Skip to content
caiolombelloPublic

About

Experimental Linux recordings, bounded AI context and notes linked to their source.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

FalaTrace

FalaTrace

Meeting notes for Linux that point back to the recording. You type short notes during the meeting. After it, the AI completes them with what was said, and every line it adds cites the moment of the recording it came from. Recording, transcription and notes run on your computer by default. FalaTrace is an experimental source alpha (0.2.0-alpha.20); see evidence and limits.

FalaTrace Studio with a synthetic meeting: the sidebar with Home, Actions, People, Clients and the meetings by day; the meeting's notes in the middle, with your lines, the lines the AI completed and a citation to the moment of each

The Studio's meeting page, with synthetic data: your notes and the AI's, each AI line citing its moment.

Install

You need Linux, Bun 1.4, FFmpeg and Git. Local processing also needs whisper.cpp and Ollama, installed separately. On Ubuntu:

git clone https://github.com/caiolombello/falatrace.git && cd falatrace
bun install --frozen-lockfile --ignore-scripts
make install-cli
sudo apt install qt6-base-dev qt6-base-dev-tools qt6-declarative-dev libmpvqt-dev
make install-studio

Open FalaTrace Studio from the application menu, or run recording-studio. The setup assistant's quick start picks local processing in one step and offers a one-minute test recording at the end; your first notes arrive a few minutes after it. Nothing records, uploads or downloads until you ask. Step by step: quickstart (português).

What it does

  • Notes and templates. A notepad next to the recording keeps the time of each line. The AI turns your notes into structured notes that follow a template (1:1, stand-up, postmortem, sales discovery and more, or your own). Your lines stay yours; AI lines are marked and cite the transcript.
  • Ask, with citations. Ask about one meeting, a person, a client or the whole library. Answers cite the moments they come from. Use local Ollama, an API key (OpenAI, OpenRouter, Anthropic, Gemini and others), an ACP agent, your own subscription CLI (claude, codex, gemini, opencode) or Sign in with ChatGPT.
  • People, clients and actions. People from confirmed speaker names and calendar guests, clients from your catalog, and the action items of every meeting in one list you can tick off.
  • Reminders and a brief. A notification shortly before a calendar meeting, with what earlier meetings with the same people left open, and a button to record or skip it.
  • Share. Send a meeting's notes to Obsidian, Notion, Slack or an AI app, after a preview.
  • Live transcript. An optional preview of the transcript while the meeting runs.
  • Retention. Keep the media, keep only the audio, or keep only the transcript once it is verified.
  • Review and export. Correct the transcript and speaker names as revisions; the original stays. Export JSON, Markdown, SRT or WebVTT.

Details: notes and templates, assistant, people, clients and actions, reminders and live transcript, sharing.

Privacy

  • Local by default. A new configuration transcribes with Whisper.cpp and writes notes with Ollama on this computer. Nothing is processed automatically until you turn it on.
  • You see every destination first. Processing, questions and shares you start name the destination before anything leaves the computer, and ask for your consent when it does. A setting that picks an external provider applies to every recording until you change it.
  • Your data stays in private files. Notes, conversations, people and action states are files only your user can read. Deleting a recording does not yet remove all of them; see privacy and data.
  • Record with permission. Confirm that the people in the call agree. Automatic call detection is opt-in and does not establish consent.

Agents. Local agents such as Claude Code, Codex and Gemini CLI can read bounded, cited context through the CLI (any processed recording, like any program you run) or a local MCP server (only the recordings you grant): skill, agent access.

Documentation

Topic Document
First use Quickstart · Início rápido
Notes, templates, summaries NOTES-AND-TEMPLATES.md
Asking questions, providers, recipes ASSISTANT.md
People, clients, action items PEOPLE-CLIENTS-ACTIONS.md
Reminders, brief, live transcript REMINDERS-AND-LIVE.md
Obsidian, Notion, Slack, AI apps SHARING.md
Where data lives, what leaves, deletion PRIVACY.md
Every setting CONFIGURATION.md · example · Studio settings
What is verified and what is not EVIDENCE-AND-LIMITS.md · compatibility
Review and export HUMAN-REVIEW-EXPORT.md
History and plans CHANGELOG.md · roadmap
Building and testing DEVELOPMENT.md · security

The landing page is in Portuguese and English. The Studio follows the system language (Portuguese or English).

Development

bun run typecheck && bun run desktop:check && bun run lint
bash scripts/test-offline.sh
bun run build

Tests run offline with a synthetic HOME; never plain bun test. The Studio journeys, release gates and harness rules are in DEVELOPMENT.md.

License

MIT for code by Caio Barbieri (LICENSE); third-party terms in THIRD-PARTY-NOTICES.md. The executable is falatrace; configuration, state and service names keep the older recording-cli identifiers (compatibility).

About

Experimental Linux recordings, bounded AI context and notes linked to their source.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages