zara

Zara is a hybrid symbolic AI assistant platform designed as a client/server architecture. It utilizes a durable runtime (zara-server) that manages conversations, memory, and tool execution, allowing multiple native clients—including a PySide6 Linux desktop app, a CLI, and an Android application—to interact via the versioned ZARA/1 protocol over ZeroMQ. Key technical capabilities include: - Symbolic Reasoning: Uses SWI-Prolog (desktop) and Trealla (Android) for deterministic intent handling and capability decisions. - LLM Integration: Supports providers such as OpenAI, Anthropic, and Ollama for open-ended conversation and reasoning. - Voice Interface: Implements a split architecture where clients handle audio I/O while the server manages VAD, STT, TTS, and turn coordination. - Plugin System: Provides a bounded runtime API for extending functionality through plugins and MCP integration. - Native UI: Features a Linux desktop copilot with system tray integration, Markdown rendering, and an optional "desktop pet" subsystem that reacts to runtime events. The project is developed primarily using Nix and is positioned as a research effort to replace vague LLM-generated machinery with explicit protocols and symbolic rules.