inference", "running huge models locally", "running large models from ssd"], "compared": [{"slug": "glm", "name": "GLM", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "flash-moe", "name": "flash-moe", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 760551456, "item_id": 47534466, "url": "https://news.ycombinator.com/item?id=47534466", "by": "louisescher", "date": "2026-03-26", "version": null, "stance": "neutral", "quote": "Also, feel free to DM me (@colibri.social) on Bluesky if you want to migrate to the Colibri PDS!", "excerpt": "Hi, I've just added an FAQ entry about this: https://colibri.social/faq#where-is-my-data-stored Also, feel free to DM me (@colibri.social) on Bluesky if you want to migrate to the Colibri PDS! We do host one ourselves.", "context": "", "use_case": "", "switched_from": "", "story": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small"}, {"id": 760551312, "item_id": 47534457, "url": "https://news.ycombinator.com/item?id=47534457", "by": "louisescher", "date": "2026-03-26", "version": null, "stance": "neutral", "quote": "Right now, Colibri is meant to function for communities that are public anyway.", "excerpt": "https://colibri.social/faq#where-is-my-data-stored I've just added a new FAQ entry to explain this in a bit more detail. > But that\u2019s not where you want your chats now is it? E2EE? And how does it keep it all private since apparently the Bluesky bros haven't figured that part out? It honestly depends. Right now, Colibri is meant to function for communities that are public anyway. If you're a streamer, an open source ", "context": "", "use_case": "Community chat on AT Protocol", "switched_from": "", "story": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small"}, {"id": 760541040, "item_id": 47533815, "url": "https://news.ycombinator.com/item?id=47533815", "by": "louisescher", "date": "2026-03-26", "version": null, "stance": "recommend", "quote": "We're doing that to allow you to update your profile from within the app.", "excerpt": "Hi! We're doing that to allow you to update your profile from within the app. Not doing anything else besides that. If you have concerns, take a look at the source code: https://github.com/colibri-social/colibri.social", "context": "", "use_case": "community chat on AT Protocol", "switched_from": "", "story": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small"}, {"id": 745355872, "item_id": 46584742, "url": "https://news.ycombinator.com/item?id=46584742", "by": "9dev", "date": "2026-01-11", "version": null, "stance": "neutral", "quote": "I'm working on a self-hostable ebook library", "excerpt": "I'm working on a self-hostable ebook library (https://github.com/colibri-hq/colibri), and currently tinkering with searching over book content. Have you written about your approach to search somewhere, perhaps? Would be very interested in learning how others go about this. Kudos for DocGlobin, looks great :-)", "context": "", "use_case": "Ebook library", "switched_from": "", "story": "Ask HN: What Are You Working On? (January 2026)"}, {"id": 745354784, "item_id": 46584674, "url": "https://news.ycombinator.com/item?id=46584674", "by": "9dev", "date": "2026-01-11", "version": null, "stance": "recommend", "quote": "Colibri[1], a self-hostable ebook library.", "excerpt": "> We just started working on a full book app that will be a private book diary, along with recommendations and insights based on your Book DNA. Think Goodreads, but rebuilt for readers who want a private space to track what they read, keep notes, and get truly personalized book recommendations (like Pandora or Spotify for books). Oh! This is very similar to the project I am working on, Colibri[1], a self-hostable ebo", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: What Are You Working On? (January 2026)"}, {"id": 745354208, "item_id": 46584638, "url": "https://news.ycombinator.com/item?id=46584638", "by": "9dev", "date": "2026-01-11", "version": null, "stance": "recommend", "quote": "An open source, self-hostable ebook library for yourself", "excerpt": "An open source, self-hostable ebook library for yourself, and/or family and friends. Will include professional and automatically gathered bibliographical metadata, direct reader connectivity, calibre sync, OPDS, an (optional) public bragging shelf, and high quality packages to go with. If you love reading, want to have an easy way to access your collection, and share it with a few people, this is for you. https://git", "context": "", "use_case": "Self-hosted ebook library", "switched_from": "", "story": "Ask HN: What Are You Working On? (January 2026)"}], "threads": [{"url": "https://news.ycombinator.com/item?id=47533283", "title": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=46577242", "title": "Ask HN: What Are You Working On? (January 2026)", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49098510", "title": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49065752", "title": "Kimi-K3 Releases on HuggingFace 6/27", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49524447", "title": "Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49449507", "title": "GLM-5.3-Flash", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {"neutral": 1, "warn": 1}, "2026-04": {}, "2026-05": {}, "2026-06": {}, "2026-07": {"recommend": 4, "neutral": 2, "warn": 1}, "2026-08": {"recommend": 1, "neutral": 2}, "2026-09": {"recommend": 1}, "2026-10": {}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 792521344, "item_id": 49532584, "url": "https://news.ycombinator.com/item?id=49532584", "by": "kzrdude", "date": "2026-09-02", "version": null, "stance": "recommend", "quote": "Colibri did that first for GLM-5.2", "excerpt": "Colibri did that first for GLM-5.2 https://github.com/JustVugg/colibri", "context": "", "use_case": "Running large models on modest hardware", "switched_from": "", "story": "Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s"}, {"id": 791270080, "item_id": 49454380, "url": "https://news.ycombinator.com/item?id=49454380", "by": "yonatan8070", "date": "2026-08-26", "version": null, "stance": "neutral", "quote": "similar to how Colibri works", "excerpt": "Speaking as someone who isn't really well versed in this, does 18B active parameters mean that you could potentially hold only the 18B parameters in RAM and stream the rest from a fast NVMe SSD for acceptable performance similar to how Colibri works? https://github.com/JustVugg/colibri", "context": "", "use_case": "streaming model weights from SSD", "switched_from": "", "story": "GLM-5.3-Flash"}, {"id": 789788912, "item_id": 49361807, "url": "https://news.ycombinator.com/item?id=49361807", "by": "akshay_akula", "date": "2026-08-19", "version": null, "stance": "neutral", "quote": "I wonder how it could work with something like https://github.com/JustVugg/colibri.", "excerpt": "This is interesting. I wonder how it could work with something like https://github.com/JustVugg/colibri.", "context": "", "use_case": "model quantization tooling", "switched_from": "", "story": "Show HN: Shoehorn \u2013 Quantize any model down to run on your machine"}, {"id": 788735826, "item_id": 49295989, "url": "https://news.ycombinator.com/item?id=49295989", "by": "wren6991", "date": "2026-08-14", "version": null, "stance": "recommend", "quote": "Colibri [2] which have better support for streaming weights from disk", "excerpt": "The thing that blows me away is it does this at one quarter the total parameter count of K3 (and 40% active parameter count). There's plenty of room at the bottom. > How are you all toying with running this kind of thing in a mega quantized way locally? Sure, let me answer that in excessive detail. I briefly tried running the UD IQ3_S quant of GLM-5.2, which is 288 GiB of weights (301 GB). Setup was: llama.cpp, 1x NV", "context": "", "use_case": "", "switched_from": "", "story": "GLM-5.3: Frontier Coding with Emergent Cyber Capabilities"}, {"id": 785672960, "item_id": 49104560, "url": "https://news.ycombinator.com/item?id=49104560", "by": "gitpusher42", "date": "2026-07-29", "version": null, "stance": "recommend", "quote": "Check for colibri, dwarf star and flash-moe.", "excerpt": "Check for colibri, dwarf star and flash-moe. they do similar things with bigger models https://github.com/JustVugg/colibri https://github.com/antirez/ds4 https://github.com/danveloper/flash-moe", "context": "", "use_case": "Large local model inference", "switched_from": "", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}, {"id": 785596768, "item_id": 49099798, "url": "https://news.ycombinator.com/item?id=49099798", "by": "gitpusher42", "date": "2026-07-29", "version": null, "stance": "recommend", "quote": "Check the Colibri and Flash-MoE projects. They\u2019re already doing that.", "excerpt": "Yeah! Check the Colibri and Flash-MoE projects. They\u2019re already doing that. https://github.com/danveloper/flash-moe https://github.com/JustVugg/colibri", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}, {"id": 785100272, "item_id": 49068767, "url": "https://news.ycombinator.com/item?id=49068767", "by": "EarthMephit", "date": "2026-07-27", "version": null, "stance": "neutral", "quote": "colibri is meant to allow running GLM 5.2 the 744B model in 32GB", "excerpt": "I haven't tried it, but colibri is meant to allow running GLM 5.2 the 744B model in 32GB by using prefetches and streaming from fast SSDs. https://github.com/JustVugg/colibri", "context": "", "use_case": "running huge models locally", "switched_from": "", "story": "Chinese chipmaker shares surge 470%"}, {"id": 785069504, "item_id": 49066844, "url": "https://news.ycombinator.com/item?id=49066844", "by": "magicalhippo", "date": "2026-07-27", "version": null, "stance": "recommend", "quote": "Or from SSD using something like Colibri[1]. Not going to be quick, but at least runable.", "excerpt": "> running it on a no GPU, but tons of RAM server Or from SSD using something like Colibri[1]. Not going to be quick, but at least runable. [1]: https://github.com/JustVugg/colibri", "context": "", "use_case": "running large models from SSD", "switched_from": "", "story": "Kimi-K3 Releases on HuggingFace 6/27"}, {"id": 785068033, "item_id": 49066752, "url": "https://news.ycombinator.com/item?id=49066752", "by": "docheinestages", "date": "2026-07-27", "version": null, "stance": "warn", "quote": "I'm aware of colibri [1], but so far I've only seen extremely slow performance.", "excerpt": "Given the frontier-level capabilities of Kimi K3, I'm wondering if it's possible to extract the core capabilities (fundamental reasoning and tool calling) of the model into a smaller one that consumer devices could run? Not sure exactly how, but either by heavy distillation or some other surgical method since Kimi has a Mixture of Experts architecture. I think it's very valuable to have a smaller model that doesn't h", "context": "", "use_case": "running large models on consumer devices", "switched_from": "", "story": "Kimi-K3 Releases on HuggingFace 6/27"}, {"id": 784860513, "item_id": 49053782, "url": "https://news.ycombinator.com/item?id=49053782", "by": "AussieWog93", "date": "2026-07-25", "version": null, "stance": "neutral", "quote": "Someone did this exact thing recently", "excerpt": "Someone did this exact thing recently, but running GLM-5.2 with something like 16GB of DRAM, a standard desktop CPU and nVME SSD. I think they got something like 10 _seconds per token_ (not tokens per second). EDIT: it was 25GB of ram and up to 20 seconds per token! https://github.com/JustVugg/colibri", "context": "", "use_case": "Local LLM inference", "switched_from": "", "story": "Running a 28.9M parameter LLM on an $8 microcontroller"}, {"id": 783792160, "item_id": 48987010, "url": "https://news.ycombinator.com/item?id=48987010", "by": "nekusar", "date": "2026-07-20", "version": null, "stance": "recommend", "quote": "Already is possible.", "excerpt": "Already is possible. On a machine with 32GB ram, and NO gpu. Just need a large SSD or NVME. Streams from disk to memory. https://github.com/JustVugg/colibri", "context": "", "use_case": "Running large models without GPU", "switched_from": "", "story": "Who's Afraid of Chinese Models?"}, {"id": 782811296, "item_id": 48925706, "url": "https://news.ycombinator.com/item?id=48925706", "by": "Aurornis", "date": "2026-07-15", "version": null, "stance": "recommend", "quote": "you should check out the project this was based on: https://github.com/JustVugg/colibri", "excerpt": "If you\u2019re interested in these projects you should check out the project this was based on: https://github.com/JustVugg/colibri It says so right in the readme. They\u2019re not hiding anything.", "context": "", "use_case": "running LLMs on CPU", "switched_from": "", "story": "Running Gemma 4 26B at 5 tokens/SEC on a 13-year-old Xeon with no GPU"}, {"id": 760658785, "item_id": 47541174, "url": "https://news.ycombinator.com/item?id=47541174", "by": "david_shi", "date": "2026-03-27", "version": null, "stance": "warn", "quote": "", "excerpt": "The first assumption has been long disproved since multiple full scale Discord data leaks. If it's a public server, it can be scraped. https://www.malwarebytes.com/blog/news/2024/04/billions-of-scraped-discord-messages-up-for-sale", "context": "", "use_case": "", "switched_from": "", "story": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small"}, {"id": 760622832, "item_id": 47538927, "url": "https://news.ycombinator.com/item?id=47538927", "by": "righthand", "date": "2026-03-26", "version": null, "stance": "neutral", "quote": "This implementation is almost entirely vibe-coded", "excerpt": "> This implementation is almost entirely vibe-coded for the purpose of being able to quickly get started with development of the main application. It will be re-written in the near future to take advantage of Tap and be reworked to include all user data storage as well as any OAuth capabilities, which currently reside within the website's backend. If you are interested in helping with this, start a discussion on this", "context": "", "use_case": "Chat platform on AT Protocol", "switched_from": "", "story": "Colibri \u2013 chat platform built on the AT Protocol for communities big and small"}]}