{"slug": "rtk", "name": "RTK", "category": "ai-coding", "category_name": "AI coding tools", "description": "CLI proxy that compresses shell command output to cut LLM coding-agent token usage.", "recommenders": 5, "commenters": 13, "warners": 4, "score": 1.736, "rising": -0.31, "first": "2026-03-31", "latest": "2026-09-12", "why": "Used to cut input tokens in coding agents like Claude Code by compressing command output, with the biggest gains on CLIs such as gh and docker.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 5, "neutral": 3, "compare": 2, "warn": 4}, "spark": [0, 0, 0, 0, 1, 1, 0, 2, 0, 0, 1, 0], "listed": true, "loved": ["Fewer input tokens", "Agent integration"], "caveat": "Small savings", "short_quote": "for agentic coding e.g. Claude Code I found this useful", "icon": true, "domain": null, "github_repo": "github.com/rtk-ai/rtk", "official_url": "https://github.com/rtk-ai/rtk", "pricing": "open source", "aliases": ["RTK", "rtk"], "verdict": [{"text": "Used to cut input tokens in coding agents like Claude Code by compressing command output", "cites": [{"id": 761491202, "item_id": 47593200, "url": "https://news.ycombinator.com/item?id=47593200", "by": "valentinconan", "date": "2026-03-31", "stance": "recommend", "quote": "It uses hooks to replace certain commands with optimized ones and helps reduce the number of input tokens.", "excerpt": "I used to use Gemini CLI and was pretty happy with it\u2026 until I started using Claude Code. I think it\u2019s, for now, the most advanced AI tool available right now. You can manage skills, the MCP, plugins, agents, hooks\u2026 It\u2019s a truly comprehensive tool, but its main drawback is the token consumption of the Anthropic model, especially in recent days. There are solutions to reduce input tokens, for example this GitHub proje", "context": "", "use_case": "reduce Claude Code token usage", "switched_from": "", "story": "Ask HN: Gemini CLI vs. Claude Code"}, {"id": 777486368, "item_id": 48592898, "url": "https://news.ycombinator.com/item?id=48592898", "by": "ilia-a", "date": "2026-06-18", "stance": "recommend", "quote": "Works fine, yeah it only compresses command output so only input tokens are affected", "excerpt": "First of all there is a way to made agents spot truncation by being aware of RTK compression and having bypass option (I use RTK_DISABLE=1) as a way of restoring original full text. Works fine, yeah it only compresses command output so only input tokens are affected in terms of \"compression\".", "context": "", "use_case": "compressing agent command output", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 775324896, "item_id": 48457806, "url": "https://news.ycombinator.com/item?id=48457806", "by": "mhrmsn", "date": "2026-06-09", "stance": "recommend", "quote": "for agentic coding e.g. Claude Code I found this useful", "excerpt": "Haven't benchmarked it, but for agentic coding e.g. Claude Code I found this useful: https://github.com/rtk-ai/rtk", "context": "", "use_case": "cutting AI token costs", "switched_from": "", "story": "Ask HN: What works for cutting AI token costs?"}]}, {"text": ", with the biggest gains on CLIs such as gh and docker.", "cites": [{"id": 794714496, "item_id": 49669656, "url": "https://news.ycombinator.com/item?id=49669656", "by": "patriciobcs", "date": "2026-09-12", "stance": "recommend", "quote": "RTK has save me a lot of usage there", "excerpt": "It seems to depend on the task you are doing. In my case, I use a lot gh and docker CLI, and RTK has save me a lot of usage there. However, using other others CLIs like grep or git doesn\u2019t seem to help that much. I use it mainly for security reviews of codebase and for this particular use case it helps a lot.", "context": "", "use_case": "token savings for security reviews", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}]}, {"text": " The most common complaint is that the savings are small or unnoticeable", "cites": [{"id": 777521536, "item_id": 48595096, "url": "https://news.ycombinator.com/item?id=48595096", "by": "Otterly99", "date": "2026-06-18", "stance": "warn", "quote": "After I used it in one session of 300k tokens, I had maybe 3k tokens saved.", "excerpt": "I completely agree with this post. After I used it in one session of 300k tokens, I had maybe 3k tokens saved. Plus, if commits really are an issue for you in term of tolen consumption, you can always ask to hand over the reigns and apply the commits yourself as a rule (unless you're operating in a loop).", "context": "", "use_case": "token compression for agents", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 766662368, "item_id": 47916398, "url": "https://news.ycombinator.com/item?id=47916398", "by": "esperent", "date": "2026-04-26", "stance": "neutral", "quote": "I have tried both of these and I have to say, I didn't notice any difference.", "excerpt": "Similar tools in the space: RTK which I guess is a close equivalent. https://github.com/rtk-ai/rtk Headroom which is complementary. https://github.com/chopratejas/headroom I have tried both of these and I have to say, I didn't notice any difference. That's not to say there wasn't a difference (rtk logs were telling me enthusiastically how many tokens I'd saved), it's just that pretty much all context is coming from r", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "8v: One CLI for you and your AI agent. Up to 66% fewer tokens"}, {"id": 788751346, "item_id": 49296959, "url": "https://news.ycombinator.com/item?id=49296959", "by": "postatic", "date": "2026-08-14", "stance": "neutral", "quote": "Has it helped? Somewhat.", "excerpt": "Oh the verbosity and the cryptic words that it uses. The other day, all of a sudden it used an acronym \"DoD\". I had no idea what it was and made me feel dumb. It's \"Definition of Done\". I don't care how widely used this acronym is, you just can't throw it in there. I've now installed quite a number of tools to combat this. Just in the last few days I've installed - https://www.codewithbullet.com - https://maki.sh - h", "context": "", "use_case": "curb Opus verbosity", "switched_from": "", "story": "Why does Opus 5 feel worse to work with?"}]}, {"text": ", and some found agents ran longer or did worse with it enabled.", "cites": [{"id": 794576672, "item_id": 49661042, "url": "https://news.ycombinator.com/item?id=49661042", "by": "kriskrunch", "date": "2026-09-11", "stance": "warn", "quote": "Empirically, I found RTK makes my agents run longer to complete similar tasks.", "excerpt": "Open question, how does this instruction in agent-rules.md look? \"Cap large/unknown command output: `COMMAND 2>&1 | head -c 4000`. Never stream full logs, tests, or large files.\" I use that instead of RTK. Empirically, I found RTK makes my agents run longer to complete similar tasks. Ponytail and Caveman seem to help somewhat.", "context": "", "use_case": "agent token savings", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 770792800, "item_id": 48174550, "url": "https://news.ycombinator.com/item?id=48174550", "by": "esperent", "date": "2026-05-17", "stance": "warn", "quote": "I got clearly better results with both off", "excerpt": "I did some evals with pi and GPT 5.5. I tested RTK on / headroom on / both on / both off (all with the standard pi system instructions and no AGENTS.md). I forget the exact tests I used (a couple of the standard agent evals that people use, one python and one typescript because those are what I use). I don't claim it was an exhaustive test, or even a good one. It's possible I could have spent a day or so tuning my AG", "context": "ran own agent evals", "use_case": "reducing agent token use", "switched_from": "", "story": "Show HN: Semble \u2013 Code search for agents that uses 98% fewer tokens than grep"}]}], "praise": [{"label": "Fewer input tokens", "summary": "Compresses command output via hooks to reduce input tokens sent to the model.", "quotes": [{"id": 761491202, "item_id": 47593200, "url": "https://news.ycombinator.com/item?id=47593200", "by": "valentinconan", "date": "2026-03-31", "stance": "recommend", "quote": "It uses hooks to replace certain commands with optimized ones and helps reduce the number of input tokens.", "excerpt": "I used to use Gemini CLI and was pretty happy with it\u2026 until I started using Claude Code. I think it\u2019s, for now, the most advanced AI tool available right now. You can manage skills, the MCP, plugins, agents, hooks\u2026 It\u2019s a truly comprehensive tool, but its main drawback is the token consumption of the Anthropic model, especially in recent days. There are solutions to reduce input tokens, for example this GitHub proje", "context": "", "use_case": "reduce Claude Code token usage", "switched_from": "", "story": "Ask HN: Gemini CLI vs. Claude Code"}, {"id": 777486368, "item_id": 48592898, "url": "https://news.ycombinator.com/item?id=48592898", "by": "ilia-a", "date": "2026-06-18", "stance": "recommend", "quote": "Works fine, yeah it only compresses command output so only input tokens are affected", "excerpt": "First of all there is a way to made agents spot truncation by being aware of RTK compression and having bypass option (I use RTK_DISABLE=1) as a way of restoring original full text. Works fine, yeah it only compresses command output so only input tokens are affected in terms of \"compression\".", "context": "", "use_case": "compressing agent command output", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 794714496, "item_id": 49669656, "url": "https://news.ycombinator.com/item?id=49669656", "by": "patriciobcs", "date": "2026-09-12", "stance": "recommend", "quote": "RTK has save me a lot of usage there", "excerpt": "It seems to depend on the task you are doing. In my case, I use a lot gh and docker CLI, and RTK has save me a lot of usage there. However, using other others CLIs like grep or git doesn\u2019t seem to help that much. I use it mainly for security reviews of codebase and for this particular use case it helps a lot.", "context": "", "use_case": "token savings for security reviews", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}]}, {"label": "Agent integration", "summary": "Commenters found it useful alongside agents like Claude Code and Pi.", "quotes": [{"id": 775324896, "item_id": 48457806, "url": "https://news.ycombinator.com/item?id=48457806", "by": "mhrmsn", "date": "2026-06-09", "stance": "recommend", "quote": "for agentic coding e.g. Claude Code I found this useful", "excerpt": "Haven't benchmarked it, but for agentic coding e.g. Claude Code I found this useful: https://github.com/rtk-ai/rtk", "context": "", "use_case": "cutting AI token costs", "switched_from": "", "story": "Ask HN: What works for cutting AI token costs?"}, {"id": 767545892, "item_id": 47971618, "url": "https://news.ycombinator.com/item?id=47971618", "by": "spqw", "date": "2026-04-30", "stance": "recommend", "quote": "(3) using rtk", "excerpt": "I got great results by (1) using pi coding agent - i often see a 20x fdifference with code cli whoch is more token hungry (2) using an advisor strategy. See https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool I use GPT-5.3-Codex-Spark for all tasks and it will ask help to GPT-5.5 when the task is complex enough (3) using rtk", "context": "", "use_case": "extending Codex sessions", "switched_from": "", "story": "Ask HN: Any good ways to extend Codex sessions?"}, {"id": 761491202, "item_id": 47593200, "url": "https://news.ycombinator.com/item?id=47593200", "by": "valentinconan", "date": "2026-03-31", "stance": "recommend", "quote": "It uses hooks to replace certain commands with optimized ones and helps reduce the number of input tokens.", "excerpt": "I used to use Gemini CLI and was pretty happy with it\u2026 until I started using Claude Code. I think it\u2019s, for now, the most advanced AI tool available right now. You can manage skills, the MCP, plugins, agents, hooks\u2026 It\u2019s a truly comprehensive tool, but its main drawback is the token consumption of the Anthropic model, especially in recent days. There are solutions to reduce input tokens, for example this GitHub proje", "context": "", "use_case": "reduce Claude Code token usage", "switched_from": "", "story": "Ask HN: Gemini CLI vs. Claude Code"}]}, {"label": "Bypass option", "summary": "Compression can be disabled with RTK_DISABLE=1 to restore full output.", "quotes": [{"id": 777486368, "item_id": 48592898, "url": "https://news.ycombinator.com/item?id=48592898", "by": "ilia-a", "date": "2026-06-18", "stance": "recommend", "quote": "Works fine, yeah it only compresses command output so only input tokens are affected", "excerpt": "First of all there is a way to made agents spot truncation by being aware of RTK compression and having bypass option (I use RTK_DISABLE=1) as a way of restoring original full text. Works fine, yeah it only compresses command output so only input tokens are affected in terms of \"compression\".", "context": "", "use_case": "compressing agent command output", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}]}, {"label": "Narrow focus", "summary": "Described as focused purely on output compression, unlike broader tools.", "quotes": [{"id": 777712688, "item_id": 48607043, "url": "https://news.ycombinator.com/item?id=48607043", "by": "bonigv", "date": "2026-06-20", "stance": "compare", "quote": "RTK on the otherhand is focused purely on the compression.", "excerpt": "Clarifying - I am not suggesting headroom and lean-ctx are only doing the tool compression. It runs elaborate subsystems towards a noble objective of saving tokens. RTK on the otherhand is focused purely on the compression. At the sametime, platforms - the providers - especially Anthropic and and Open AI have transparently absorbed many of the big ticket token saving items these tools are cleverly managing. That is a", "context": "", "use_case": "tool output compression", "switched_from": "", "story": "Ask HN: Will we start seeing tools for LLM use?"}]}], "caveats": [{"label": "Small savings", "summary": "Several users saw little or no noticeable token reduction in practice.", "quotes": [{"id": 777521536, "item_id": 48595096, "url": "https://news.ycombinator.com/item?id=48595096", "by": "Otterly99", "date": "2026-06-18", "stance": "warn", "quote": "After I used it in one session of 300k tokens, I had maybe 3k tokens saved.", "excerpt": "I completely agree with this post. After I used it in one session of 300k tokens, I had maybe 3k tokens saved. Plus, if commits really are an issue for you in term of tolen consumption, you can always ask to hand over the reigns and apply the commits yourself as a rule (unless you're operating in a loop).", "context": "", "use_case": "token compression for agents", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 766662368, "item_id": 47916398, "url": "https://news.ycombinator.com/item?id=47916398", "by": "esperent", "date": "2026-04-26", "stance": "neutral", "quote": "I have tried both of these and I have to say, I didn't notice any difference.", "excerpt": "Similar tools in the space: RTK which I guess is a close equivalent. https://github.com/rtk-ai/rtk Headroom which is complementary. https://github.com/chopratejas/headroom I have tried both of these and I have to say, I didn't notice any difference. That's not to say there wasn't a difference (rtk logs were telling me enthusiastically how many tokens I'd saved), it's just that pretty much all context is coming from r", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "8v: One CLI for you and your AI agent. Up to 66% fewer tokens"}, {"id": 788751346, "item_id": 49296959, "url": "https://news.ycombinator.com/item?id=49296959", "by": "postatic", "date": "2026-08-14", "stance": "neutral", "quote": "Has it helped? Somewhat.", "excerpt": "Oh the verbosity and the cryptic words that it uses. The other day, all of a sudden it used an acronym \"DoD\". I had no idea what it was and made me feel dumb. It's \"Definition of Done\". I don't care how widely used this acronym is, you just can't throw it in there. I've now installed quite a number of tools to combat this. Just in the last few days I've installed - https://www.codewithbullet.com - https://maki.sh - h", "context": "", "use_case": "curb Opus verbosity", "switched_from": "", "story": "Why does Opus 5 feel worse to work with?"}]}, {"label": "Worse agent results", "summary": "Some found agents took longer or scored better with it turned off.", "quotes": [{"id": 794576672, "item_id": 49661042, "url": "https://news.ycombinator.com/item?id=49661042", "by": "kriskrunch", "date": "2026-09-11", "stance": "warn", "quote": "Empirically, I found RTK makes my agents run longer to complete similar tasks.", "excerpt": "Open question, how does this instruction in agent-rules.md look? \"Cap large/unknown command output: `COMMAND 2>&1 | head -c 4000`. Never stream full logs, tests, or large files.\" I use that instead of RTK. Empirically, I found RTK makes my agents run longer to complete similar tasks. Ponytail and Caveman seem to help somewhat.", "context": "", "use_case": "agent token savings", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 770792800, "item_id": 48174550, "url": "https://news.ycombinator.com/item?id=48174550", "by": "esperent", "date": "2026-05-17", "stance": "warn", "quote": "I got clearly better results with both off", "excerpt": "I did some evals with pi and GPT 5.5. I tested RTK on / headroom on / both on / both off (all with the standard pi system instructions and no AGENTS.md). I forget the exact tests I used (a couple of the standard agent evals that people use, one python and one typescript because those are what I use). I don't claim it was an exhaustive test, or even a good one. It's possible I could have spent a day or so tuning my AG", "context": "ran own agent evals", "use_case": "reducing agent token use", "switched_from": "", "story": "Show HN: Semble \u2013 Code search for agents that uses 98% fewer tokens than grep"}]}, {"label": "Task dependent", "summary": "Helps with some CLIs but not grep, git, or file reads.", "quotes": [{"id": 794714496, "item_id": 49669656, "url": "https://news.ycombinator.com/item?id=49669656", "by": "patriciobcs", "date": "2026-09-12", "stance": "recommend", "quote": "RTK has save me a lot of usage there", "excerpt": "It seems to depend on the task you are doing. In my case, I use a lot gh and docker CLI, and RTK has save me a lot of usage there. However, using other others CLIs like grep or git doesn\u2019t seem to help that much. I use it mainly for security reviews of codebase and for this particular use case it helps a lot.", "context": "", "use_case": "token savings for security reviews", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 766662368, "item_id": 47916398, "url": "https://news.ycombinator.com/item?id=47916398", "by": "esperent", "date": "2026-04-26", "stance": "neutral", "quote": "I have tried both of these and I have to say, I didn't notice any difference.", "excerpt": "Similar tools in the space: RTK which I guess is a close equivalent. https://github.com/rtk-ai/rtk Headroom which is complementary. https://github.com/chopratejas/headroom I have tried both of these and I have to say, I didn't notice any difference. That's not to say there wasn't a difference (rtk logs were telling me enthusiastically how many tokens I'd saved), it's just that pretty much all context is coming from r", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "8v: One CLI for you and your AI agent. Up to 66% fewer tokens"}]}, {"label": "Skepticism", "summary": "Some call token-saving hacks like it snake oil or expect many caveats.", "quotes": [{"id": 794505952, "item_id": 49656622, "url": "https://news.ycombinator.com/item?id=49656622", "by": "aeneas_ory", "date": "2026-09-11", "stance": "warn", "quote": "All of these \"hacks\" are snakeoil", "excerpt": "All of these \"hacks\" are snakeoil and I think deep down we all know. Whether it's caveman, RTK, or whatever other vibe-coded productivity/token cost saving hacks/skills/claude.md. What I had success with (although benchmarks are older) is to index the codebase with a dedicated local code embedding model. It's a bit expensive on the CPU side but in my benchmarks it reduced token use and wall clock time significantly. ", "context": "", "use_case": "token savings", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 776422608, "item_id": 48526413, "url": "https://news.ycombinator.com/item?id=48526413", "by": "brunoluiz", "date": "2026-06-14", "stance": "neutral", "quote": "which decreases the amount of token used by tool calls and, although it seems an interesting idea, I am pretty sure there are many caveats", "excerpt": "100% with the author on that one, albeit the performance decay seems to depend on the type of task for me. Simple plumbing tasks seem to run okay with longer running contexts. Also, some colleagues were playing around with RTK (https://github.com/rtk-ai/rtk), which decreases the amount of token used by tool calls and, although it seems an interesting idea, I am pretty sure there are many caveats. Although, I believe ", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "Don't trust large context windows"}]}, {"label": "Alternatives preferred", "summary": "One commenter preferred Headroom for its accuracy benchmarks and transparency.", "quotes": [{"id": 777501729, "item_id": 48593858, "url": "https://news.ycombinator.com/item?id=48593858", "by": "jvican", "date": "2026-06-18", "stance": "compare", "quote": "I liked their approach a lot better than RTK", "excerpt": "An ex colleague is working on Headroom, a much more legit alternative to RTK. They provide accuracy benchmarks in the repo and are transparent about the compression algorithms used for the different output types. I liked their approach a lot better than RTK and thought it might be relevant for you. https://github.com/chopratejas/headroom", "context": "", "use_case": "LLM token compression", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}]}], "who": [], "tags": ["for Claude Code users", "for agentic coding", "for heavy CLI users"], "use_cases": ["reducing token usage", "token savings for security reviews", "agent token savings", "token savings", "curb opus verbosity", "tool output compression"], "compared": [{"slug": "claude-code", "name": "Claude Code", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "headroom", "name": "headroom", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "caveman", "name": "caveman", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "pi", "name": "Pi", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "gpt-5", "name": "GPT-5", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 751595904, "item_id": 46974744, "url": "https://news.ycombinator.com/item?id=46974744", "by": "patrick4urcloud", "date": "2026-02-11", "stance": "recommend", "quote": "So I built RTK. It wraps commands and strips the junk before it reaches your context", "excerpt": "I use Claude Code all day and got tired of watching it eat tokens on noisy CLI output. git status on a big repo, find across a project, cargo test with 200 lines of compilation \u2014 none of that noise is useful to the model. So I built RTK. It wraps commands and strips the junk before it reaches your context: rtk git status # instead of git status rtk find \"*.rs\" . # compact results rtk ls . # token-optimized tree rtk c", "context": "uses Claude Code all day", "use_case": "saving tokens in AI coding agents", "switched_from": "", "story": "Show HN: RTK \u2013 Wrap your CLI commands, save 60-90% of tokens in AI coding agents"}], "threads": [{"url": "https://news.ycombinator.com/item?id=49656471", "title": "RTK reports token savings, but our cost benchmarks disagree", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=48588755", "title": "The Token Compression Illusion: Why I'm Skeptical of RTK", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49296740", "title": "Why does Opus 5 feel worse to work with?", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48606997", "title": "Ask HN: Will we start seeing tools for LLM use?", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48524620", "title": "Don't trust large context windows", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48457585", "title": "Ask HN: What works for cutting AI token costs?", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {"recommend": 1}, "2026-04": {"neutral": 1, "recommend": 1}, "2026-05": {"warn": 1}, "2026-06": {"recommend": 2, "neutral": 1, "compare": 2, "warn": 1}, "2026-07": {}, "2026-08": {"neutral": 1}, "2026-09": {"warn": 2, "recommend": 1}, "2026-10": {}}, "quotes": [{"id": 794714496, "item_id": 49669656, "url": "https://news.ycombinator.com/item?id=49669656", "by": "patriciobcs", "date": "2026-09-12", "stance": "recommend", "quote": "RTK has save me a lot of usage there", "excerpt": "It seems to depend on the task you are doing. In my case, I use a lot gh and docker CLI, and RTK has save me a lot of usage there. However, using other others CLIs like grep or git doesn\u2019t seem to help that much. I use it mainly for security reviews of codebase and for this particular use case it helps a lot.", "context": "", "use_case": "token savings for security reviews", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 794576672, "item_id": 49661042, "url": "https://news.ycombinator.com/item?id=49661042", "by": "kriskrunch", "date": "2026-09-11", "stance": "warn", "quote": "Empirically, I found RTK makes my agents run longer to complete similar tasks.", "excerpt": "Open question, how does this instruction in agent-rules.md look? \"Cap large/unknown command output: `COMMAND 2>&1 | head -c 4000`. Never stream full logs, tests, or large files.\" I use that instead of RTK. Empirically, I found RTK makes my agents run longer to complete similar tasks. Ponytail and Caveman seem to help somewhat.", "context": "", "use_case": "agent token savings", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 794505952, "item_id": 49656622, "url": "https://news.ycombinator.com/item?id=49656622", "by": "aeneas_ory", "date": "2026-09-11", "stance": "warn", "quote": "All of these \"hacks\" are snakeoil", "excerpt": "All of these \"hacks\" are snakeoil and I think deep down we all know. Whether it's caveman, RTK, or whatever other vibe-coded productivity/token cost saving hacks/skills/claude.md. What I had success with (although benchmarks are older) is to index the codebase with a dedicated local code embedding model. It's a bit expensive on the CPU side but in my benchmarks it reduced token use and wall clock time significantly. ", "context": "", "use_case": "token savings", "switched_from": "", "story": "RTK reports token savings, but our cost benchmarks disagree"}, {"id": 788751346, "item_id": 49296959, "url": "https://news.ycombinator.com/item?id=49296959", "by": "postatic", "date": "2026-08-14", "stance": "neutral", "quote": "Has it helped? Somewhat.", "excerpt": "Oh the verbosity and the cryptic words that it uses. The other day, all of a sudden it used an acronym \"DoD\". I had no idea what it was and made me feel dumb. It's \"Definition of Done\". I don't care how widely used this acronym is, you just can't throw it in there. I've now installed quite a number of tools to combat this. Just in the last few days I've installed - https://www.codewithbullet.com - https://maki.sh - h", "context": "", "use_case": "curb Opus verbosity", "switched_from": "", "story": "Why does Opus 5 feel worse to work with?"}, {"id": 777712688, "item_id": 48607043, "url": "https://news.ycombinator.com/item?id=48607043", "by": "bonigv", "date": "2026-06-20", "stance": "compare", "quote": "RTK on the otherhand is focused purely on the compression.", "excerpt": "Clarifying - I am not suggesting headroom and lean-ctx are only doing the tool compression. It runs elaborate subsystems towards a noble objective of saving tokens. RTK on the otherhand is focused purely on the compression. At the sametime, platforms - the providers - especially Anthropic and and Open AI have transparently absorbed many of the big ticket token saving items these tools are cleverly managing. That is a", "context": "", "use_case": "tool output compression", "switched_from": "", "story": "Ask HN: Will we start seeing tools for LLM use?"}, {"id": 777521536, "item_id": 48595096, "url": "https://news.ycombinator.com/item?id=48595096", "by": "Otterly99", "date": "2026-06-18", "stance": "warn", "quote": "After I used it in one session of 300k tokens, I had maybe 3k tokens saved.", "excerpt": "I completely agree with this post. After I used it in one session of 300k tokens, I had maybe 3k tokens saved. Plus, if commits really are an issue for you in term of tolen consumption, you can always ask to hand over the reigns and apply the commits yourself as a rule (unless you're operating in a loop).", "context": "", "use_case": "token compression for agents", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 777501729, "item_id": 48593858, "url": "https://news.ycombinator.com/item?id=48593858", "by": "jvican", "date": "2026-06-18", "stance": "compare", "quote": "I liked their approach a lot better than RTK", "excerpt": "An ex colleague is working on Headroom, a much more legit alternative to RTK. They provide accuracy benchmarks in the repo and are transparent about the compression algorithms used for the different output types. I liked their approach a lot better than RTK and thought it might be relevant for you. https://github.com/chopratejas/headroom", "context": "", "use_case": "LLM token compression", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 777486368, "item_id": 48592898, "url": "https://news.ycombinator.com/item?id=48592898", "by": "ilia-a", "date": "2026-06-18", "stance": "recommend", "quote": "Works fine, yeah it only compresses command output so only input tokens are affected", "excerpt": "First of all there is a way to made agents spot truncation by being aware of RTK compression and having bypass option (I use RTK_DISABLE=1) as a way of restoring original full text. Works fine, yeah it only compresses command output so only input tokens are affected in terms of \"compression\".", "context": "", "use_case": "compressing agent command output", "switched_from": "", "story": "The Token Compression Illusion: Why I'm Skeptical of RTK"}, {"id": 776422608, "item_id": 48526413, "url": "https://news.ycombinator.com/item?id=48526413", "by": "brunoluiz", "date": "2026-06-14", "stance": "neutral", "quote": "which decreases the amount of token used by tool calls and, although it seems an interesting idea, I am pretty sure there are many caveats", "excerpt": "100% with the author on that one, albeit the performance decay seems to depend on the type of task for me. Simple plumbing tasks seem to run okay with longer running contexts. Also, some colleagues were playing around with RTK (https://github.com/rtk-ai/rtk), which decreases the amount of token used by tool calls and, although it seems an interesting idea, I am pretty sure there are many caveats. Although, I believe ", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "Don't trust large context windows"}, {"id": 775324896, "item_id": 48457806, "url": "https://news.ycombinator.com/item?id=48457806", "by": "mhrmsn", "date": "2026-06-09", "stance": "recommend", "quote": "for agentic coding e.g. Claude Code I found this useful", "excerpt": "Haven't benchmarked it, but for agentic coding e.g. Claude Code I found this useful: https://github.com/rtk-ai/rtk", "context": "", "use_case": "cutting AI token costs", "switched_from": "", "story": "Ask HN: What works for cutting AI token costs?"}, {"id": 770792800, "item_id": 48174550, "url": "https://news.ycombinator.com/item?id=48174550", "by": "esperent", "date": "2026-05-17", "stance": "warn", "quote": "I got clearly better results with both off", "excerpt": "I did some evals with pi and GPT 5.5. I tested RTK on / headroom on / both on / both off (all with the standard pi system instructions and no AGENTS.md). I forget the exact tests I used (a couple of the standard agent evals that people use, one python and one typescript because those are what I use). I don't claim it was an exhaustive test, or even a good one. It's possible I could have spent a day or so tuning my AG", "context": "ran own agent evals", "use_case": "reducing agent token use", "switched_from": "", "story": "Show HN: Semble \u2013 Code search for agents that uses 98% fewer tokens than grep"}, {"id": 767545892, "item_id": 47971618, "url": "https://news.ycombinator.com/item?id=47971618", "by": "spqw", "date": "2026-04-30", "stance": "recommend", "quote": "(3) using rtk", "excerpt": "I got great results by (1) using pi coding agent - i often see a 20x fdifference with code cli whoch is more token hungry (2) using an advisor strategy. See https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool I use GPT-5.3-Codex-Spark for all tasks and it will ask help to GPT-5.5 when the task is complex enough (3) using rtk", "context": "", "use_case": "extending Codex sessions", "switched_from": "", "story": "Ask HN: Any good ways to extend Codex sessions?"}, {"id": 766662368, "item_id": 47916398, "url": "https://news.ycombinator.com/item?id=47916398", "by": "esperent", "date": "2026-04-26", "stance": "neutral", "quote": "I have tried both of these and I have to say, I didn't notice any difference.", "excerpt": "Similar tools in the space: RTK which I guess is a close equivalent. https://github.com/rtk-ai/rtk Headroom which is complementary. https://github.com/chopratejas/headroom I have tried both of these and I have to say, I didn't notice any difference. That's not to say there wasn't a difference (rtk logs were telling me enthusiastically how many tokens I'd saved), it's just that pretty much all context is coming from r", "context": "", "use_case": "reducing token usage", "switched_from": "", "story": "8v: One CLI for you and your AI agent. Up to 66% fewer tokens"}, {"id": 761491202, "item_id": 47593200, "url": "https://news.ycombinator.com/item?id=47593200", "by": "valentinconan", "date": "2026-03-31", "stance": "recommend", "quote": "It uses hooks to replace certain commands with optimized ones and helps reduce the number of input tokens.", "excerpt": "I used to use Gemini CLI and was pretty happy with it\u2026 until I started using Claude Code. I think it\u2019s, for now, the most advanced AI tool available right now. You can manage skills, the MCP, plugins, agents, hooks\u2026 It\u2019s a truly comprehensive tool, but its main drawback is the token consumption of the Anthropic model, especially in recent days. There are solutions to reduce input tokens, for example this GitHub proje", "context": "", "use_case": "reduce Claude Code token usage", "switched_from": "", "story": "Ask HN: Gemini CLI vs. Claude Code"}]}