Open-source · macOS-first · local + SSH

Taviraq

A safer AI terminal for people who live in shell.

Taviraq understands your terminal output, helps debug local and SSH sessions, and pauses before risky commands touch your shell. Bring OpenAI-compatible providers, Ollama, or LM Studio.

Real terminal sessions Agent safety gate BYO model
agent mode
Taviraq agent mode demo in a terminal window
Local PTY Use Taviraq as a normal terminal first.
System SSH Profiles use your keys, agents, config, and jump hosts.
Command risk model Risky or unclear commands require confirmation.
Keychain storage API keys stay in the macOS Keychain.

Core workflows

Turn terminal confusion into a safe next action.

01

Explain failing output

Select text, use recent output, or share the current session so the assistant can explain errors without a browser detour.

02

Get one next command

Ask for a focused next step, review what it does, then approve execution only when the command makes sense.

03

Troubleshoot over SSH

Diagnose remote shells with context, while destructive or ambiguous commands still stop at an in-app safety gate.

Safety and control

The assistant can help. You keep the shell.

Taviraq is designed around explicit context and approval. It does not need to be another chat window pretending to be a terminal.

Agent mode · review before execution
Proposed command
rm -rf dist build .cache
Needs confirmation Deletes local files. Review paths before running.

Provider-neutral

Connect OpenAI-compatible APIs, Ollama, or LM Studio. Use separate models for chat and command-risk classification.

Context-aware

Choose selected text, recent output, or the current session. The assistant sees what you decide to share.

Terminal-first

Tabs, searchable output, clickable links, themes, command snippets, prompts, and a compact assistant sidebar.

Local by default

Non-secret settings live in app data. Secrets stay in Keychain. Local models are one provider choice away.

Install

Download the latest macOS build or build from source.

Current release builds are unsigned test builds, so macOS may show an unidentified developer warning.

git clone https://github.com/Doka-NT/Taviraq.git
cd Taviraq
make build

Roadmap

Built toward safe AI-assisted operations.

  1. Trust readiness Signed builds, notarization, Homebrew cask, clearer privacy docs.
  2. Daily AI workflow Better context picker, quick actions, runbooks, model presets.
  3. SSH and DevOps wedge Read-only-first remote mode, protected command policies, audit history.
  4. Extensible control surface MCP, external CLI agent integrations, prompt packs, shared workflows.