How to Configure DeepSeek-Reasonix: Build, Install, and Run Setup Guide

Configure DeepSeek-Reasonix by compiling the static Go binary with make build, generating a TOML configuration file via reasonix setup, and setting your LLM API key in the environment variable specified by the api_key_env field.

DeepSeek-Reasonix is a Go-based AI coding assistant that drives one or two LLM models via a static, CGO-free binary. To configure DeepSeek-Reasonix for your workflow, you must understand its layered configuration architecture defined in internal/config/*, which merges settings from command flags, project-level TOML files, user preferences, and system defaults.

Building the Static Binary

The build process produces a single static executable named reasonix that can be cross-compiled for all major operating systems and architectures.

Start by cloning the repository and navigating to the main source directory:

git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix/main-v2

The Makefile wraps the go build command with CGO_ENABLED=0 and injects version information via ldflags. Compile the binary locally:

make build

This generates ./bin/reasonix and ./bin/reasonix-plugin-example. For cross-platform distribution across Darwin, Linux, and Windows (both amd64 and arm64), run:

make cross

Output binaries are placed in ./dist/reasonix-<os>-<arch>[.exe].

Install the binary system-wide on Linux or macOS:

sudo install -Dm755 bin/reasonix /usr/local/bin/reasonix

Configuration Architecture and Files

DeepSeek-Reasonix uses a TOML-based configuration system implemented in internal/config/* that follows a hierarchical merge order: command-line flags override project-level settings, which override user-level settings, which finally override defaults.

The configuration manages:

• Provider settings (internal/provider/openai/openai.go): OpenAI-compatible wire protocols for DeepSeek, MiniMax, Anthropic, and other services
• Tool registry (internal/tool/registry.go): Built-in tools (file I/O, bash, grep) and external MCP plugins
• Permission policies (internal/permission/*): Per-call allow/ask/deny policies
• Sandbox constraints (internal/sandbox/*): Filesystem confinement rules

Initial Setup and TOML Configuration

Run the first-time wizard to generate a minimal reasonix.toml in your project root:

reasonix setup

This creates a configuration file following the structure defined in the source examples:

default_model = "deepseek-flash"

[[providers]]
name        = "deepseek-flash"
kind        = "openai"
base_url    = "https://api.deepseek.com"
model       = "deepseek-v4-flash"
api_key_env = "DEEPSEEK_API_KEY"

The api_key_env field specifies which environment variable contains the API key. Export this in your shell:

export DEEPSEEK_API_KEY=sk-<your-key>

Alternatively, the wizard can store credentials in Reasonix's home .env file.

Configuring the Agent and Two-Model Collaboration

The Agent layer, described in docs/SPEC.md under sections 3.4 Agent and 3.5 Two-model collaboration, handles the LLM chat loop and tool calls. The entry point in cmd/reasonix/main.go delegates to internal/cli/cli.go for session orchestration.

To use a specific model for a single task, bypass the default configuration:

reasonix run --model deepseek-pro "add unit tests for this function"

Running DeepSeek-Reasonix

Start an interactive session where the TUI prompts you to run /init to generate AGENTS.md (project memory):

reasonix

For non-interactive execution, pipe commands or use direct arguments:

echo "explain this code" | reasonix run
reasonix run "implement the TODOs in main.go"

The Desktop UI built with Bubble Tea in desktop/* automatically detects whether it is running in a terminal or graphical environment. Launch the same binary in either context:

reasonix  # Adapts to terminal vs. desktop automatically

Permission and Sandbox Configuration

Configure security policies through the TOML configuration to govern tool execution:

• Permission policies (internal/permission/*): Define per-tool approval levels (allow, ask, deny)
• Sandbox constraints (internal/sandbox/*): Restrict filesystem access to specific directories

These settings are loaded during the initialization phase in internal/cli/cli.go and enforced for each tool call registered in internal/tool/registry.go.

Summary

• Build the static binary using make build or make cross for distribution
• Configure providers in reasonix.toml with name, kind, base_url, model, and api_key_env fields
• Authenticate by exporting the API key environment variable specified in your provider configuration
• Initialize projects with reasonix setup and /init to generate AGENTS.md
• Run interactively with reasonix, non-interactively with reasonix run, or in desktop mode via the Bubble Tea UI in desktop/*

Frequently Asked Questions

Where is the DeepSeek-Reasonix configuration file stored?

The configuration follows a hierarchy: command-line flags take precedence over ./reasonix.toml (project-level), which overrides user-level settings, which finally fall back to defaults. Run reasonix setup to generate the initial project configuration file.

How do I configure multiple LLM providers in DeepSeek-Reasonix?

Add multiple [[providers]] blocks to your reasonix.toml file. Each provider requires name, kind (set to "openai" for OpenAI-compatible APIs), base_url, model, and api_key_env fields. Reference different providers at runtime using the --model flag.

Can I run DeepSeek-Reasonix without installing it system-wide?

Yes. After running make build, execute the binary directly from ./bin/reasonix without installing to /usr/local/bin. The binary is statically linked with CGO_ENABLED=0, requiring no external dependencies or system installation.

How do I configure permissions for tool execution?

Permission policies are defined in internal/permission/* and configured via the TOML settings. You can set per-tool policies to allow, ask, or deny to control whether the agent can execute file operations, bash commands, or other registered tools from internal/tool/registry.go without prompting.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}