silicato 0.1.0rc2

Linux push-to-talk voice bridge to terminal AI agents running in tmux.

Silicato

Silicato is a Linux voice bridge from push-to-talk microphone input to a selected AI-agent tmux pane.

Current release candidate: 0.1.0rc2.

Supported Platforms

• Official support: TUXEDO OS 24.04 LTS (Ubuntu noble base)
• Best effort: Ubuntu 24.04-compatible Linux environments

Validated baseline for this RC:

• OS: TUXEDO OS 24.04.4 LTS
• Python: 3.12.3
• tmux: 3.4
• Agent CLIs: Codex CLI 0.107.0 and Claude Code (manual validation)

Release Scope

0.1.0rc2 is intentionally limited to the current feature set:

• Push-to-talk transcription flow
• Optional preview mode (--preview) for send/edit/retry/skip/quit
• Required tmux + terminal agent CLI workflow
• Existing target resolution, config persistence, JSONL logging, and doctor diagnostics

Out of scope for this RC:

• Always-on mode
• Spoken replies (TTS)
• Non-Linux platform support

Capabilities

• Local speech capture via arecord
• Local transcription via faster-whisper
• Language selection: en, de, auto
• tmux pane target selection with persisted default target
• Normal mode: direct send after transcription
• Preview mode: review/edit/retry/skip/quit before send
• Runtime diagnostics via silicato --doctor
• Local JSONL turn logging

Limitations

• Linux and tmux are mandatory
• A terminal agent CLI (for example Codex or Claude Code) must run in a tmux pane
• No always-on listening mode
• No spoken output mode
• Turn logs may contain transcript text and stay local on disk

Install

Install from PyPI (0.1.0rc2)

pipx install silicato==0.1.0rc2

Source install

python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -U pip
make install-dev

Final release install (after 0.1.0 publish)

pipx install silicato

Maintainer-only TestPyPI validation steps live in docs/dev/release-rc.md.

Quick Start

Install required system packages:

sudo apt update
sudo apt install -y alsa-utils ffmpeg python3-venv tmux

Run diagnostics:

silicato --doctor

Run normal mode:

silicato

Run preview mode:

silicato --preview

Normal mode sends directly after transcription. Preview mode requires explicit confirmation.

Known Issues

Agent submit timing in tmux

Status:

• Silicato mitigates this by splitting text send and submit key send with a short delay.
• The issue has been observed with terminal agent UIs in tmux on Linux.

Symptom:

• Transcript text appears in the agent input but is not submitted until manual Enter.

Current mitigation in Silicato:

1. Send transcript text first.
2. Wait briefly.
3. Send Enter as a separate tmux operation.

If you still see this behavior, open a bug report and include:

• OS distribution and version
• terminal emulator
• tmux version
• Agent CLI + version (for example Codex CLI or Claude Code)
• exact reproduction steps

Commands

make install
make install-dev
make hooks
make test-arch
make check-rules
make test-rules-fast
make test-rules
make check
make test-fast
make test
make gate

Documentation

User docs:

• Quickstart
• Capabilities
• Dependencies
• Alpha Preview
• Feedback Template

Developer docs:

• Architecture
• Patterns Quickstart
• Dependency Rules
• Business Rules
• RC Release Runbook

Project docs:

• Changelog
• Contributing
• Security

License

MIT. See LICENSE.