Hacker News: jahala

New comment by jahala in "Show HN: Kyushu – A self-hostable WASM sandbox for JavaScript workers"

jahala — Sun, 07 Jun 2026 13:05:04 +0000

Thanks for that!

New comment by jahala in "Show HN: Kyushu – A self-hostable WASM sandbox for JavaScript workers"

jahala — Sun, 07 Jun 2026 12:47:54 +0000

Loving the customer testimonials :D ..

If someone feels like an eli5 - What are the use-cases for something like this?

New comment by jahala in "Show HN: Lowfat – pluggable CLI filter that saved 91.8% of my LLM tokens"

jahala — Sat, 06 Jun 2026 08:18:28 +0000

Yup, this is hitting it on the nose. But, despite the cost - the benchmark is the vital ingredient that cant be skipped. Otherwise, you don't know if what you're building is actually helping the agent rather than hindering it.

On the previous large benchmark run, i proved 40-50% cost reduction per correct answer.

I'm not sure why the vendors aren't using token filtering/compression more in their tooling, but perhaps they don't mind users feeding them more data and using more data.

New comment by jahala in "Show HN: Lowfat – pluggable CLI filter that saved 91.8% of my LLM tokens"

jahala — Sat, 06 Jun 2026 08:14:59 +0000

No I don't have the funds to benchmark the competition, but would be happy to put the numbers up if any token whales feel like having a go.

https://github.com/jahala/tilth/tree/main/benchmark

New comment by jahala in "Show HN: Lowfat – pluggable CLI filter that saved 91.8% of my LLM tokens"

jahala — Fri, 05 Jun 2026 21:08:18 +0000

This is the reason, when I built a tool in the same space, I chose to benchmark with cost per correct answer.

Reducing tokens and also turns is quite worthless if the LLM doesn’t solve what you put it to do.

New comment by jahala in "Does anyone else find Hacker News visually exhausting?"

jahala — Mon, 25 May 2026 11:42:47 +0000

Nope.

New comment by jahala in "Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep"

jahala — Sun, 17 May 2026 22:26:05 +0000

This looks great! I built a tool in the same space- and I found that the biggest challenge was often to get the agent to prefer to use the tool over bash tools. What’s your experience with that?

New comment by jahala in "Accelerando (2005)"

jahala — Sat, 16 May 2026 14:12:54 +0000

I absolutely LOVE Accelerando. I've recommended it to everyone I meet for years.

If you're looking for other great sci-fi reads:

John Ringo - Live free or die

John Varley - Titan (-> Wizard / Demon)

Charles Stross - Singularity Sky

Vernor Vinge - A Fire Upon the Deep / A Deepness in the Sky

Robert Heinlein - Stranger in a Strange Land

Dan Simmons - Hyperion

Alastair Reynolds - Revelation Space / The Prefect

Orson Scott Card - Enders game

Isaac Asimov - Foundation

New comment by jahala in "LLM Wiki – example of an "idea file""

jahala — Sun, 05 Apr 2026 08:42:52 +0000

I did a proof of concept for self-updating html files (polyglot bash/html) some weeks ago. It actually works quite well, with simple prompting it seems to not just go in circles (https://github.com/jahala/o-o)

New comment by jahala in "We replaced RAG with a virtual filesystem for our AI documentation assistant"

jahala — Sat, 04 Apr 2026 20:48:29 +0000

I built tilth (https://github.com/jahala/tilth) much for this reason. Couldn't bother with RAG, but the agents kept using too many tokens - and too many turns - for finding what it needed. So I combined ripgrep and tree-sitter and some fiddly bits, and now agents find things faster and with ~40% less token use (benchmarked).

New comment by jahala in "Show HN: Walkie-Clawkie – Push-to-talk between AI agents, one file, zero deps"

jahala — Thu, 02 Apr 2026 12:17:21 +0000

I got tired of copy-pasting between agents for simple coordination. Everything I found was a framework or a hosted service. I just needed them to talk.

Walkie-Clawkie is a single JS file, zero dependencies. It’s an MCP server that gives agents walkie_send and walkie_agents. Same machine: file mailboxes. Cross machine: HTTP relay you expose however you want. Unknown agents need human approval before they can get through.

https://github.com/jahala/walkie-clawkie

Show HN: Walkie-Clawkie – Push-to-talk between AI agents, one file, zero deps

jahala — Thu, 02 Apr 2026 12:16:53 +0000

Article URL: https://github.com/jahala/walkie-clawkie

Comments URL: https://news.ycombinator.com/item?id=47613418

Points: 1

# Comments: 1

New comment by jahala in "Show HN: Tilth v0.5.0 –> ~40% cheaper AI code navigation (160 runs, 3 models)"

jahala — Tue, 10 Mar 2026 16:50:54 +0000

Hey @joknoll - in the benchmarks, I'm seeing very positive results with Haiku, getting quicker and more correct answers. So I think you're absolutely right that harness improvements will be a natural part of "sharpening" most models - especially the smaller ones with less reasoning capability.

New comment by jahala in "Ask HN: What Are You Working On? (March 2026)"

jahala — Mon, 09 Mar 2026 12:34:54 +0000

Have been tinkering with a few different projects:

- Tilth: Smart(er) code reading for humans and AI agents. Reduces LLM token use and cost by ~40% (benchmarked) https://github.com/jahala/tilth

- mrkd: Native macOS .md viewer (+preview in Finder) that imports iTerm2 and VSCode themes https://github.com/jahala/mrkd

- O-O: Self-updating articles concept with polyglot (bash/html) files. No server, no database. https://github.com/jahala/o-o

Show HN: Tilth v0.5.0 –> ~40% cheaper AI code navigation (160 runs, 3 models)

jahala — Sun, 08 Mar 2026 17:35:07 +0000

Smart code reading for humans and AI agents. Tilth is what happens when you give ripgrep, tree-sitter, and cat a shared brain.

—

v0.5.0 was about figuring out why models weren’t using tilth tools consistently — even when they were available.

Results vs baseline (built-in tools only):

Sonnet 4.6: -44% $/correct (84% → 94% accuracy, 31% fewer turns)

Opus 4.6: -39% $/correct (91% → 92% accuracy, 37% fewer turns)

Haiku 4.5: -38% $/correct (54% → 73% accuracy, 7% fewer turns)

—

https://github.com/jahala/tilth/

Full results: https://github.com/jahala/tilth/blob/main/benchmark/README.m...

— PS: I don't have the budget to run the benchmark a lot (especially with Opus), so if any token whales has capacity to run some benchmarks, please feel free to PR results.

Comments URL: https://news.ycombinator.com/item?id=47299191

Points: 4

# Comments: 2

New comment by jahala in "Claude Code told me what tools it needs to work faster"

jahala — Fri, 06 Mar 2026 00:00:35 +0000

I made tilth much for the same reasons. Its MIT licence, open source and saves you 30% token use (benchmarked)

https://github.com/jahala/tilth

New comment by jahala in "Show HN: Reduce LLM token use by ~30% with this MCP/CLI tool(Claude benchmarked)"

jahala — Thu, 05 Mar 2026 16:03:21 +0000

All contributions are welcome, especially more benchmarks for other models!

Show HN: Reduce LLM token use by ~30% with this MCP/CLI tool(Claude benchmarked)

jahala — Thu, 05 Mar 2026 15:59:32 +0000

Smart code reading for humans and AI agents. Tilth is what happens when you give ripgrep, tree-sitter, and cat a shared brain. --

v0.4.4: Added adaptive 2nd-hop impact analysis to callers search — when a function has ≤10 unique callers, tilth automatically traces callers-of-callers in a single scan. First full 26-task Opus baseline (previously 5 hard tasks only). Haiku adoption improved from 42% to 78%, flipping Haiku from a cost regression to -38% $/correct.

v0.4.5: Bumped TOKEN_THRESHOLD from 3500 to 6000 estimated tokens (~24KB), so mid-sized files return full content instead of an outline that agents then read back via 5–7 sequential --section calls. Fixed two major regressions: gin_radix_tree (+35% → ~tie) and rg_search_dispatch (+90% → -26% win). Sonnet hit 100% accuracy (52/52) and -34% $/correct overall.

https://github.com/jahala/tilth/

Full results: https://github.com/jahala/tilth/blob/main/benchmark/README.m...

-- PS: I dont have the budget to run the benchmark a lot (especially with Opus), so if any token whales has capacity to run some benchmarks, please feel free to PR results.

Comments URL: https://news.ycombinator.com/item?id=47263220

Points: 2

# Comments: 1

New comment by jahala in "Buddhist concept of suffering explained with AI generated video and music (HQ)"

jahala — Tue, 03 Mar 2026 23:30:05 +0000

Amazed by what can now be produced by a single (?) individual working with AI.

Buddhist concept of suffering explained with AI generated video and music (HQ)

jahala — Tue, 03 Mar 2026 23:30:05 +0000

Article URL: https://www.youtube.com/watch?v=Vyq0kbjFjdA

Comments URL: https://news.ycombinator.com/item?id=47240635

Points: 1

# Comments: 1