←
Terp21,545 posts
Terp avatar

Terp

@OnlyTerp

Be kind to others, everyone deserves it

◷ Joined July 2009▧ 2,300 Photos and videos
Terp avatar
Terp@OnlyTerp·Nov 13

Your AI coding agent is slow because it keeps waiting on itself. Search, wait, read, wait, edit, wait, test, wait. So I built Turnpack. It folds those steps together. Half the round trips, roughly half the time, 36% cheaper in my test. It also lets the AI test its edits before they stick. If something breaks, every file snaps back to how it was. Free and open source. Works with Claude Code, Codex, Cursor, Gemini CLI, Grok, anything MCP :) github.com/OnlyTerp/turnpack

0:48
Terp avatar
Terp@OnlyTerp·14h

Tell me why new zealand gets gta 6 in 3 and half hours from now but i gotta wait 20 more hours while trying to dodge spoilers.... rockstar why wouldn't we globally sync this roll out are you insane?

Terp avatar
Terp@OnlyTerp·19h

Day 2 of 30 Days of Terp’s Tech: Connect Anything If you’re running a ton of agents across different harnesses and subscriptions, this is for you. I built Connect Anything so any agent (on any harness, on any sub you already pay for) can spawn and work with agents on any other harness. Local agents are great until your machine becomes the bottleneck. Then you’re stuck manually jumping between tools, copy-pasting answers, and trying to keep everything in sync. Not anymore. Your agents can now spin up cloud or local agents against any of your existing subscriptions. Want to throw a bunch of them at a big task and spread the usage across everything you already pay for? Do it. Free and open source: github.com/OnlyTerp/connect-… Day 3 tomorrow. Keep the requests coming.

0:23
Terp avatar
Terp@OnlyTerp·Nov 17

I have been a fan of the cognition team for a long time, anyone I talk to in the space will tell you I have told them to try devin out. This is no different, give them a shot, they provide so much value & the team is full of amazing people

Meet Devin Bots. From writing firmware to merging thousands of PRs a week, a Devin Bot leads a team of agents to get the job done.

1:33
Terp avatar
Terp@OnlyTerp·Nov 16

Day 1 of 30 Days of Terp’s Tech: ADHD Your AI has ADHD. Give it a hard problem and it grabs the first idea and runs. ADHD makes it stay on the problem, and the whole topic around it, for as long as it takes. No tools, no browsing: just raw reasoning, math and code, round after round. A second model keeps a live notebook, and every claim links back to the round that made it. This technique has really helped me in my own projects with solving complex issues that models typically struggle or drift with. Works with the models you already pay for: Claude, Codex, Devin, or any OpenAI-compatible endpoint. MCP server + skill for Claude Code, Codex and other MCP apps. It opens inside your harness built in browser so you don't have to open anything externally to use it :) Free and open source: Day 2 tomorrow. Keep the requests coming.

0:45
Terp avatar
Terp@OnlyTerp·Nov 16

Morning yapp

3:05
Terp avatar
Terp@OnlyTerp·Nov 16

Apperently qwen is best ai lawyer in the world? Qwen law coming soon 🤣

Post attachment

📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: qwen.ai/blog?id=qwen4-max - API: qwencloud.com/models/qwen4-m… - Token Plan: qwencloud.com/pricing/token-… More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀

Post attachment
Terp avatar
Terp@OnlyTerp·Nov 15

What do you want built? For the next 30 days I’m dropping a new tool, tip, or guide every single day: 30 Days of Terp’s Tech Reply with requests. I’ll pick some, build them for free, and post them open source Plus a backlog of goodies I’ve been sitting on for way too long

Terp avatar
Terp@OnlyTerp·Nov 15

Fixed my sleep schedule, so morning yaps are back 😌

2:53
Terp avatar
Terp@OnlyTerp·Nov 14

2:39
Terp@OnlyTerpNov 13

Your AI coding agent is slow because it keeps waiting on itself. Search, wait, read, wait, edit, wait, test, wait. So I built Turnpack. It folds those steps together. Half the round trips, roughly half the time, 36% cheaper in my test. It also lets the AI test its edits before they stick. If something breaks, every file snaps back to how it was. Free and open source. Works with Claude Code, Codex, Cursor, Gemini CLI, Grok, anything MCP :) github.com/OnlyTerp/turnpack

0:48
Terp avatar
Terp@OnlyTerp·Nov 13

How I got those numbers: I gave Claude Code the same small coding task 3 times with Turnpack and 3 times without, then compared. The test script is in the repo if you want to run it on your own setup. Credits: Video by fable 5.2 XH in claude app

Terp avatar
Terp@OnlyTerp·Nov 13

I just got back online now, not sure what happened

Terp avatar
Terp@OnlyTerp·Nov 13

Claude is down. Reports coming everywhere

Post attachment
Terp avatar
Terp@OnlyTerp·Nov 13

“Garbage speeds compared to anything else on the market” from 4 runs of one code review. Hydrogen 84.5 tok/s vs Sol 34.0 (OpenRouter) • Latency: 1.53s vs 7.47s • Artificial Analysis has Hydrogen on the time-vs-intelligence Pareto frontier • Hydrogen (high) scores 51 in 6.0 min/task. Sol (xhigh) scores 52 in 9.9. Same intelligence, 40% less time • Hallucination rate: 8% vs 52% Sol can win a single coding task in one harness. That’s a result about that task and that setup, not “the market.” Speed only means something next to what the model does with the time. The frustrating part is Theo knows better. You don’t accidentally leave out thinking level, output length, and quality on a speed chart. Rage bait pays better than nuance, I get it.

Post attachmentPost attachmentPost attachment

“Built for speed” Still garbage speeds compared to anything else on the market. Note that I ran Sol and Opus at regular tier! Didn’t even use fast/ultrafast.

Post attachment