IdeaWell

Project ideas from Hacker News discussions.

Show HN: TERMy – A fast terminal assistant that does not use LLMs

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Deterministic output vs. LLM unpredictability
Many commenters value TERMy’s claimed determinism for safety and trustworthiness, contrasting it with the inherent variability of LLMs.
- “I love the fact that the output is (at least claimed to be) deterministic. I can't trust an LLM to do the right thing after I deploy it to production, because their output is non‑deterministic by design.” – gurjeet
- “It isn’t … by design … If you do greedy decoding … it will be 100% always deterministic.” – kouteiheika
- “Sure, every time you prompt ‘what is the weather in kansas’ you’ll get the same output, but … ‘what is the weather in kansas right now’ … gets a different output … there are infinite variations in what the llm can output.” – kennywinker

2. Speed, low resource usage, and privacy
The tool’s ability to run instantly on a CPU (no GPU needed) is highlighted as a major advantage over larger LLMs.
- “I really like it … gives the user a win on privacy and speed.” – mpalmer
- “TERMy is more or less a calculator that knows english :) so it can run on your CPU and respond instantly!” – gioscarab
- “Models like … are not fast and reliable enough … if you have only 4GB of VRAM.” – kouteiheika
- “I tried functiongemma … it is for sure faster … the problem is that is not reliable enough for a terminal assistant.” – gioscarab

3. Rule‑based/template matching vs. LLM fallback / self‑learning
Discussion centers on TERMy’s deterministic, preset‑response model and whether augmenting it with LLMs for unseen queries (or self‑updating) would improve usefulness while preserving safety.
- “TERMy … can only answer predetermined responses … makes it safe and predictable for a use case like a terminal assistant.” – gioscarab
- “Have you considered making TERMy leverage an LLM for unseen or low‑confidence queries, and then generate the config and update itself to make future similar queries deterministic?” – mbil
- “It would make sense to have this integrate with a self‑learning routine … writes NPC‑Forge recipes … without an LLM.” – dmos62
- “I think that TERMy's design is now feasible because of the availability of LLMs. They make the dataset development feasible.” – gioscarab

🚀 Project Ideas

ForgeLearn: Auto‑update NPC‑Forge patterns from terminal usage

Summary

  • A background agent that records successful command executions, extracts natural‑language → command mappings, and automatically generates or refines NPC‑Forge pattern files for TERMy.
  • Core value proposition: turns ad‑hoc terminal assistance into a self‑improving, deterministic tool that grows smarter without sacrificing speed or privacy.

Details

Key Value
Target Audience Developers and sysadmins who use TERMy or similar NL‑to‑shell assistants and want the tool to learn from their workflow
Core Feature Continuous observation of terminal sessions, confidence‑scored pattern generation, PR‑style submission to a shared NPC‑Forge repo
Tech Stack Go (for low‑overhead hooking), SQLite for local logging, Jinja2‑style templating for NPC files, GitHub Actions for CI validation
Difficulty Medium
Monetization Hobby

Notes

  • HN users praised determinism: “I love the fact that the output is (at least claimed to be) deterministic.” – gurjeet, and expressed interest in self‑learning: “It would make sense to have this integrate with a self‑learning routine… ” – dmos62.
  • Enables community‑driven dataset growth while keeping the assistant fast and safe, addressing the frustration that “even if it fails only 1 time out of 10 becomes useless.” – gioscarab.

TermInstant: Zero‑latency NL → Shell via deterministic trie

Summary

  • A standalone command‑lookup service that stores a curated, community‑maintained trie of natural‑language phrases to shell commands (with argument placeholders) for instant, deterministic retrieval.
  • Core value proposition: provides sub‑millisecond response times and zero hallucination risk by replacing LLM inference with a pure data‑structure lookup.

Details

Key Value
Target Audience Power users who need immediate, reliable command suggestions (e.g., DevOps engineers, CLI enthusiasts)
Core Feature Trie‑based phrase matching with fallback to user‑defined aliases; supports optional arguments and flags
Tech Stack Rust (for fast trie), WASM plugin for terminal integration, optional SQLite persistence for user overrides
Difficulty Low
Monetization Hobby

Notes

  • Commenters highlighted speed and privacy as key advantages: “Using very colloquial terms TERMy is more or less a calculator that knows english :) so it can run on your CPU and respond instantly!” – gioscarab.
  • Solves the pain of slow LLMs: “Models like … are not fast and reliable enough for general use…” – kouteiheika, while preserving deterministic output prized by HN users.

ShellGuard: Sandboxed dry‑run and approval layer for NL‑to‑shell tools

Summary

  • A lightweight wrapper that intercepts suggested commands from any NL‑to‑shell assistant (TERMy, Warp AI, etc.), runs them in an isolated, read‑only sandbox or as a dry‑run, shows the expected effect, and requires explicit user approval before execution.
  • Core value proposition: eliminates the fear of accidental damage from assistant mistakes, making deterministic or probabilistic assistants safe for production use.

Details

Key Value
Target Audience Teams and individuals who use NL‑to‑shell assistants in scripts or production environments and need safety guarantees
Core Feature Sandbox execution (using bubblewrap or gVisor), diff‑preview of filesystem changes, mandatory approval prompt
Tech Stack Go (wrapper), bubblewrap for Linux sandbox, optional macOS sandbox via sandbox-exec, configurable approval UI (CLI or tmux popup)
Difficulty Medium
Monetization Revenue‑ready: “Per‑seat SaaS $5/mo for team audit logs”

Notes

  • HN discussion emphasized risk: “Even if it fails only 1 time out of 10 becomes useless.” – gioscarab, and desire for safety: “I can't trust an LLM to do the right thing after I deploy it to production…” – gurjeet.
  • Provides practical utility by letting users adopt NL assistants without fear, encouraging broader adoption and discussion around safe AI‑assisted CLI workflows.

Read Later