<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: Bnjoroge</title><link>https://news.ycombinator.com/user?id=Bnjoroge</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Sat, 01 Aug 2026 02:23:51 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=Bnjoroge" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by Bnjoroge in "Show HN: BitBang – Reach machines behind NAT from a browser, no account"]]></title><description><![CDATA[
<p>how's this different from iroh?</p>
]]></description><pubDate>Sat, 01 Aug 2026 02:10:09 +0000</pubDate><link>https://news.ycombinator.com/item?id=49130457</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49130457</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49130457</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Show HN: Kedge – Full-stack cloud with forkable VM snapshots and global SQLite"]]></title><description><![CDATA[
<p>certainly sounded like one of</p>
]]></description><pubDate>Fri, 31 Jul 2026 16:21:42 +0000</pubDate><link>https://news.ycombinator.com/item?id=49125145</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49125145</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49125145</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Google fixed more Chrome bugs in June than over the past two years, thanks to AI"]]></title><description><![CDATA[
<p>What model is “AI”?</p>
]]></description><pubDate>Fri, 31 Jul 2026 15:54:12 +0000</pubDate><link>https://news.ycombinator.com/item?id=49124669</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49124669</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49124669</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Git worktrees are not an isolation boundary for coding agents"]]></title><description><![CDATA[
<p>til, thanks!</p>
]]></description><pubDate>Fri, 31 Jul 2026 14:56:20 +0000</pubDate><link>https://news.ycombinator.com/item?id=49123994</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49123994</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49123994</guid></item><item><title><![CDATA[New comment by Bnjoroge in "DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis"]]></title><description><![CDATA[
<p>Using tokens to evaluate models is an outdated approach. Cost per task is what matters. Not all tokens are created equal</p>
]]></description><pubDate>Fri, 31 Jul 2026 13:14:52 +0000</pubDate><link>https://news.ycombinator.com/item?id=49122723</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49122723</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49122723</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Git worktrees are not an isolation boundary for coding agents"]]></title><description><![CDATA[
<p>Why a clone over a worktree? The obvious disadvantage I can see is you dont share your git objects with a clone and for really large repos that becomes an issue</p>
]]></description><pubDate>Thu, 30 Jul 2026 16:01:56 +0000</pubDate><link>https://news.ycombinator.com/item?id=49111903</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49111903</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49111903</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Launch HN: Prized (YC S26) – Let non-engineer staff build secure internal tools"]]></title><description><![CDATA[
<p>Demo looks cool, congrats on the ship! Probably missed but is it possible to externally share/view the built artifact outside of prized?</p>
]]></description><pubDate>Thu, 30 Jul 2026 15:29:24 +0000</pubDate><link>https://news.ycombinator.com/item?id=49111390</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=49111390</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49111390</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Qwen 3.8"]]></title><description><![CDATA[
<p>i'll take that 100% over llm slop.</p>
]]></description><pubDate>Mon, 20 Jul 2026 03:10:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=48973959</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48973959</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48973959</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Qwen 3.8"]]></title><description><![CDATA[
<p>what exactly do they do to "tune" it? It's not Deepseek's official harness, and going direct with deepseek using ur own harness is stupid cheap</p>
]]></description><pubDate>Sun, 19 Jul 2026 22:36:07 +0000</pubDate><link>https://news.ycombinator.com/item?id=48972311</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48972311</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48972311</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Qwen 3.8"]]></title><description><![CDATA[
<p>I use these guys. <a href="https://crof.ai/tos" rel="nofollow">https://crof.ai/tos</a>. Their prices match deepseek, pretty dang fast, and support ZDR. Maybe they serve quantized models but I havent seen a drop in quality from my evals using them vs Deepseek or even for other models.</p>
]]></description><pubDate>Sun, 19 Jul 2026 22:33:33 +0000</pubDate><link>https://news.ycombinator.com/item?id=48972281</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48972281</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48972281</guid></item><item><title><![CDATA[New comment by Bnjoroge in "G# – A modern .NET language with Go, Kotlin, and Swift ergonomics"]]></title><description><![CDATA[
<p>It’s a pretty safe assumption that most recent projects use AI in some capacity. Models have gotten pretty good</p>
]]></description><pubDate>Thu, 16 Jul 2026 03:49:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=48930138</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48930138</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48930138</guid></item><item><title><![CDATA[New comment by Bnjoroge in "TS-2026-009: Insecure argument handling in Tailscale SSH permitted root access"]]></title><description><![CDATA[
<p>Did you try Headscale? <a href="https://github.com/juanfont/headscale" rel="nofollow">https://github.com/juanfont/headscale</a> or netbird? The latter has been great for me.</p>
]]></description><pubDate>Wed, 15 Jul 2026 13:03:08 +0000</pubDate><link>https://news.ycombinator.com/item?id=48920231</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48920231</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48920231</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Show HN: Learn by rebuilding Redis, Git, a database from scratch"]]></title><description><![CDATA[
<p>I’ll take that over something AI generated that’s not free</p>
]]></description><pubDate>Sun, 12 Jul 2026 00:13:05 +0000</pubDate><link>https://news.ycombinator.com/item?id=48877065</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48877065</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48877065</guid></item><item><title><![CDATA[New comment by Bnjoroge in "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"]]></title><description><![CDATA[
<p>Yea I cant convince myself to pay api pricing unless my employer is doing it</p>
]]></description><pubDate>Thu, 09 Jul 2026 14:20:40 +0000</pubDate><link>https://news.ycombinator.com/item?id=48846341</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48846341</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48846341</guid></item><item><title><![CDATA[New comment by Bnjoroge in "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"]]></title><description><![CDATA[
<p>I’ve unfortunately had to temper my excitement with Cognition’s models/products given the amount of unwarranted hype they created with Devin on first release, but hopefully this is good.</p>
]]></description><pubDate>Wed, 08 Jul 2026 19:14:50 +0000</pubDate><link>https://news.ycombinator.com/item?id=48836181</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48836181</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48836181</guid></item><item><title><![CDATA[Show HN: Run multiple Docker Compose instances for your agents]]></title><description><![CDATA[
<p>Hey folks.<p>While running multiple agents and git worktrees with a Docker Compose setup, I kept running into host port, container name, and volume collisions. Before building this tool, I tried some combinations of -p to namespace the projects/hash the worktree path for a unique name, managing .env files with port variables and Caddy for reverse-proxying, but I still had some issues getting everything to play together automatically without me making any changes to my compose files or thinking about it. I also tried a bunch of tools that try to address this but you end up with a Docker-in-docker setup, or they requires you to rewrite your compose files.<p>Docktree solves these issues by dynamically generating Compose overrides without editing your compose file. It auto-allocates unique host ports, isolates networks and named volumes, and rewrites explicit container names so different branches of the same stack run simultaneously without manual tweaks. There's also an option where, for certain services like postgres/mysql that support "logical isolation" you can have only one database container, and run multiple databases in them isolated instead of isolated containers.<p>The CLI features a built-in reverse proxy that handles routing by hostname automatically (accessing apps via http://<worktree-name>.localhost), Cloudflare/ngrok sharing tunnels to expose worktrees externally. These are pretty useful since I can just have agents work on something end to end and return me a link I can use to investigate their work or they can send me screenshots/walkthroughs when combined with something like Vercel's agent-browser. Tool also has two skills and --json outputs for your agents to autonomously drive them. These are still containers at the end of day though so the existing docker tools   agents are familiar with work out of the box.<p>- Repository: <a href="https://github.com/Bnjoroge1/docktree" rel="nofollow">https://github.com/Bnjoroge1/docktree</a>
 - Website: <a href="https://docktree.dev" rel="nofollow">https://docktree.dev</a></p>
<hr>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=48767816">https://news.ycombinator.com/item?id=48767816</a></p>
<p>Points: 2</p>
<p># Comments: 0</p>
]]></description><pubDate>Thu, 02 Jul 2026 21:48:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=48767816</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48767816</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48767816</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Show HN: role-model, a router for hybrid local/cloud AI"]]></title><description><![CDATA[
<p>Interesting, any benchmarks for how well it works?</p>
]]></description><pubDate>Wed, 01 Jul 2026 01:37:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=48741424</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48741424</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48741424</guid></item><item><title><![CDATA[New comment by Bnjoroge in "Show HN: Smart model routing directly in Claude, Codex and Cursor"]]></title><description><![CDATA[
<p>Looks cool, but I couldnt find any benchmarks for how well this does. Any links?</p>
]]></description><pubDate>Sun, 28 Jun 2026 02:46:07 +0000</pubDate><link>https://news.ycombinator.com/item?id=48703869</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48703869</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48703869</guid></item><item><title><![CDATA[New comment by Bnjoroge in "The unbearable cheapness of open weight models"]]></title><description><![CDATA[
<p>It was a rough heuristic for how much more opus/5.5 would presumably cost if you extrapolate from glm5.2 prices. In any case input tokens for 5.4/4.6 are 70-100% more expensive, cached about 3-100%, and output tokens anywhere from 60%-240% more as per all their current api pricing. I highly doubt 5.4/4.6 are that much more expensive to serve given how cheap and commoditized inference has become, and how comparable they are perfomance wise.</p>
]]></description><pubDate>Fri, 26 Jun 2026 14:51:12 +0000</pubDate><link>https://news.ycombinator.com/item?id=48687337</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48687337</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48687337</guid></item><item><title><![CDATA[New comment by Bnjoroge in "The unbearable cheapness of open weight models"]]></title><description><![CDATA[
<p>You can make a fairly decent assumption by calculating the margin on serving glm 5.2, and adding say 30% extra costs and it still leaves a healthy margin</p>
]]></description><pubDate>Fri, 26 Jun 2026 13:35:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=48686498</link><dc:creator>Bnjoroge</dc:creator><comments>https://news.ycombinator.com/item?id=48686498</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48686498</guid></item></channel></rss>