Bring your own model

An AI terminal for Ollama and LM Studio

Taviraq is provider-neutral. Point it at a local model running in Ollama or LM Studio and keep assistant traffic on your own machine — while still getting explicit context, command-risk checks, and approval before risky commands run.

Download for macOS View source

Local models, treated as a first-class provider

Connect OpenAI-compatible APIs, Ollama, or LM Studio. Taviraq lets you use separate models for chat and for command-risk classification, so you can run a capable local chat model and a smaller, fast model for risk scoring — or the same one for both.

Safety still applies with local models

The safety layer is independent of the provider. Agent mode checks built-in protected-command patterns and then asks the command-risk model before auto-execution. If classification fails or cannot be parsed, the command is treated as risky and requires confirmation. Risky or unclear commands pause in an in-app confirmation modal before they touch your shell.

Why a local-model terminal

Running models locally keeps prompts and terminal context on your hardware, avoids per-token API costs, and works offline. Combined with local secret masking and Keychain-stored keys, Taviraq is a practical choice when privacy and control matter.

Honest tradeoffs: Taviraq is macOS-first and early-stage. Local model quality depends on the model you run; smaller models may classify command risk less reliably, so keep approvals on for anything destructive.