Deepseek Tui
Coding agent for DeepSeek models that runs in your terminal
Last verified:
What is Deepseek Tui?
DeepSeek TUI is an open-source command-line coding agent that lives in your terminal and is built on the DeepSeek V4 family. It is designed to help with real development work rather than casual chat: it can read files, edit code, run shell commands, call MCP servers, search the web, and report back with multi-step tool use.
The tool is positioned as a terminal-native assistant for people who want an agentic workflow directly inside their shell. The website compares it to other coding agents in the same category, emphasizing a loop of reading, acting, testing, and iterating instead of one-shot answers. It also highlights sandboxing and approval controls so users can keep risky actions gated.
A major part of the product is its mode system. Plan mode is read-only, Agent mode is the default multi-step mode that asks for approval on risky actions, and YOLO mode auto-approves everything and lifts the workspace boundary when trust is enabled. The docs also describe separate approval settings, sandboxing support on macOS, Linux, and Windows, and a configurable interface for workflow control.
It is aimed at developers, power users, and teams that want an AI coding assistant embedded in the terminal, especially users already comfortable with shells, git, and local workspaces. The site also shows support for multiple providers, skills, hooks, sub-agents, recursive LMs, and MCP integrations, which makes it suitable for users who want a highly configurable and extensible coding environment.
Deepseek Tui pricing
Pricing model: Freemium
The website presents DeepSeek TUI as free and open source under the MIT license, with no paid tiers or subscription plans listed on the site. The homepage shows an install command using cargo, says the first run sets up ~/.deepseek/, and notes that other installation methods exist if you do not have Rust. Pricing details for the underlying DeepSeek API or any third-party providers are not described on the site itself.
Deepseek Tui pros
- Runs directly in the terminal
- Open-source project
- Built on DeepSeek V4 family
- Edits files in place
- Runs shell commands
- Supports MCP servers
- Has read-only Plan mode
- Has default Agent mode for multi-step work
- Has YOLO mode for full automation
- Approval system for risky actions
- Sandbox support on macOS
- Sandbox support on Linux
- Restricted-token support on Windows
- Supports 9 built-in providers
- Can switch providers from config or command line
- Supports web search and URL fetching
- Supports git diagnostics and testing tools
- Has sub-agents for persistent parallel sessions
- Supports recursive LM / Python REPL workflows
- Offers hooks for session and tool events
Deepseek Tui cons
- Requires Rust 1.88+ for the main install path
- Terminal-first interface may feel complex for beginners
- Risky shell actions still need approval in normal use
- YOLO mode removes workspace boundaries
- Can be too powerful for simple chat-only tasks
- Depends on API keys and provider setup
- Feature set may be overwhelming for casual users
- Sandbox and trust settings require careful configuration
- Some operations rely on local MCP server setup
- Best fit is coding and development work, not general-purpose consumer use
Frequently asked questions about Deepseek Tui
What is DeepSeek TUI?
DeepSeek TUI is an open-source terminal coding agent built on the DeepSeek V4 family. It is meant to act like a real coding assistant in the shell: it can read and edit files, run commands, search, test, and work through multi-step tasks instead of only answering questions.
How does it work inside the terminal?
You interact with it from a TUI, and the agent runs a loop that sends requests to DeepSeek models while using tools such as file reading, editing, shell execution, web search, and MCP calls. The site describes it as a coding agent with approval prompts and sandbox protections around potentially dangerous actions.
What modes does it support?
It has three operating modes: Plan, Agent, and YOLO. Plan is read-only and can inspect files and URLs without writing or shelling out; Agent is the default multi-step mode; YOLO auto-approves actions and enables trust mode, so the workspace boundary is lifted.
What tools can it use?
The docs list file tools, search tools, shell tools, git and diagnostics tools, sub-agents, recursive LM tools, and MCP tools. It can read and write files, grep code, run tests, inspect git status and diffs, open parallel sessions, and access auto-registered MCP tools.
Does it support sandboxing?
Yes. The website says it respects sandbox boundaries and supports seatbelt on macOS, landlock on Linux, and restricted tokens or AppContainer on Windows. The default workspace boundary can also be lifted for a session using trust settings.
Which providers does it support?
The docs show 9 built-in providers, including DeepSeek, NVIDIA NIM, OpenAI, OpenRouter, Novita, Fireworks, sglang, vLLM, and Ollama. You can switch providers with a config command, and each provider uses its own API key environment variable.
How do approvals work?
Approvals and modes are separate. In the default suggested approval setup, the agent asks before risky operations; auto mode approves all tool calls; and never mode blocks anything that is not safe or read-only.
Can it use MCP servers?
Yes. The site says DeepSeek TUI speaks the Model Context Protocol both as a client and as a server. It can load MCP servers from ~/.deepseek/mcp.json, and server tools appear as mcp_<server>_<tool> in the tool surface.
How do I install it?
The homepage shows a cargo install command for the CLI and then launching it with deepseek. It also says the first run creates the ~/.deepseek/ directory and points users to other installation methods if they do not have Rust.
Who is it for?
It is mainly for developers and power users who want an AI coding agent inside the terminal. The product is especially suited to people comfortable with shells, git, local workspaces, configuration files, and approval-based automation.