{"slug": "crush", "name": "Crush", "category": "ai-coding", "category_name": "AI coding tools", "description": "Terminal-based AI coding agent from Charmbracelet, written in Go, supporting many LLM providers.", "recommenders": 9, "commenters": 11, "warners": 0, "score": 8.964, "rising": 2.04, "first": "2026-04-09", "latest": "2026-09-29", "why": "HN commenters like Crush for its clean, fast terminal UI and single Go binary, and for running open or local models across swappable providers.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 10, "neutral": 3, "compare": 0, "warn": 0}, "spark": [0, 0, 0, 0, 0, 1, 0, 3, 1, 4, 1, 0], "listed": true, "loved": ["Open models", "Provider flexibility"], "caveat": "Defaults", "short_quote": "Crush is pretty nice. It\u2019s become my daily driver for personal projects.", "icon": true, "domain": null, "github_repo": "github.com/charmbracelet/crush", "official_url": "https://github.com/charmbracelet/crush", "pricing": "free", "aliases": ["Crush"], "verdict": [{"text": "HN commenters like Crush for its clean, fast terminal UI and single Go binary,", "cites": [{"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 788904688, "item_id": 49306543, "url": "https://news.ycombinator.com/item?id=49306543", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about", "excerpt": "Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash) Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on. But I will bite. What does pi do today ", "context": "works on bootstrapping, compilers, kernel debugging", "use_case": "OS bootstrapping and kernel debugging", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788744897, "item_id": 49296556, "url": "https://news.ycombinator.com/item?id=49296556", "by": "brhenc", "date": "2026-08-14", "stance": "recommend", "quote": "As someone who lives in the terminal the web interface doesn't feel as nice to use as things like aichat, crush, or antigravity-cli.", "excerpt": "Tried it this morning - it works with openrouter, feels a little less verbose than antigravity CLI which should help with cost. I would be happier if it wasn't javascript-based, but it's open source so it should be doable to use the models with things like https://github.com/charmbracelet/crush and optimize a little how these alternatives work with deepseek models. For prototyping the web interface is OK, but I don't", "context": "lives in the terminal", "use_case": "terminal coding harness", "switched_from": "", "story": "DeepSeek Harness official website launched"}]}, {"text": " and for running open or local models across swappable providers.", "cites": [{"id": 766287793, "item_id": 47892987, "url": "https://news.ycombinator.com/item?id=47892987", "by": "nickdothutton", "date": "2026-04-24", "stance": "recommend", "quote": "Having some success with Qwen+Crush and have been more productive.", "excerpt": "Switched to local models after quality dropped off a cliff and token consumption seemed to double. Having some success with Qwen+Crush and have been more productive.", "context": "", "use_case": "local coding agent", "switched_from": "Claude", "story": "I Cancelled Claude: Token Issues, Declining Quality, and Poor Support"}, {"id": 788811153, "item_id": 49300697, "url": "https://news.ycombinator.com/item?id=49300697", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Use 3.6 27b as a daily driver for months with charmbracelet crush.", "excerpt": "Use 3.6 27b as a daily driver for months with charmbracelet crush. Gemma 26b-A3b is not even remotely comparable in terms of coding for me. YMMV depending on how you work, what harness you use, etc I suppose.", "context": "", "use_case": "coding harness", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 776357041, "item_id": 48522315, "url": "https://news.ycombinator.com/item?id=48522315", "by": "saratogacx", "date": "2026-06-13", "stance": "recommend", "quote": "I'm using crush and have had a good experience", "excerpt": "I've been using GLM-5/5.1 for about 6 months and it has been a fairly capable model. I've seen a lot of mixed opinions that tend to align with harness usage so it is worth trying out a couple with a model before writing it off. For example, I'm using crush and have had a good experience while others using CC have had a much more mixed experience. For task complexity, I treat it as I would sonnet with the same care in", "context": "", "use_case": "coding harness", "switched_from": "", "story": "GLM 5.2 Is Out"}, {"id": 783809106, "item_id": 48988069, "url": "https://news.ycombinator.com/item?id=48988069", "by": "ricardobeat", "date": "2026-07-20", "stance": "recommend", "quote": "Really easy to do with Crush/OpenCode.", "excerpt": "I just did this five minutes ago. Started a task with Kimi K3, noticed cost going up, switched to Minimax M3 continuing from the same context. Really easy to do with Crush/OpenCode. It works quite well most of the time. For very deeply technical tasks, where exploration involves multiple agents (and blowing up context limits a few times), I'll have a plan written down to disk that will be much longer than 2k tokens.", "context": "", "use_case": "switching models mid-task", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 789496898, "item_id": 49343556, "url": "https://news.ycombinator.com/item?id=49343556", "by": "dongkeren", "date": "2026-08-18", "stance": "recommend", "quote": "you can also try different harness such as Crush and Pi", "excerpt": "You'd better take harness and provider as different parts. In your case, I think the major issue is OpenCode go provier, not necessariy OpenCode harness. If you just want to have an alternative provider, I would recommend OpenRouter, which has provider fallback and model fallback. You can try OpenCode + OpenRouter instead of OpenCode + OpenCode Go; Or you can also try different harness such as Crush and Pi, in this w", "context": "", "use_case": "agent harness", "switched_from": "", "story": "Ask HN: What's the Best Agent Harness?"}]}, {"text": " The most cited caveats are defaults that some users patch out, like command blocking and mouse grabbing,", "cites": [{"id": 798377856, "item_id": 49898616, "url": "https://news.ycombinator.com/item?id=49898616", "by": "ThomasGlanzmann", "date": "2026-09-29", "stance": "recommend", "quote": "I use crush (https://github.com/charmbracelet/crush) with the following patches:", "excerpt": "I use crush (https://github.com/charmbracelet/crush) with the following patches: curl https://tg.st/u/0001-fix-unblock-all-commands-in-bash-tool.patch | git am curl https://tg.st/u/0002-feat-add-light-theme-with-auto-detection-for-white-b.patch | git am curl https://tg.st/u/0003-feat-enable-yolo-mode-by-default.patch | git am curl https://tg.st/u/0004-fix-disable-mouse-grabbing-to-restore-native-termina.patch | git a", "context": "", "use_case": "agent harness", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 787834992, "item_id": 49239687, "url": "https://news.ycombinator.com/item?id=49239687", "by": "ThomasGlanzmann", "date": "2026-08-09", "stance": "recommend", "quote": "I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops.", "excerpt": "I do the same. I also use crush/deepseek a lot. I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops. For example to configure netapp storage systems and had not a single incident. Of course I also have backups of my systems and have everything under revision control. Depending on the task I let it run on a throw away VM, production system, lab system.", "context": "devops, runs agents on throwaway VMs with backups", "use_case": "devops and agentic tasks", "switched_from": "", "story": "Auto mode is now the default in Claude Code"}]}, {"text": " and subagents that cannot use edit tools.", "cites": [{"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}]}], "praise": [{"label": "Open models", "summary": "Commenters report good results pairing it with open or local models such as Qwen, GLM and DeepSeek.", "quotes": [{"id": 788811153, "item_id": 49300697, "url": "https://news.ycombinator.com/item?id=49300697", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Use 3.6 27b as a daily driver for months with charmbracelet crush.", "excerpt": "Use 3.6 27b as a daily driver for months with charmbracelet crush. Gemma 26b-A3b is not even remotely comparable in terms of coding for me. YMMV depending on how you work, what harness you use, etc I suppose.", "context": "", "use_case": "coding harness", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 766287793, "item_id": 47892987, "url": "https://news.ycombinator.com/item?id=47892987", "by": "nickdothutton", "date": "2026-04-24", "stance": "recommend", "quote": "Having some success with Qwen+Crush and have been more productive.", "excerpt": "Switched to local models after quality dropped off a cliff and token consumption seemed to double. Having some success with Qwen+Crush and have been more productive.", "context": "", "use_case": "local coding agent", "switched_from": "Claude", "story": "I Cancelled Claude: Token Issues, Declining Quality, and Poor Support"}, {"id": 776357041, "item_id": 48522315, "url": "https://news.ycombinator.com/item?id=48522315", "by": "saratogacx", "date": "2026-06-13", "stance": "recommend", "quote": "I'm using crush and have had a good experience", "excerpt": "I've been using GLM-5/5.1 for about 6 months and it has been a fairly capable model. I've seen a lot of mixed opinions that tend to align with harness usage so it is worth trying out a couple with a model before writing it off. For example, I'm using crush and have had a good experience while others using CC have had a much more mixed experience. For task complexity, I treat it as I would sonnet with the same care in", "context": "", "use_case": "coding harness", "switched_from": "", "story": "GLM 5.2 Is Out"}]}, {"label": "Provider flexibility", "summary": "Easy to switch models or providers mid-task and combine with services like OpenRouter.", "quotes": [{"id": 783809106, "item_id": 48988069, "url": "https://news.ycombinator.com/item?id=48988069", "by": "ricardobeat", "date": "2026-07-20", "stance": "recommend", "quote": "Really easy to do with Crush/OpenCode.", "excerpt": "I just did this five minutes ago. Started a task with Kimi K3, noticed cost going up, switched to Minimax M3 continuing from the same context. Really easy to do with Crush/OpenCode. It works quite well most of the time. For very deeply technical tasks, where exploration involves multiple agents (and blowing up context limits a few times), I'll have a plan written down to disk that will be much longer than 2k tokens.", "context": "", "use_case": "switching models mid-task", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 789496898, "item_id": 49343556, "url": "https://news.ycombinator.com/item?id=49343556", "by": "dongkeren", "date": "2026-08-18", "stance": "recommend", "quote": "you can also try different harness such as Crush and Pi", "excerpt": "You'd better take harness and provider as different parts. In your case, I think the major issue is OpenCode go provier, not necessariy OpenCode harness. If you just want to have an alternative provider, I would recommend OpenRouter, which has provider fallback and model fallback. You can try OpenCode + OpenRouter instead of OpenCode + OpenCode Go; Or you can also try different harness such as Crush and Pi, in this w", "context": "", "use_case": "agent harness", "switched_from": "", "story": "Ask HN: What's the Best Agent Harness?"}, {"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}]}, {"label": "Terminal UI", "summary": "Praised as the most user-friendly and prettiest terminal interface, with a usable scroll view.", "quotes": [{"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 788904688, "item_id": 49306543, "url": "https://news.ycombinator.com/item?id=49306543", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about", "excerpt": "Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash) Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on. But I will bite. What does pi do today ", "context": "works on bootstrapping, compilers, kernel debugging", "use_case": "OS bootstrapping and kernel debugging", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788744897, "item_id": 49296556, "url": "https://news.ycombinator.com/item?id=49296556", "by": "brhenc", "date": "2026-08-14", "stance": "recommend", "quote": "As someone who lives in the terminal the web interface doesn't feel as nice to use as things like aichat, crush, or antigravity-cli.", "excerpt": "Tried it this morning - it works with openrouter, feels a little less verbose than antigravity CLI which should help with cost. I would be happier if it wasn't javascript-based, but it's open source so it should be doable to use the models with things like https://github.com/charmbracelet/crush and optimize a little how these alternatives work with deepseek models. For prototyping the web interface is OK, but I don't", "context": "lives in the terminal", "use_case": "terminal coding harness", "switched_from": "", "story": "DeepSeek Harness official website launched"}]}, {"label": "Daily driver", "summary": "Used daily for demanding work including devops, compilers and kernel debugging.", "quotes": [{"id": 787834992, "item_id": 49239687, "url": "https://news.ycombinator.com/item?id=49239687", "by": "ThomasGlanzmann", "date": "2026-08-09", "stance": "recommend", "quote": "I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops.", "excerpt": "I do the same. I also use crush/deepseek a lot. I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops. For example to configure netapp storage systems and had not a single incident. Of course I also have backups of my systems and have everything under revision control. Depending on the task I let it run on a throw away VM, production system, lab system.", "context": "devops, runs agents on throwaway VMs with backups", "use_case": "devops and agentic tasks", "switched_from": "", "story": "Auto mode is now the default in Claude Code"}, {"id": 788904688, "item_id": 49306543, "url": "https://news.ycombinator.com/item?id=49306543", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about", "excerpt": "Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash) Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on. But I will bite. What does pi do today ", "context": "works on bootstrapping, compilers, kernel debugging", "use_case": "OS bootstrapping and kernel debugging", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788811153, "item_id": 49300697, "url": "https://news.ycombinator.com/item?id=49300697", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Use 3.6 27b as a daily driver for months with charmbracelet crush.", "excerpt": "Use 3.6 27b as a daily driver for months with charmbracelet crush. Gemma 26b-A3b is not even remotely comparable in terms of coding for me. YMMV depending on how you work, what harness you use, etc I suppose.", "context": "", "use_case": "coding harness", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}]}, {"label": "Lightweight", "summary": "Ships as one tiny Go binary without NPM dependencies and runs fast.", "quotes": [{"id": 788904688, "item_id": 49306543, "url": "https://news.ycombinator.com/item?id=49306543", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about", "excerpt": "Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash) Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on. But I will bite. What does pi do today ", "context": "works on bootstrapping, compilers, kernel debugging", "use_case": "OS bootstrapping and kernel debugging", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}]}, {"label": "Hackable", "summary": "Users patch it to change defaults like themes, yolo mode and prompts.", "quotes": [{"id": 798377856, "item_id": 49898616, "url": "https://news.ycombinator.com/item?id=49898616", "by": "ThomasGlanzmann", "date": "2026-09-29", "stance": "recommend", "quote": "I use crush (https://github.com/charmbracelet/crush) with the following patches:", "excerpt": "I use crush (https://github.com/charmbracelet/crush) with the following patches: curl https://tg.st/u/0001-fix-unblock-all-commands-in-bash-tool.patch | git am curl https://tg.st/u/0002-feat-add-light-theme-with-auto-detection-for-white-b.patch | git am curl https://tg.st/u/0003-feat-enable-yolo-mode-by-default.patch | git am curl https://tg.st/u/0004-fix-disable-mouse-grabbing-to-restore-native-termina.patch | git a", "context": "", "use_case": "agent harness", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 787834992, "item_id": 49239687, "url": "https://news.ycombinator.com/item?id=49239687", "by": "ThomasGlanzmann", "date": "2026-08-09", "stance": "recommend", "quote": "I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops.", "excerpt": "I do the same. I also use crush/deepseek a lot. I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops. For example to configure netapp storage systems and had not a single incident. Of course I also have backups of my systems and have everything under revision control. Depending on the task I let it run on a throw away VM, production system, lab system.", "context": "devops, runs agents on throwaway VMs with backups", "use_case": "devops and agentic tasks", "switched_from": "", "story": "Auto mode is now the default in Claude Code"}]}], "caveats": [{"label": "Defaults", "summary": "Some users patch out default command blocking, safety prompts, mouse grabbing and the project init prompt.", "quotes": [{"id": 798377856, "item_id": 49898616, "url": "https://news.ycombinator.com/item?id=49898616", "by": "ThomasGlanzmann", "date": "2026-09-29", "stance": "recommend", "quote": "I use crush (https://github.com/charmbracelet/crush) with the following patches:", "excerpt": "I use crush (https://github.com/charmbracelet/crush) with the following patches: curl https://tg.st/u/0001-fix-unblock-all-commands-in-bash-tool.patch | git am curl https://tg.st/u/0002-feat-add-light-theme-with-auto-detection-for-white-b.patch | git am curl https://tg.st/u/0003-feat-enable-yolo-mode-by-default.patch | git am curl https://tg.st/u/0004-fix-disable-mouse-grabbing-to-restore-native-termina.patch | git a", "context": "", "use_case": "agent harness", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 787834992, "item_id": 49239687, "url": "https://news.ycombinator.com/item?id=49239687", "by": "ThomasGlanzmann", "date": "2026-08-09", "stance": "recommend", "quote": "I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops.", "excerpt": "I do the same. I also use crush/deepseek a lot. I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops. For example to configure netapp storage systems and had not a single incident. Of course I also have backups of my systems and have everything under revision control. Depending on the task I let it run on a throw away VM, production system, lab system.", "context": "devops, runs agents on throwaway VMs with backups", "use_case": "devops and agentic tasks", "switched_from": "", "story": "Auto mode is now the default in Claude Code"}]}, {"label": "Subagent limits", "summary": "Subagents cannot have edit tools, which limits agent orchestration.", "quotes": [{"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}]}, {"label": "Telemetry, docs", "summary": "One user found local model support unclear, noted telemetry that must be disabled in config, and disliked a maintainer's handling of an issue.", "quotes": [{"id": 763289040, "item_id": 47705565, "url": "https://news.ycombinator.com/item?id=47705565", "by": "Ms-J", "date": "2026-04-09", "stance": "neutral", "quote": "Thanks for the recommendation, I took a look", "excerpt": "Thanks for the recommendation, I took a look and maybe you can answer a few questions that I couldn't find a clear answer when doing some quick searching. Regarding local models, can it use them? I found this discussion: https://github.com/charmbracelet/crush/discussions/775 I didn't appreciate the meow maintainer's attitude converting it into a discussion and ignoring the issue even to this day. \"It does have intern", "context": "Linux user", "use_case": "offline local-model coding agent", "switched_from": "", "story": "Best Open Source Offline AI Agent"}]}], "who": [{"by": "lrvick", "context": "works on bootstrapping, compilers, kernel debugging", "url": "https://news.ycombinator.com/item?id=49306543", "date": "2026-08-14"}, {"by": "brhenc", "context": "lives in the terminal", "url": "https://news.ycombinator.com/item?id=49296556", "date": "2026-08-14"}, {"by": "ThomasGlanzmann", "context": "devops, runs agents on throwaway VMs with backups", "url": "https://news.ycombinator.com/item?id=49239687", "date": "2026-08-09"}, {"by": "Ms-J", "context": "Linux user", "url": "https://news.ycombinator.com/item?id=47705565", "date": "2026-04-09"}], "tags": ["for terminal users", "for local models", "for open-weight models", "bring your own provider", "for devops"], "use_cases": ["coding agent", "agent harness", "coding harness", "coding agent harness", "os bootstrapping and kernel debugging", "terminal coding harness"], "compared": [{"slug": "opencode", "name": "OpenCode", "co": 5, "switched_from": 0, "switched_to": 0}, {"slug": "deepseek", "name": "DeepSeek", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "claude-code", "name": "Claude Code", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "pi", "name": "Pi", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "zcode", "name": "ZCode", "co": 1, "switched_from": 0, "switched_to": 1}, {"slug": "minimax", "name": "MiniMax", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "codex", "name": "Codex", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49299605", "title": "Qwen 3.8 27B is out: open weights, best local dense model yet", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48916512", "title": "You only need the frontier model for one single edit", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48567759", "title": "GLM-5.2 is the new leading open weights model on Artificial Analysis", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49896604", "title": "Dots: Always-on agents", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49896586", "title": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49333747", "title": "Ask HN: What's the Best Agent Harness?", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {"neutral": 1, "recommend": 1}, "2026-05": {}, "2026-06": {"recommend": 3, "neutral": 1}, "2026-07": {"recommend": 1}, "2026-08": {"recommend": 4}, "2026-09": {"recommend": 1, "neutral": 1}, "2026-10": {}}, "quotes": [{"id": 798406804, "item_id": 49900425, "url": "https://news.ycombinator.com/item?id=49900425", "by": "reachableceo", "date": "2026-09-29", "stance": "neutral", "quote": "When I moved from crush to Zcode", "excerpt": "I mean it\u2019s not exactly tin foil hat. I suspect when we see the s1 drop for these companies , the business plan will be essentially exactly what you just said. OpenWebUI has allowed me to avoid this. Combined with OpenTerminal. I am using zcode with GLM5.3 flash from z.ai due to the extra usage / air drops etc . However nothing I\u2019m doing is tied to that harness. When I moved from crush to Zcode , the first thing I di", "context": "", "use_case": "coding agent harness", "switched_from": "", "story": "Dots: Always-on agents"}, {"id": 798377856, "item_id": 49898616, "url": "https://news.ycombinator.com/item?id=49898616", "by": "ThomasGlanzmann", "date": "2026-09-29", "stance": "recommend", "quote": "I use crush (https://github.com/charmbracelet/crush) with the following patches:", "excerpt": "I use crush (https://github.com/charmbracelet/crush) with the following patches: curl https://tg.st/u/0001-fix-unblock-all-commands-in-bash-tool.patch | git am curl https://tg.st/u/0002-feat-add-light-theme-with-auto-detection-for-white-b.patch | git am curl https://tg.st/u/0003-feat-enable-yolo-mode-by-default.patch | git am curl https://tg.st/u/0004-fix-disable-mouse-grabbing-to-restore-native-termina.patch | git a", "context": "", "use_case": "agent harness", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 789496898, "item_id": 49343556, "url": "https://news.ycombinator.com/item?id=49343556", "by": "dongkeren", "date": "2026-08-18", "stance": "recommend", "quote": "you can also try different harness such as Crush and Pi", "excerpt": "You'd better take harness and provider as different parts. In your case, I think the major issue is OpenCode go provier, not necessariy OpenCode harness. If you just want to have an alternative provider, I would recommend OpenRouter, which has provider fallback and model fallback. You can try OpenCode + OpenRouter instead of OpenCode + OpenCode Go; Or you can also try different harness such as Crush and Pi, in this w", "context": "", "use_case": "agent harness", "switched_from": "", "story": "Ask HN: What's the Best Agent Harness?"}, {"id": 788904688, "item_id": 49306543, "url": "https://news.ycombinator.com/item?id=49306543", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about", "excerpt": "Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash) Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on. But I will bite. What does pi do today ", "context": "works on bootstrapping, compilers, kernel debugging", "use_case": "OS bootstrapping and kernel debugging", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788811153, "item_id": 49300697, "url": "https://news.ycombinator.com/item?id=49300697", "by": "lrvick", "date": "2026-08-14", "stance": "recommend", "quote": "Use 3.6 27b as a daily driver for months with charmbracelet crush.", "excerpt": "Use 3.6 27b as a daily driver for months with charmbracelet crush. Gemma 26b-A3b is not even remotely comparable in terms of coding for me. YMMV depending on how you work, what harness you use, etc I suppose.", "context": "", "use_case": "coding harness", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788744897, "item_id": 49296556, "url": "https://news.ycombinator.com/item?id=49296556", "by": "brhenc", "date": "2026-08-14", "stance": "recommend", "quote": "As someone who lives in the terminal the web interface doesn't feel as nice to use as things like aichat, crush, or antigravity-cli.", "excerpt": "Tried it this morning - it works with openrouter, feels a little less verbose than antigravity CLI which should help with cost. I would be happier if it wasn't javascript-based, but it's open source so it should be doable to use the models with things like https://github.com/charmbracelet/crush and optimize a little how these alternatives work with deepseek models. For prototyping the web interface is OK, but I don't", "context": "lives in the terminal", "use_case": "terminal coding harness", "switched_from": "", "story": "DeepSeek Harness official website launched"}, {"id": 788545873, "item_id": 49284117, "url": "https://news.ycombinator.com/item?id=49284117", "by": "lrvick", "date": "2026-08-13", "stance": "recommend", "quote": "I am looking at a Crush client running beside neovim in a herdr session", "excerpt": "Are there really daily driver Linux users that want to run a proprietary gui based LLM solution that Sam Altman gets to keylog AND have full remote code execution on their system? This feels like trolling. Meanwhile I am looking at a Crush client running beside neovim in a herdr session I am accessing via ssh to a machine in my garage with 4 local GPUs. I think we Linux users have got this without Sams help.", "context": "Linux user with 4 local GPUs", "use_case": "local LLM client", "switched_from": "", "story": "ChatGPT Desktop (Codex Desktop) for Linux"}, {"id": 787834992, "item_id": 49239687, "url": "https://news.ycombinator.com/item?id=49239687", "by": "ThomasGlanzmann", "date": "2026-08-09", "stance": "recommend", "quote": "I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops.", "excerpt": "I do the same. I also use crush/deepseek a lot. I patched crush to remove all safetyguards and enable yolo mode by default. I use it on a daily basis including devops. For example to configure netapp storage systems and had not a single incident. Of course I also have backups of my systems and have everything under revision control. Depending on the task I let it run on a throw away VM, production system, lab system.", "context": "devops, runs agents on throwaway VMs with backups", "use_case": "devops and agentic tasks", "switched_from": "", "story": "Auto mode is now the default in Claude Code"}, {"id": 784026960, "item_id": 49001685, "url": "https://news.ycombinator.com/item?id=49001685", "by": "ricardobeat", "date": "2026-07-21", "stance": "recommend", "quote": "It has the most user-friendly UI of them all, a really usable scroll view.", "excerpt": "It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast. The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.", "context": "", "use_case": "one-shot prompts with open models", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 783809106, "item_id": 48988069, "url": "https://news.ycombinator.com/item?id=48988069", "by": "ricardobeat", "date": "2026-07-20", "stance": "recommend", "quote": "Really easy to do with Crush/OpenCode.", "excerpt": "I just did this five minutes ago. Started a task with Kimi K3, noticed cost going up, switched to Minimax M3 continuing from the same context. Really easy to do with Crush/OpenCode. It works quite well most of the time. For very deeply technical tasks, where exploration involves multiple agents (and blowing up context limits a few times), I'll have a plan written down to disk that will be much longer than 2k tokens.", "context": "", "use_case": "switching models mid-task", "switched_from": "", "story": "You only need the frontier model for one single edit"}, {"id": 781183648, "item_id": 48823978, "url": "https://news.ycombinator.com/item?id=48823978", "by": "ricardobeat", "date": "2026-07-07", "stance": "recommend", "quote": "Crush is pretty nice. It\u2019s become my daily driver for personal projects.", "excerpt": "Crush is pretty nice. It\u2019s become my daily driver for personal projects. They maintain a list of supported providers so most of the time no configuration is needed. OpenCode is also fine but more finnicky - if you\u2019re ok with desktop apps, give OpenChamber a try.", "context": "", "use_case": "switching LLM providers for coding", "switched_from": "", "story": "Ask HN: What is your AI harness that lets you switch LLM models easily?"}, {"id": 777180848, "item_id": 48573803, "url": "https://news.ycombinator.com/item?id=48573803", "by": "granra", "date": "2026-06-17", "stance": "recommend", "quote": "I use a coding agent (crush), prompt it to brainstorm an implementation with me", "excerpt": "I'm not entirely sure what an agentic workflow could mean today but I think so. I use a coding agent (crush), prompt it to brainstorm an implementation with me (or sometimes I know exactly how I want to implement it but ask it to challenge it), correct any wrong assumptions or request the implementation to look differently than suggested if I don't like it. Then finally when I'm positive I've cleared the most importa", "context": "", "use_case": "coding agent", "switched_from": "", "story": "GLM-5.2 is the new leading open weights model on Artificial Analysis"}, {"id": 777166272, "item_id": 48572892, "url": "https://news.ycombinator.com/item?id=48572892", "by": "granra", "date": "2026-06-17", "stance": "neutral", "quote": "I use crush normally.", "excerpt": "That's what I've been doing. I use crush normally. While the codebase are by no means huge, they're not tiny either.", "context": "", "use_case": "coding agent", "switched_from": "", "story": "GLM-5.2 is the new leading open weights model on Artificial Analysis"}, {"id": 776752354, "item_id": 48547022, "url": "https://news.ycombinator.com/item?id=48547022", "by": "ThomasGlanzmann", "date": "2026-06-15", "stance": "recommend", "quote": "I use crush as coding agent but removed the blocked commands because I also do a lot of system administration. Love it.", "excerpt": "I do the same. deepseekv4 fast for the 90% of the tasks, if it can't lift it, I use deepseekv4 pro. I use crush as coding agent but removed the blocked commands because I also do a lot of system administration. Love it. I use 8 USD in 7 weeks and use it quiet extensively for all sorts of things, programming, system administration, google search replacement, investments, you name it.", "context": "does a lot of system administration", "use_case": "coding agent", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776357041, "item_id": 48522315, "url": "https://news.ycombinator.com/item?id=48522315", "by": "saratogacx", "date": "2026-06-13", "stance": "recommend", "quote": "I'm using crush and have had a good experience", "excerpt": "I've been using GLM-5/5.1 for about 6 months and it has been a fairly capable model. I've seen a lot of mixed opinions that tend to align with harness usage so it is worth trying out a couple with a model before writing it off. For example, I'm using crush and have had a good experience while others using CC have had a much more mixed experience. For task complexity, I treat it as I would sonnet with the same care in", "context": "", "use_case": "coding harness", "switched_from": "", "story": "GLM 5.2 Is Out"}, {"id": 775055541, "item_id": 48440971, "url": "https://news.ycombinator.com/item?id=48440971", "by": "slopinthebag", "date": "2026-06-07", "stance": "recommend", "quote": "you could download a different harness (Crush, OpenCode, Pi etc) and load $5 in credits and test it for yourself", "excerpt": "I used ~16,000,000 input tokens yesterday on v4 pro, ~15,000,000 were cache hits, and I spent $0.47. Output tokens were negligible. However that's with Zed's harness, I'm not sure what you would get with Claude Code. It's not as knowledgeable as the most expensive American models and makes more mistakes, so you need to constrain its scope more. That suits my workflow, half the time I have it generate code in the chat", "context": "", "use_case": "AI coding harness", "switched_from": "", "story": "DeepSeek V4 Pro beats GPT-5.5 Pro on precision"}, {"id": 766287793, "item_id": 47892987, "url": "https://news.ycombinator.com/item?id=47892987", "by": "nickdothutton", "date": "2026-04-24", "stance": "recommend", "quote": "Having some success with Qwen+Crush and have been more productive.", "excerpt": "Switched to local models after quality dropped off a cliff and token consumption seemed to double. Having some success with Qwen+Crush and have been more productive.", "context": "", "use_case": "local coding agent", "switched_from": "Claude", "story": "I Cancelled Claude: Token Issues, Declining Quality, and Poor Support"}, {"id": 763289040, "item_id": 47705565, "url": "https://news.ycombinator.com/item?id=47705565", "by": "Ms-J", "date": "2026-04-09", "stance": "neutral", "quote": "Thanks for the recommendation, I took a look", "excerpt": "Thanks for the recommendation, I took a look and maybe you can answer a few questions that I couldn't find a clear answer when doing some quick searching. Regarding local models, can it use them? I found this discussion: https://github.com/charmbracelet/crush/discussions/775 I didn't appreciate the meow maintainer's attitude converting it into a discussion and ignoring the issue even to this day. \"It does have intern", "context": "Linux user", "use_case": "offline local-model coding agent", "switched_from": "", "story": "Best Open Source Offline AI Agent"}]}