UnitaryLab CLI User Manual

The UnitaryLab CLI is UnitaryLab’s local terminal command-line tool. Running the unitarylab command in your terminal opens an interactive conversation environment (REPL), where the agent assists you with model conversations, file reading and writing, and invocation of built-in skills such as quantum algorithms and PDE solvers.

This manual is written for users who interact with the CLI directly. It covers the full workflow, from startup and day-to-day conversation to skill invocation, session management, and troubleshooting. If you are using the web product, please refer to the Unitary Studio Web User Manual.

Intended Audience

• Developers who prefer working in the terminal and want to handle day-to-day conversation and file collaboration from the command line.
• Researchers who need to keep conversation history and working directory structure on their local machine.
• Engineers who need to embed the UnitaryLab agent into local scripts and workflows.
• Anyone else with a strong interest in quantum computing and CLI tools.

Key Capabilities

• Local interactive conversation: Engage in multi-turn conversations with the agent in your terminal. Conversation history is persisted locally.
• Multiple conversation modes: Three modes — agent (full tool chain), ask (retrieval-based Q&A), and plan (planning-oriented) — to fit different scenarios.
• Multi-model support: Compatible with DashScope, OpenAI, and any OpenAI-compatible service (such as SiliconFlow, a local vLLM instance, and others). You can add multiple models during configuration and switch between them at any time.
• Tool-call approval: Before the agent invokes an operational tool, a confirmation dialog appears with a clear description. Approve or deny with a single key. Approval cache is intelligently scoped.
• Streaming status and cancellation: The current task phase is displayed in real time. Press Esc to cancel the current inference at any time.
• Sessions bound to working directories: Each session automatically records its most recently used working directory, so resuming a session puts you right back where you left off. Sessions can be deleted or cleared as needed.
• Skill invocation: Use /<skill_name> commands to drive the agent to load built-in skills such as quantum algorithms, PDE solvers, and code workspaces. Supports multi-level directory browsing, Tab completion, and skill detail inspection.
• Local account login: Sign in with your account and password. Your login state is persisted locally, and the CLI automatically prompts you to log in again when your credentials expire.

Chapter Navigation

• Quick Start: From zero to your first conversation.
• Account Guide: Logging in, logging out, and checking your login status.
• Model Configuration Guide: Configuring the LLM provider, API Base, API Key, and model list.
• Interactive Conversation Guide: In-session prompts, conversation modes, runtime switching, and file-open operations.
• Session and Working Directory Guide: Switching sessions, working directories, and local history.
• Skill Invocation Guide: How to invoke built-in skills.
• Troubleshooting Guide: Common issues and diagnostic approaches.

Recommendations

• First-time users: Follow the Quick Start guide to complete configuration and login before reading the other chapters.
• Already conversing successfully: Consult the relevant chapters as needed. The in-session /help command is always the fastest way to see the available commands.
• Running into problems: Start with the Troubleshooting Guide. You can also describe the symptoms directly to the agent in a conversation, and it will attempt to diagnose the issue on its own.