Have an LLM write your ticket-triage rules once, then run them locally for free
STROT uses an LLM only once to compile your intent, such as "triage support tickets into the right workflow", into a small local decision model. After that every record is classified, scored or routed on your own hardware as typed JSON with a confidence score, and low-confidence records abstain instead of guessing.
Source
solohunter001 on Hacker News · 2026-10-19 · Project link
Written in our own words. Credit to the original author. Link checked 2026-10-19.
What they did
STROT uses an LLM only once to compile your intent, such as "triage support tickets into the right workflow", into a small local decision model. After that every record is classified, scored or routed on your own hardware as typed JSON with a confidence score, and low-confidence records abstain instead of guessing.
Result (as claimed)
No per-record API cost after the first compile; MIT licensed with 106 passing tests.
Tools they used
Cost: Claude Opus 5.5: from $4 in / $20 out per 1M tokens.
Set it up
Map the task
Write the trigger, the inputs, the decisions and the output. Note which steps need a human yes.
Connect the tools
Set up Python, Claude Opus 5.5 with the narrowest permissions that work. Use a test account or sandbox first.
Build the workflow
Create the steps in order. Add an approval step before anything sends, pays or deletes.
Run it on real but low-stakes data
Run a few times, read every output, and fix prompts or rules that misfire.
Schedule and monitor
Turn on the schedule or trigger, and add a daily summary or alert so you notice failures.
Copy the full setup prompt
Paste into your agent and it will walk you through the setup, asking before anything that sends, posts or spends.
You are helping me set up this AI use case end to end in Claude Code. Goal: Have an LLM write your ticket-triage rules once, then run them locally for free What the original author did: STROT uses an LLM only once to compile your intent, such as "triage support tickets into the right workflow", into a small local decision model. After that every record is classified, scored or routed on your own hardware as typed JSON with a confidence score, and low-confidence records abstain instead of guessing. Reported result (as claimed by the author): No per-record API cost after the first compile; MIT licensed with 106 passing tests. Source: https://news.ycombinator.com/item?id=50148396 Tools they used: Python, Claude Opus 5.5 Category: Automation & business Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Codex. Goal: Have an LLM write your ticket-triage rules once, then run them locally for free What the original author did: STROT uses an LLM only once to compile your intent, such as "triage support tickets into the right workflow", into a small local decision model. After that every record is classified, scored or routed on your own hardware as typed JSON with a confidence score, and low-confidence records abstain instead of guessing. Reported result (as claimed by the author): No per-record API cost after the first compile; MIT licensed with 106 passing tests. Source: https://news.ycombinator.com/item?id=50148396 Tools they used: Python, Claude Opus 5.5 Category: Automation & business Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Grok Bot. Goal: Have an LLM write your ticket-triage rules once, then run them locally for free What the original author did: STROT uses an LLM only once to compile your intent, such as "triage support tickets into the right workflow", into a small local decision model. After that every record is classified, scored or routed on your own hardware as typed JSON with a confidence score, and low-confidence records abstain instead of guessing. Reported result (as claimed by the author): No per-record API cost after the first compile; MIT licensed with 106 passing tests. Source: https://news.ycombinator.com/item?id=50148396 Tools they used: Python, Claude Opus 5.5 Category: Automation & business Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
The Daily 5
Get five new use cases tomorrow morning
Free. Real sources. Unsubscribe any time.
Related use cases
Bridge Claude Code to Grok Bot through a webhook routine
With no direct API between them, the author found one door after reading 37 doc pages: a Grok Bot routine with a webhook trigger. They wrapped it in 670 lines of dependency-free Python with 56 tests; sends are idempotent and results come back as files.
Have an LLM write HTML/CSS graphics and burn them into video without a browser
Geneva is a Rust take on the ffmpeg CLI with a native HTML/CSS compositor. Prompt your agent for an intro or animation (the author finds Opus 5.5 good at this), let it write the HTML/CSS and render it onto video in one pass, faster than browser-based tools.
Make a short animated film as one HTML file with a critic agent reviewing every scene
A creator had Claude Code with Opus 5.5 build a short motion-graphics film entirely as an 85 KB HTML file with sound and no assets or libraries. A separate critic agent reviewed every scene three times before it was kept.
Turn your kids' vacation idea into a game with Claude
After a visit to a military airbase his kids asked to make a game of it, so a dad used leftover Claude credits to build HALO A400, a high-altitude jump game with stunts and even a parachute malfunction. Everything is procedural and made with Claude Opus 5.5.