Settings & Model Catalog
Configure local GGUF weights, interactive `/config` settings, and standalone vocabulary tokenizers.
Configuration Directory Layout
Tacet stores settings, session logs, prompt history, and MCP definitions in the platform config directory:
- Unix / macOS:
$XDG_CONFIG_HOME/tacetor~/.tacetor~/.config/tacet - Windows:
%APPDATA%\Tacet - Configuration File:
config.toml(contains default model package, theme settings, search provider) - MCP Config:
mcp.json(contains registered MCP server definitions)
Standalone GGUF Vocabulary Tokenizer
`tacet-engine` includes a custom GGUF tokenizer that reads embedded vocabulary tensors directly from `.gguf` weight files without requiring external `tokenizer.json` files. This allows Tacet to load single-file model downloads seamlessly.
Verified Default Catalog & Hardware Inference Floor
Before attempting model execution, Tacet performs binary capability checks to verify CPU vector extensions (AVX2, AVX512, NEON) and GPU acceleration backends (Metal, CUDA), preventing trade-offs of working engines for invalid downloads.