# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/49848269

# Ollaya – Ollama for open\-source, Jev\-style decision models

579 points · 139 comments

[Full discussion](<https://news.ycombinator.com/item?id=49848269>)

[Read original](<https://ollaya.dev/>)

Category: [Infrastructure & Efficiency](<https://hacksnap.live/?category=infrastructure-efficiency>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Low\. Estimated from 10 comments\.

15 comments for the summary\.

Peak rank: \#2

Time in Top 10: 24\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

217 recorded rank observations from 2026\-09\-26T08:01:15\.746052\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 168 recorded Hacksnap ranks:

2026\-09\-30T18:03:34\.83179\+00:00: rank \#31

2026\-09\-30T19:02:02\.291882\+00:00: rank \#31

2026\-09\-30T20:02:49\.392147\+00:00: rank \#32

2026\-09\-30T21:03:42\.990832\+00:00: rank \#30

2026\-09\-30T22:02:11\.294387\+00:00: rank \#31

2026\-09\-30T23:01:52\.722953\+00:00: rank \#31

2026\-10\-01T08:02:24\.797064\+00:00: rank \#31

2026\-10\-01T09:01:35\.899477\+00:00: rank \#34

2026\-10\-01T10:01:12\.485692\+00:00: rank \#34

2026\-10\-01T11:00:47\.508128\+00:00: rank \#34

2026\-10\-01T12:00:42\.477917\+00:00: rank \#33

2026\-10\-01T13:00:51\.124852\+00:00: rank \#33

2026\-10\-01T14:00:59\.01153\+00:00: rank \#33

2026\-10\-01T15:00:51\.048047\+00:00: rank \#34

2026\-10\-01T16:00:53\.034654\+00:00: rank \#35

2026\-10\-01T16:36:37\.180071\+00:00: rank \#36

2026\-10\-01T17:00:23\.382263\+00:00: rank \#36

2026\-10\-01T18:01:44\.046011\+00:00: rank \#37

2026\-10\-01T19:01:47\.420738\+00:00: rank \#38

2026\-10\-01T20:01:39\.585083\+00:00: rank \#37

2026\-10\-01T21:01:45\.872085\+00:00: rank \#38

2026\-10\-01T22:01:37\.214899\+00:00: rank \#37

2026\-10\-01T23:04:11\.395288\+00:00: rank \#40

2026\-10\-02T08:02:26\.819307\+00:00: rank \#41

2026\-10\-02T09:02:51\.51649\+00:00: rank \#42

2026\-10\-02T10:02:12\.346099\+00:00: rank \#42

2026\-10\-02T11:02:22\.794376\+00:00: rank \#42

2026\-10\-02T12:01:14\.681578\+00:00: rank \#42

2026\-10\-02T13:02:29\.511591\+00:00: rank \#42

2026\-10\-02T14:01:40\.936738\+00:00: rank \#41

2026\-10\-02T15:02:30\.225476\+00:00: rank \#41

2026\-10\-02T16:02:47\.060695\+00:00: rank \#40

2026\-10\-02T17:02:31\.351493\+00:00: rank \#42

2026\-10\-02T18:01:22\.271899\+00:00: rank \#43

2026\-10\-02T19:01:23\.681417\+00:00: rank \#43

2026\-10\-02T20:01:33\.801055\+00:00: rank \#45

2026\-10\-02T21:02:55\.453316\+00:00: rank \#48

2026\-10\-02T22:01:50\.233139\+00:00: rank \#49

2026\-10\-02T23:01:20\.268516\+00:00: rank \#47

2026\-10\-03T08:01:30\.186188\+00:00: rank \#48

2026\-10\-03T09:01:57\.91467\+00:00: rank \#45

2026\-10\-03T10:00:55\.732001\+00:00: rank \#44

2026\-10\-03T11:01:22\.538596\+00:00: rank \#44

2026\-10\-03T12:01:08\.789384\+00:00: rank \#45

2026\-10\-03T13:00:35\.273559\+00:00: rank \#45

2026\-10\-03T14:00:48\.425606\+00:00: rank \#45

2026\-10\-03T15:01:04\.472963\+00:00: rank \#45

2026\-10\-03T16:01:56\.56175\+00:00: rank \#46

2026\-10\-03T17:01:36\.651036\+00:00: rank \#43

2026\-10\-03T18:01:58\.24207\+00:00: rank \#43

2026\-10\-03T19:01:19\.196589\+00:00: rank \#42

2026\-10\-03T20:00:57\.313759\+00:00: rank \#40

2026\-10\-03T21:01:04\.44439\+00:00: rank \#38

2026\-10\-03T21:20:01\.585706\+00:00: rank \#37

2026\-10\-03T22:00:15\.503126\+00:00: rank \#37

2026\-10\-03T23:00:55\.584832\+00:00: rank \#37

2026\-10\-04T08:01:00\.327459\+00:00: rank \#36

2026\-10\-04T09:00:59\.321023\+00:00: rank \#37

2026\-10\-04T10:00:42\.975336\+00:00: rank \#37

2026\-10\-04T11:01:03\.76577\+00:00: rank \#37

2026\-10\-04T12:01:45\.307985\+00:00: rank \#37

2026\-10\-04T13:00:49\.176267\+00:00: rank \#37

2026\-10\-04T14:00:56\.952543\+00:00: rank \#37

2026\-10\-04T15:00:54\.739637\+00:00: rank \#37

2026\-10\-04T16:00:51\.403054\+00:00: rank \#37

2026\-10\-04T17:00:43\.166713\+00:00: rank \#37

2026\-10\-04T18:00:46\.315535\+00:00: rank \#36

2026\-10\-04T19:00:54\.768841\+00:00: rank \#36

2026\-10\-04T20:00:51\.070442\+00:00: rank \#37

2026\-10\-04T21:00:28\.550532\+00:00: rank \#36

2026\-10\-04T22:01:36\.217651\+00:00: rank \#38

2026\-10\-04T23:01:09\.400767\+00:00: rank \#38

2026\-10\-05T08:01:51\.387003\+00:00: rank \#37

2026\-10\-05T09:01:40\.528475\+00:00: rank \#36

2026\-10\-05T10:00:56\.290772\+00:00: rank \#36

2026\-10\-05T11:01:05\.919518\+00:00: rank \#36

2026\-10\-05T12:00:56\.590881\+00:00: rank \#35

2026\-10\-05T13:01:05\.197156\+00:00: rank \#37

2026\-10\-05T14:01:04\.085718\+00:00: rank \#37

2026\-10\-05T15:00:50\.814571\+00:00: rank \#37

2026\-10\-05T16:02:09\.54139\+00:00: rank \#36

2026\-10\-05T17:00:35\.336076\+00:00: rank \#36

2026\-10\-05T18:00:20\.445491\+00:00: rank \#36

2026\-10\-05T19:02:43\.604975\+00:00: rank \#37

2026\-10\-05T20:01:09\.777459\+00:00: rank \#37

2026\-10\-05T21:01:47\.650671\+00:00: rank \#38

2026\-10\-05T22:02:59\.93099\+00:00: rank \#38

2026\-10\-06T08:02:11\.210914\+00:00: rank \#38

2026\-10\-06T08:02:14\.354697\+00:00: rank \#38

2026\-10\-06T09:01:21\.688765\+00:00: rank \#40

2026\-10\-06T10:00:58\.047773\+00:00: rank \#40

2026\-10\-06T11:00:41\.689929\+00:00: rank \#41

2026\-10\-06T12:00:41\.7705\+00:00: rank \#42

2026\-10\-06T13:01:24\.509519\+00:00: rank \#42

2026\-10\-06T14:00:26\.950817\+00:00: rank \#42

2026\-10\-06T15:00:49\.066414\+00:00: rank \#43

2026\-10\-06T16:00:58\.166747\+00:00: rank \#45

2026\-10\-06T17:01:17\.009511\+00:00: rank \#45

2026\-10\-06T18:01:43\.791429\+00:00: rank \#46

2026\-10\-06T19:00:46\.048609\+00:00: rank \#45

2026\-10\-06T20:01:14\.687294\+00:00: rank \#45

2026\-10\-06T21:02:13\.414382\+00:00: rank \#44

2026\-10\-06T22:00:40\.926271\+00:00: rank \#43

2026\-10\-06T23:03:26\.867445\+00:00: rank \#44

2026\-10\-07T08:01:41\.339804\+00:00: rank \#45

2026\-10\-07T09:02:38\.026986\+00:00: rank \#46

2026\-10\-07T10:01:35\.069531\+00:00: rank \#46

2026\-10\-07T11:02:07\.770084\+00:00: rank \#46

2026\-10\-07T12:01:14\.996697\+00:00: rank \#45

2026\-10\-07T13:00:47\.45912\+00:00: rank \#44

2026\-10\-07T14:01:18\.307739\+00:00: rank \#44

2026\-10\-07T15:01:57\.541362\+00:00: rank \#46

2026\-10\-07T16:01:28\.474022\+00:00: rank \#45

2026\-10\-07T17:01:15\.036908\+00:00: rank \#46

2026\-10\-07T18:00:53\.300079\+00:00: rank \#45

2026\-10\-07T19:01:49\.456657\+00:00: rank \#46

2026\-10\-07T20:01:59\.032487\+00:00: rank \#46

2026\-10\-07T21:05:36\.655392\+00:00: rank \#48

2026\-10\-07T22:02:27\.105241\+00:00: rank \#49

2026\-10\-07T23:01:17\.452894\+00:00: rank \#48

2026\-10\-08T08:03:55\.083808\+00:00: rank \#48

2026\-10\-08T09:03:58\.68898\+00:00: rank \#46

2026\-10\-08T10:04:05\.364235\+00:00: rank \#47

2026\-10\-08T11:02:04\.410307\+00:00: rank \#46

2026\-10\-08T12:02:36\.39618\+00:00: rank \#46

2026\-10\-08T13:03:09\.206401\+00:00: rank \#46

2026\-10\-08T14:03:30\.113384\+00:00: rank \#47

2026\-10\-08T15:02:54\.444758\+00:00: rank \#45

2026\-10\-08T16:03:13\.067807\+00:00: rank \#44

2026\-10\-08T17:03:47\.92465\+00:00: rank \#43

2026\-10\-08T18:01:50\.360431\+00:00: rank \#43

2026\-10\-08T19:02:55\.719274\+00:00: rank \#44

2026\-10\-08T20:01:36\.359022\+00:00: rank \#44

2026\-10\-08T21:03:34\.000719\+00:00: rank \#44

2026\-10\-08T22:01:13\.459732\+00:00: rank \#43

2026\-10\-08T23:01:25\.318392\+00:00: rank \#42

2026\-10\-09T08:01:53\.314021\+00:00: rank \#43

2026\-10\-09T09:02:59\.93426\+00:00: rank \#43

2026\-10\-09T10:02:22\.652527\+00:00: rank \#42

2026\-10\-09T11:02:04\.807282\+00:00: rank \#42

2026\-10\-09T12:02:16\.769295\+00:00: rank \#42

2026\-10\-09T13:02:30\.35034\+00:00: rank \#44

2026\-10\-09T14:01:34\.215134\+00:00: rank \#43

2026\-10\-09T15:01:27\.564392\+00:00: rank \#42

2026\-10\-09T16:02:06\.561056\+00:00: rank \#43

2026\-10\-09T17:03:06\.673648\+00:00: rank \#45

2026\-10\-09T18:01:02\.851052\+00:00: rank \#45

2026\-10\-09T19:01:11\.250291\+00:00: rank \#45

2026\-10\-09T20:01:32\.91907\+00:00: rank \#45

2026\-10\-09T21:02:08\.376144\+00:00: rank \#47

2026\-10\-09T22:01:19\.253075\+00:00: rank \#49

2026\-10\-09T23:01:31\.990912\+00:00: rank \#49

2026\-10\-10T08:01:00\.593376\+00:00: rank \#48

2026\-10\-10T09:01:45\.527975\+00:00: rank \#50

2026\-10\-10T10:00:53\.799353\+00:00: rank \#50

2026\-10\-10T11:00:35\.65851\+00:00: rank \#49

2026\-10\-10T12:01:11\.169647\+00:00: rank \#49

2026\-10\-10T13:00:54\.008377\+00:00: rank \#47

2026\-10\-10T14:01:43\.208766\+00:00: rank \#49

2026\-10\-10T15:01:24\.825188\+00:00: rank \#49

2026\-10\-10T16:01:10\.213328\+00:00: rank \#48

2026\-10\-10T17:00:48\.84189\+00:00: rank \#47

2026\-10\-10T18:00:57\.55132\+00:00: rank \#47

2026\-10\-10T19:00:52\.379463\+00:00: rank \#46

2026\-10\-10T20:00:21\.993912\+00:00: rank \#46

2026\-10\-10T21:01:10\.657738\+00:00: rank \#46

2026\-10\-10T22:00:50\.433569\+00:00: rank \#45

2026\-10\-10T23:01:00\.95027\+00:00: rank \#46

Ollaya offers a compelling local, TypeSafe\-compatible deployment story for fast decision models, but the discussion suggests its value depends on whether small open models like Laya can meet quality needs or whether users will fine\-tune them for narrow tasks\.

## The brief

Ollaya is a local runtime and server for open\-weight decision models that answer typed questions about text or JSON in a single forward pass, returning calibrated probabilities rather than generated tokens\. It is positioned as private, open source, and TypeSafe\-compatible, with a model lineup including laya, decider, nli, gliclass, qwen3guard, decision, kev, and von\. The page claims millisecond latency on local NVIDIA GPUs, CPU support everywhere, desktop/CLI/Docker distribution, pinned open weights, and an Apache\-2\.0 runtime\.

- Ollaya runs decision models locally and returns typed answers such as choice, score, and yes/no with calibrated probabilities in a single forward pass\.
- It is drop\-in compatible with TypeSafe's API: it serves /v1/systemone and /v1/models, and the official TypeSafe Python SDK 0\.7\.1 reportedly works unchanged against a local server\.
- The model catalog includes laya as the fastest option, decider as more accurate, nli and gliclass as zero\-shot classifiers, qwen3guard for safety, and decision/kev/von as additional decision\-model variants\.
- The page claims Laya can answer a five\-question request in about 8–10 ms end\-to\-end on an RTX 4090, compared with 236–276 ms median for TypeSafe's hosted Jev API in third\-party benchmarks, while noting the setups differ\.
- Ollaya emphasizes privacy and local deployment: ONNX Runtime on CPU or NVIDIA GPU, server bound to 127\.0\.0\.1 by default, open weights pinned to Hugging Face commits and checked by sha256, and no per\-token fees\.
- Platform support includes desktop apps and command\-line tools for macOS, Windows, and Linux, plus Docker; NVIDIA GPU acceleration is available on Linux, Windows, WSL 2, and Docker, while other GPUs run on CPU\.

## Discussion themes

### The 'decision model' framing drew skepticism

Some commenters argued the example is just text classification, that 'decision model' is marketing jargon, and that text classification is equivalent to what Jev does\.

Sources: [Comment 49848482](<https://news.ycombinator.com/item?id=49848482>) · [Comment 49848695](<https://news.ycombinator.com/item?id=49848695>) · [Comment 49848784](<https://news.ycombinator.com/item?id=49848784>)

### Laya was reported as weaker than Jev

One commenter said Laya performs significantly worse, is less confident, and makes wrong decisions on complex queries\. The developer confirmed Laya is weaker than Jev and described it as a speed trade\-off, while another commenter said Jev generalizes better but Laya can be cheaply fine\-tuned\.

Sources: [Comment 49848537](<https://news.ycombinator.com/item?id=49848537>) · [Comment 49848595](<https://news.ycombinator.com/item?id=49848595>) · [Comment 49848596](<https://news.ycombinator.com/item?id=49848596>)

### Practical utility was questioned, then partly illustrated

A commenter installed Ollaya, found the examples worked, but asked what it is useful for and raised enum\-maintenance concerns\. Another pointed to tools like smart grep and SQL\-based shell history as practical uses, and the original commenter found those examples decent and appreciated the speed/token benefits\.

Sources: [Comment 49850429](<https://news.ycombinator.com/item?id=49850429>) · [Comment 49850583](<https://news.ycombinator.com/item?id=49850583>) · [Comment 49850716](<https://news.ycombinator.com/item?id=49850716>)

### Alternative approach: train a classifier if you have evals

One commenter suggested that with an evaluation set, the smarter move is to train a classifier rather than rely on a generic decision model\.

Sources: [Comment 49848386](<https://news.ycombinator.com/item?id=49848386>)

### Competitive and business\-model concerns

Commenters noted Ollama could add decision\-model support at any time, and that open\-source copies can quickly erode AI startup advantages, forcing TypeSafe to compete on results, UX, support, or custom solutions\.

Sources: [Comment 49848421](<https://news.ycombinator.com/item?id=49848421>) · [Comment 49850052](<https://news.ycombinator.com/item?id=49850052>)

### Technical distinction from rerankers and iterative production path

One commenter did not see the difference between an instruct\-based reranker and laya/jev\. Another explained that laya/jev provide a zero\-shot classifier requiring no training, and outlined an iterative path from LLM prompting to rubric\-based decisions to a custom fine\-tuned classifier\.

Sources: [Comment 49849302](<https://news.ycombinator.com/item?id=49849302>) · [Comment 49849617](<https://news.ycombinator.com/item?id=49849617>)

## Sources & coverage

AI-generated summary · 2026\-09\-26T08:00:34\.469609\+00:00

Based on 15 of 15 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.