A personal lexicon for voice-to-agents. It maps what speech-to-text actually writes (Ashler, Cooper Nettie's, sass) back to what you meant (Ashlr.AI, Kubernetes, SaaS) before an agent sees the prompt. This page runs the real matcher in your browser.
Dictation uses your browser's own speech recognizer. In Chrome that is a cloud service: it streams your microphone audio to Google. The transcript it returns is then corrected here, on this page.
Pick one. Each takes about a minute. Ordinary correction stays on the machine. lexicon voice downloads a whisper model the first time, lexicon pack fetches a registry URL you chose, and the GitHub importer calls the GitHub API through your existing gh login.
Then run lexicon setup. It seeds your lexicon with your name and company, harvests the current repo, registers the MCP server and hooks in Claude Code, Codex, Cursor and the other clients it finds, and exports to your dictation app. Run from a terminal, the script runs it for you; under sudo or without a terminal it only installs.
From the latest release. Checksums are in SHA256SUMS next to them.
Extensions load unpacked from chrome://extensions or about:debugging.
LexiconBar is ad-hoc signed: right-click, Open on first launch.
Walkthrough in docs/QUICKSTART.md.
If a name came out right, star the repo. That is how the next person who dictates a company name finds it.
One YAML file of the words you say. It rewrites a transcript before an agent reads it. It is not a dictation app.
curl -fsSL https://ashlrai.github.io/lexicon/install.sh | sh, then lexicon setup --dry-run --json. The script runs setup only in a terminal. Under sudo, without a terminal, or with LEXICON_NO_SETUP=1, it prints run: lexicon setup.
No. The matcher runs in your browser. Dictation is the exception: Chrome's speech recognizer sends audio to Google. The corrected text stays on the page.
The longer answers are on lexicon.ashlr.ai and in llms-full.txt.