Sparking Embers.
Find a file
2026-05-06 21:21:57 -04:00
src Imnprove Todos UI. 2026-05-06 21:21:57 -04:00
.gitignore first commit 2026-05-02 00:08:27 -04:00
Cargo.toml Add transcription. 2026-05-05 22:46:25 -04:00
README.md Add transcription. 2026-05-05 22:46:25 -04:00

Catalyst Guard

A native desktop AI harness built in Rust. Catalyst Guard provides an integrated workspace for running AI agents, managing tasks, taking notes, chatting with multiple providers, and transcribing audio — all from a single offline-capable application.

Features

Screen Description
Home Dashboard and quick navigation
Todos Kanban board for task tracking (canvas-based, built for zoom/pan)
Agents Define and configure AI agents with per-agent tool access and artifact directories
Notes Markdown vault with wiki-link support, file tree, and live preview
Chat Multi-provider streaming chat with native function-calling and image attachments
File Explorer 3-pane Finder-style explorer scoped to agent artifact directories
Transcriptions Audio recording and transcription via cloud APIs or on-device whisper.cpp
Settings Provider credentials, model selection, and transcription configuration

Supported AI Providers

  • Anthropic (Claude)
  • OpenAI (GPT / Whisper)
  • Lemonade (local LLM server)
  • Ollama (local model runner)

Agent Tools

Agents can be granted access to:

  • Filesystem — sandboxed read/write within the agent's artifact directory
  • FilesystemBroad — broader filesystem access (with deny-list sandbox)
  • OfficeFiles — generate .xlsx and .docx files from markdown
  • WebSearch, CodeExecution, Terminal — planned

Building

Requirements

  • Rust toolchain (stable, 1.75+)
  • C++ compiler and cmake (required only for local-whisper feature)

Default build (cloud transcription only)

cargo build --release

The binary is placed at target/release/catalyst-guard.

With on-device transcription (CPU)

Enables whisper.cpp via whisper-rs. Requires cmake and a C++ compiler.

cargo build --release --features local-whisper

With NVIDIA GPU acceleration

Requires the CUDA toolkit in addition to the above.

cargo build --release --features cuda

With Vulkan GPU acceleration

Requires the Vulkan SDK and a compatible driver.

cargo build --release --features vulkan

macOS (Apple Silicon)

Metal acceleration is enabled automatically when local-whisper is active — no extra flag needed.

cargo build --release --features local-whisper

Data Locations

Platform Config Database
Linux / macOS ~/.config/catalyst-guard/config.toml ~/.local/share/catalyst-guard/todos.db
Windows %APPDATA%\catalyst-guard\config.toml %LOCALAPPDATA%\catalyst-guard\todos.db

The config file is created automatically on first run with sensible defaults.

Whisper Models

When using on-device transcription, models are downloaded through Settings → Transcription. Models are stored in the application data directory and are not bundled with the binary.