deepseek

AI coding assistant using DeepSeek Coder models

TLDR

Start interactive mode

$ deepseek

Run a single prompt

$ deepseek chat "[write a Python function for binary search]"

Start with a specific model

$ deepseek --model [deepseek-coder:6.7b]

Set up the local environment (Ollama)

$ deepseek setup

Force the CLI to use local mode

$ deepseek --local

Display help

$ deepseek --help

deepseek is a command-line AI coding assistant that leverages DeepSeek Coder models to assist with code completion, generation, refactoring, debugging, and code review across many programming languages.It can run in two modes. The recommended local mode uses Ollama to run DeepSeek Coder models on your own machine, keeping code private and avoiding API costs. Cloud mode uses the DeepSeek API and requires an API key.Without arguments it launches an interactive session; the chat subcommand runs a single prompt and exits, which is convenient for scripting.

PARAMETERS

chat prompt

Send a single prompt and print the response.

setup

Set up the local Ollama environment and models.

--model model

Model to use (e.g., deepseek-coder:6.7b, deepseek-coder:1.3b).

--local

Force the CLI to use local Ollama models instead of the cloud API.

--help

Display help information.

CONFIGURATION

DEEPSEEK_API_KEY (environment variable)

API key used in cloud mode.

DEEPSEEK_USE_LOCAL (environment variable)

Set to true/false to select local or cloud mode.

CAVEATS

Local mode requires Ollama and a downloaded DeepSeek Coder model (sizes range from ~1GB to ~19GB). Cloud mode requires a DeepSeek API key and may incur costs. Installed via npm as the `run-deepseek-cli` package (requires Node.js 18+); this is a community project, not an official DeepSeek release.

RESOURCES

Source code