Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/t0msilver/localvoxtral/agents-mdgit clone --depth 1 https://github.com/T0mSIlver/localvoxtralWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/t0msilver/localvoxtral/agents-md)<a href="https://agentmods.dev/instructions/t0msilver/localvoxtral/agents-md"><img src="https://agentmods.dev/badge/instructions/t0msilver/localvoxtral/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.02917 | $0.02917 |
| Opus 5 | $0.01458 | $0.01458 |
| Sonnet 5 | $0.00583 | $0.00583 |
| Haiku 4.5 | $0.00292 | $0.00292 |
Grade A, and why
localvoxtral AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 193 lines — stays where its author put it; the contents beside it link to each section on GitHub.
localvoxtral — agent guide
Native macOS menu bar app for realtime dictation (Swift 6.2 strict concurrency,
SwiftPM, macOS 15+). Streams mic audio to an OpenAI Realtime-compatible backend
(the bundled localvoxtral-speechd helper in managed mode; vLLM or any
compatible server in External URL mode), merges partial transcripts, and
inserts text into the focused app — either live ("Live Auto-Paste") or via an
overlay committed on stop ("Overlay Buffer", supports replacement dictionary +
LLM polishing). Subsystem map: docs/architecture.md.
Build & test — read this first on a non-Mac dev box
This repo only compiles on macOS. From a Linux box, the inner loop is the Mac
build host over SSH (no commit needed — it rsyncs the working tree). The host
is machine-local config, set once per clone (never committed):
git config localvoxtral.buildhost <ssh-destination>.
./scripts/remote-build.sh # build + unit tests
./scripts/remote-build.sh test --filter TextMergingAlgorithmsTests
./scripts/remote-build.sh integration # realtime pipeline vs the live speechd STT service
./scripts/remote-build.sh eval-llm # default polish prompt eval vs a live chat/completions server
./scripts/remote-build.sh package # build the .app bundle (also builds both MLX helpers)
./scripts/remote-build.sh integration-polishd [hf-repo] # bundled polish helper vs real model + eval baseline (run package first)
./scripts/remote-build.sh integration-speechd [hf-repo] # packaged speech helper vs real audio/model (run package first)
./scripts/remote-build.sh integration-herdr [ssh-dest] # remote-herdr join vs a LIVE herdr over a real ssh -L (hermetic by default)
./scripts/remote-build.sh eval-e2e [EvalRecordings/agent-dictation/<set>] # agent-dictation E2E eval (run package first)
./scripts/remote-build.sh dogfood # build the instrumented tree + run the context-capture suite
./scripts/remote-build.sh dogfood-package # package an instrumented .app for hand-dogfooding
./scripts/remote-build.sh build --package-path PolishHelper # compile a helper package alone (no Metal kernels)
./scripts/remote-build.sh test --package-path PolishHelper # helper unit tests (Metal-free); same for SpeechHelper
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 193 lines · 2,917 tokens per session scan A e58d9c2afe5c
localvoxtral AGENTS.md is an instructions file published in the GitHub repository T0mSIlver/localvoxtral (54 stars, last pushed 4d ago), licensed MIT. It adds 2,917 tokens to every session, about $0.0146 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
OpenSuperMLX AGENTS.md
AGENTS.md instructions for axot/OpenSuperMLX, covering agents.md — opensupermlx, agent behavioral guidelines, 1. think before coding, 2. simplicity first and 3. surgical changes.
OpenSuperMLX CLAUDE.md
Claude Code instructions for axot/OpenSuperMLX, a project described as: macOS app for real-time audio transcription powered by MLX on Apple Silicon.
speech-swift AGENTS.md
AGENTS.md instructions for soniqo/speech-swift, covering agent instructions, communication style, workflow, running tests and benchmarks — sequential and memory-aware and git conventions.
Vocello AGENTS.md
AGENTS.md instructions for PowerBeef/Vocello, covering agents.md — vocello (qwenvoice), product and authority, start and resume work, hard invariants and domain routing.
Logue CLAUDE.md
Instructions for bitwize-ai/Logue, covering logue — development guidelines, project overview, running the build you just made, where documentation goes and dependencies.
NarrateClaude CLAUDE.md
Instructions for nicedreamzapp/NarrateClaude, covering narrate mode — always on, core rule, how to narrate, voice and this is a general-purpose session.