deepx-code CLI
A terminal coding agent in a single Go binary for DeepSeek and other OpenAI-compatible models, with a code graph and local OCR
- 392 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent writes files and executes commands; auto mode performs these actions automatically
- The repository’s install script modifies the binary and the user’s shell profile
- Stores API keys in ~/.deepx/model.yaml and sends code and prompts to the chosen model provider
Install
Manual install
curl -fsSL https://raw.githubusercontent.com/itmisx/deepx-code/main/scripts/install.sh | bash && exec $SHELLFor macOS and Linux. A PowerShell variant exists for Windows, and deepx upgrade updates the binary.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent writes files and executes commands; auto mode performs these actions automatically
- The repository’s install script modifies the binary and the user’s shell profile
- Stores API keys in ~/.deepx/model.yaml and sends code and prompts to the chosen model provider
This is third-party code. Review the repository files before installing.
What it does
deepx-code is a terminal agent for working with code. It installs as one Go binary with no Node or Python runtime, starts with the deepx command in a project folder and opens an interactive interface. DeepSeek, Xiaomi MiMo, Kimi and Qwen come preconfigured, and any other OpenAI-compatible model can be added by hand. A router helps select a model for each turn: flash for simpler requests and pro for harder ones. A model can also be pinned manually. A built-in symbol graph handles code navigation: definitions, callers, interface implementations and the blast radius of an edit. A local PaddleOCR engine extracts screenshot text without a multimodal model. It does not interpret charts, colors or interface layout. In `review` mode, file writes and shell commands require confirmation. `auto` runs them automatically, while `plan` disables these tools. It supports MCP and Claude skill directories, multi-agent plans and reusable JS workflows. The deepx exec command runs a single task without the interface, so it fits scripts and CI.
Who it is for. For developers who want a terminal agent with a choice of OpenAI-compatible APIs, a symbol graph and screenshot text extraction.
Good fit when
- You use DeepSeek through its API and want to use prefix caching to reduce repeated context processing
- You need an agent that finds callers and interface implementations in a large repository through a symbol graph without grepping files
- You need to read text from an error screenshot without a multimodal model, which the local OCR covers
- The same multi-step procedure has to run on a schedule or in CI, which deepx exec and the JS workflows are for
- You already keep skills in a Claude directory and want to reuse them in another agent
Not a fit when
- You need Anthropic models: the agent targets OpenAI-compatible APIs
- You need to work offline: model calls use an API, and routing and OCR models require a prior download
- You want an agent with a graphical window: deepx lives in the terminal
- The job is small and one-off: setting up a provider and a key takes longer than the edit
Example request
Find every place that calls the order handling function and show what changing its signature would affect.Limitations
The provider key is stored in `~/.deepx/model.yaml`, and API usage is billed separately. Routing and OCR models require additional downloads. Until the router is available, each turn starts with flash; local OCR does not make the agent independent of its model API.
The `native` sandbox uses OS mechanisms where available. It keeps network access and most reads, and allows writes to the project, temporary directories and several caches. If isolation is unavailable, it may fall back to a limited blacklist. The installer updates the shell profile’s executable path. The selected model provider receives prompts, code and tool results.
How to disable. Delete the ~/.local/bin/deepx binary. Settings, keys and sessions live in ~/.deepx and can be removed separately when no longer needed.
FAQ
How do I install deepx-code?
On macOS and Linux run the install script from the repository through curl and bash, on Windows through PowerShell. The binary lands in ~/.local/bin/deepx and updates with deepx upgrade. A Gitee mirror exists.
Which models does deepx-code support?
DeepSeek (deepseek-v4-flash and deepseek-v4-pro), Xiaomi MiMo, Kimi and Qwen come preset. The /provider command switches between them, and any model with an OpenAI-compatible API can be set in ~/.deepx/model.yaml.
Can I reuse my Claude Code skills?
Yes. The skills directory can be the same as Claude's, for example ~/.claude/skills. Project skills go into the .deepx/skills folder.
How does deepx-code differ from Claude Code?
It is a separate open-source agent under the MIT license that runs on DeepSeek, MiMo, Kimi and Qwen models. The code graph and local OCR are built in, and MCP and skills are supported.
How do I run a task without the interface?
Use deepx exec with the task text. The result prints to stdout and data can be piped in, for example cat error.log | deepx exec with a question. The key must be configured in interactive mode first.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal coding agent tuned for the deepseek-v4 model: deep thinking, reasoning effort control, Agent Skills and MCP
An open-source TypeScript terminal coding agent tuned for DeepSeek models and cheap caching
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
Terminal coding agent in Go for DeepSeek, Kimi, GLM and OpenAI, built around prefix caching and confirmed writes and commands
Lightweight Python terminal coding agent: ReAct loop, file-based context, guarded bash and skills, works with DeepSeek and OpenAI
A coding agent with a Python core and three front-ends: a script CLI, a terminal TUI and an Electron app, for OpenAI-style APIs
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The code uses MIT. The prebuilt binary needs neither Node.js nor Python; source builds require Go 1.25.8 according to `go.mod`. Session history is stored under `~/.deepx/sessions/`, with separate model and routing settings.