Costa Terminal
One Terminal. Every Model.
A native AI terminal that routes queries across local and cloud providers using an ML classifier. Built with Rust and Tauri for zero-overhead performance. Currently in development.
Every Provider, One Interface
The ML router picks the best model for each query. Local models handle simple tasks for free. Cloud models handle the rest. You never have to choose manually.
Local models (Qwen, Llama, Gemma, Apriel, GPT-OSS)
Free 70B inference at 840 tok/s
Free Flash tier, math and reasoning
Free Devstral for code tasks
Full Claude Code sessions with tool visibility
Built for Local-First AI
7 route classes
ML Query Router
A pre-trained ONNX classifier routes every query to the optimal provider. Local models handle system tasks for free. Complex reasoning escalates to cloud. The router retrains on your usage patterns.
Full tool visibility
Claude Code Sessions
Full Claude Code integration with persistent session IDs. See every tool call, file edit, and bash command in a live activity panel. Cost tracking per session.
Auto-detection
Hardware-Aware Onboarding
A 4-step wizard detects your GPU, VRAM, installed Ollama models, and API keys. Suggests optimal models based on your hardware. Pulls missing models directly.
Native performance
Streaming Responses
Native Tauri events stream tokens from any provider with consistent latency. No Electron, no web overhead. Rust backend with direct HTTP/2 connections.
DB-driven config
Provider Management
Add, remove, and configure providers from the settings panel. Enable/disable individual models. Set tier priorities. All stored in a local SQLite database.
18+ workers
Worker Pool Discovery
On startup, the app discovers all available providers and builds a worker pool. 10+ local workers from Ollama, 8+ cloud workers from API keys. Concurrent inference across all of them.
Tech Stack
Rust + Tauri v2
SolidJS + Tailwind
ONNX (ort crate)
Tauri Events (NDJSON)
SQLite (rusqlite)
CLI subprocess + session IDs
Part of the Costa OS Ecosystem
Costa Terminal is a standalone app that works on any Linux system with Ollama. It ships as a core component of Costa OS, where it integrates with the MCP server, voice pipeline, and agent system.
Learn about Costa OS