DeepSeek TUI logo

DeepSeek TUI

Free

DeepSeek TUI - Terminal Coding Agent for DeepSeek V4 Models

Developer ToolsFreeFree tier
#deepseek#terminal#coding-agent#rust#tui#ai-agent#mcp#cli#code-editor#git
Inputs: textOutputs: text
Starting Price
Free
Type
Saas
DeepSeek TUI screenshot

About DeepSeek TUI

DeepSeek TUI, also known as CodeWhale, is a local-first, terminal-native coding agent designed for DeepSeek V4 and other open frontier models. Built in Rust, it provides a fast keyboard-driven Text User Interface (TUI) that gives the model direct access to the user's workspace. The agent can read and edit files, execute shell commands, search the web, manage git repositories, and orchestrate sub-agents, all within the terminal environment. It supports 1M-token context windows with intelligent compaction and a thinking-mode streaming feature to handle complex, long-running tasks. The project is hosted on GitHub and is available as a free, open-source tool for developers who prefer working in the command line.

Key Features

Native RLM (rlm_query) — fan out 1-16 cheap flash model calls in parallel for batched reasoning
Thinking-mode streaming — see DeepSeek's chain-of-thought as it reasons about code
1M-token context with intelligent automatic compaction
Three interaction modes: Plan (read-only), Agent (approval), YOLO (auto-approved)
Workspace rollback with side-git snapshots and turn-level revert
HTTP/SSE runtime API for headless agent workflows
MCP protocol support for external tools and extensions
Session save and resume for long checkpointed sessions
Live cost tracking with per-turn and session-level token estimates
Skill system with installable community skills from GitHub repos

Pros & Cons

Pros
  • Free and open-source, with full code transparency and community contributions
  • Terminal-native design offers fast, keyboard-driven interaction suited for experienced developers
  • Rust-based implementation likely provides good performance and low resource usage
  • Local-first approach allows operation without cloud dependency for core agent logic, enhancing privacy
  • Large context window (1M tokens) with compaction helps manage extended sessions and complex projects
Cons
  • Terminal-only interface may have a steep learning curve for developers not comfortable with CLI tools
  • Requires access to DeepSeek V4 or compatible models, which may involve API costs or self-hosting setup
  • Full feature set, including web search and sub-agent orchestration, may need additional configuration and API keys
  • Output quality and reliability depend on the underlying model and prompt design; results may vary
  • Not a managed SaaS — users must install, configure, and maintain the tool themselves

Best For

undefined: undefinedundefined: undefinedundefined: undefined

Alternatives to DeepSeek TUI

FAQ

What models does CodeWhale support?
CodeWhale supports any model from DeepSeek, Claude, GPT, Kimi, GLM, and 30+ providers. It also works with local models served via vLLM, SGLang, or Ollama.
Is CodeWhale free?
Yes, CodeWhale is completely free and open-source, licensed under MIT.
How do I install CodeWhale?
You can install it via npm (`npm install -g codewhale`), Cargo, Docker, Nix, Scoop, prebuilt archives, or Android/Termux. See docs/INSTALL.md for full details.
Can I use CodeWhale without a terminal?
While it is terminal-native, you can also use the local web client by running `codewhale web` for a browser-based UI.
Does CodeWhale support model switching?
Yes, you can switch models mid-task using the `/model` command in the TUI, which swaps both the provider and model.
Is CodeWhale related to DeepSeek TUI?
Yes, CodeWhale was formerly named DeepSeek TUI. Your existing config and sessions carry over automatically.