# DeepSeek CLI
A terminal-native TUI and CLI for [DeepSeek](https://platform.deepseek.com) models, built in Rust.
[](https://github.com/Hmbown/DeepSeek-TUI/actions/workflows/ci.yml)
[](https://crates.io/crates/deepseek-tui)
For DeepSeek models (current and future model IDs). Not affiliated with DeepSeek Inc.
## What is this
A terminal-native agent loop that gives DeepSeek the tools it needs to actually write code: file editing, shell execution, web search, git operations, task tracking, and MCP server integration. Coherence-aware memory compaction keeps long sessions on track without blowing up the context window.
Three modes:
- **Plan** — design-first, proposes before acting
- **Agent** — multi-step autonomous tool use
- **YOLO** — full auto-approve, no guardrails
Sub-agent orchestration is in there too (background workers, parallel tool calls). Still shaking out the rough edges.
## Install
```bash
# From crates.io (requires Rust 1.85+)
cargo install deepseek-tui --locked
# Or from source
git clone https://github.com/Hmbown/DeepSeek-TUI.git
cd DeepSeek-TUI && cargo install --path . --locked
```
## Setup
Create `~/.deepseek/config.toml`:
```toml
api_key = "YOUR_DEEPSEEK_API_KEY"
```
Then run:
```bash
deepseek
```
**Tab** switches modes, **F1** opens help, **Esc** cancels a running request.
## Usage
```bash
deepseek # interactive TUI
deepseek -p "explain this in 2 sentences" # one-shot prompt
deepseek --yolo # agent mode, all tools auto-approved
deepseek doctor # check your setup
```
## Model IDs
Common model IDs: `deepseek-chat`, `deepseek-reasoner`.
Any valid `deepseek-*` model ID is accepted (including future releases). To see live IDs from your configured endpoint:
```bash
deepseek models
```
## Configuration
Everything lives in `~/.deepseek/config.toml`. See [config.example.toml](config.example.toml) for the full set of options.
Environment overrides: `DEEPSEEK_API_KEY`, `DEEPSEEK_BASE_URL`.
## Docs
Detailed docs are in the [docs/](docs/) folder — architecture, modes, MCP integration, runtime API, etc.
## License
MIT