DeepSeekX

Chinese README: README.zh-CN.md

DeepSeekX is a downstream adaptation of OpenAI Codex CLI for DeepSeek API usage. It keeps the local terminal coding-agent workflow while defaulting to DeepSeek-oriented provider, model, packaging, and user-facing behavior.

The original upstream OpenAI Codex README is preserved as CODEX_README.md to reduce conflicts during future upstream syncs.

Install

Install from npm:

npm install -g @meomeo-dev/[email protected]

Run:

deepseekx

For one-off usage:

npx @meomeo-dev/[email protected]

Homebrew is not supported yet. GitHub Releases provide release notes and optional tarball downloads.

Quickstart

Set a DeepSeek API key and start DeepSeekX:

export DEEPSEEK_API_KEY="sk-..."
deepseekx

DeepSeekX uses its own user directory and does not read the original OpenAI Codex user directory by default.

Configuration

User-level config lives at DEEPSEEKX_HOME/config.toml. If DEEPSEEKX_HOME is not set, the default user directory is ~/.deepseekx, so the default file is:

~/.deepseekx/config.toml

Project-level config lives at the project root:

.deepseekx/config.toml

Use project config for repository-specific model, sandbox, approval, MCP, provider, or profile settings that should not become global user defaults.

Configuration precedence:

command-line overrides -> project .deepseekx/config.toml ->
user DEEPSEEKX_HOME/config.toml -> built-in defaults

User Config

Create ~/.deepseekx/config.toml:

model_provider = "deepseek"
model = "deepseek-v4-pro"
approval_policy = "on-request"
sandbox_mode = "workspace-write"

[model_providers.deepseek]
name = "DeepSeek"
base_url = "https://api.deepseek.com"
env_key = "DEEPSEEK_API_KEY"
wire_api = "chat"
requires_openai_auth = false

For a custom DeepSeek-compatible provider, keep the provider id in the deepseek-* namespace:

model_provider = "deepseek-vendor-a"
model = "deepseek-v4-flash"

[model_providers.deepseek-vendor-a]
name = "Vendor A DeepSeek"
base_url = "https://vendor.example/v1"
env_key = "VENDOR_DEEPSEEK_API_KEY"
wire_api = "chat"
requires_openai_auth = false

Project Config

Create .deepseekx/config.toml in a repository:

model = "deepseek-v4-flash"
approval_policy = "on-request"
sandbox_mode = "workspace-write"

[profiles.pro]
model_provider = "deepseek"
model = "deepseek-v4-pro"
approval_policy = "on-request"
sandbox_mode = "workspace-write"

[profiles.flash]
model_provider = "deepseek"
model = "deepseek-v4-flash"
approval_policy = "on-request"
sandbox_mode = "workspace-write"

Run with a profile:

deepseekx --profile pro

Models

DeepSeekX currently exposes:

• deepseek-v4-pro
• deepseek-v4-flash

DeepSeek model metadata, base instructions, and personality templates are maintained inside the DeepSeek provider catalog rather than loaded from the OpenAI model catalog at runtime.

Differences From OpenAI Codex

• The CLI command is deepseekx.
• The npm package is @meomeo-dev/deepseekx.
• User config defaults to ~/.deepseekx.
• Project config defaults to .deepseekx/config.toml.
• DeepSeek auth uses DEEPSEEK_API_KEY.
• OpenAI login and unsupported IDE/app commands are hidden or disabled.
• Some upstream Codex features may stay unavailable until they are adapted for DeepSeekX.

Release Channels

• npm is the primary install channel.
• GitHub Releases provide release notes and optional tarballs.
• Homebrew is not currently supported.

License

This repository is licensed under the Apache-2.0 License.

DeepSeekX is a downstream fork of OpenAI Codex. See CODEX_README.md for the preserved upstream README.