dcode

dcode is an AI-powered terminal coding agent built in Go with a full Bubbletea TUI. It supports 20+ AI providers, autonomous tool use, multi-step undo/redo, live thinking display, and a rich @ file-picker — all without leaving your terminal.

Features

• 20+ AI providers — Anthropic, OpenAI, Google Gemini, GitHub Copilot, Azure OpenAI, AWS Bedrock, Groq, OpenRouter, xAI, DeepSeek, Mistral, DeepInfra, Cerebras, Together AI, Cohere, Perplexity, GitLab, Cloudflare, Replicate, and any OpenAI-compatible endpoint
• Autonomous tool use — reads, writes, edits files; searches with glob/grep; executes shell commands; browses the web; calls MCP servers
• Rich TUI — Bubbletea interface with Glamour-rendered markdown, Chroma syntax highlighting, scrollable history, and a light-green blinking cursor
• Thinking display — live streaming of extended-thinking / reasoning with opencode-style spinner and topic extraction
• Multi-step undo/redo — Ctrl+Z / Ctrl+Shift+Z backed by git snapshots; also /undo and /redo slash commands
• Copy code blocks — Ctrl+Y + digit copies the Nth code block from the last response; /copy N also works
• @ file picker — type @ to attach any file (text, images, or directories); text files are injected as fenced code blocks, images as inline data
• Session persistence — all conversations saved as JSON with import/export and share
• MCP support — connects to local and remote Model Context Protocol servers
• Git worktrees — dcode worktree create <name> for isolated parallel sessions on separate branches
• Non-interactive mode — dcode run "prompt" for scripting and CI/CD pipelines
• Shell completions — bash, zsh, fish, and PowerShell via dcode completion

Installation

From a release (recommended)

Download the binary for your platform from the Releases page and put it on your PATH:

# Linux (x86_64)
curl -L https://github.com/Dhanuzh/dcode/releases/latest/download/dcode_linux_x86_64.tar.gz | tar xz
sudo mv dcode /usr/local/bin/

# macOS (Apple Silicon)
curl -L https://github.com/Dhanuzh/dcode/releases/latest/download/dcode_darwin_arm64.tar.gz | tar xz
sudo mv dcode /usr/local/bin/

Via go install

go install github.com/Dhanuzh/dcode/cmd/dcode@latest

Build from source

git clone https://github.com/Dhanuzh/dcode.git
cd dcode
make build        # builds ./dcode with version/commit ldflags
make install      # copies to /usr/local/bin (requires sudo)

Quick Start

# First run — dcode guides you through authentication
dcode

# Or authenticate explicitly
dcode auth login

# Non-interactive one-shot prompt
dcode run "explain the main function in cmd/dcode/main.go"

# Use a specific provider and model
dcode --provider openai --model gpt-4o

Authentication

dcode supports three credential sources, in priority order:

1. Stored credentials (recommended) — dcode auth login saves keys to ~/.config/dcode/credentials.json with 0600 permissions
2. Environment variables — e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_API_KEY
3. Config file — ~/.config/dcode/dcode.json or project-local .dcode/dcode.json

Provider-specific OAuth flows:

dcode auth anthropic   # Anthropic OAuth PKCE flow (no API key needed)
dcode auth copilot     # GitHub Copilot device-code OAuth
dcode auth openai      # API key prompt

List configured providers:

dcode auth list

Configuration

Global config lives at ~/.config/dcode/dcode.json. Override per-project with .dcode/dcode.json in your repo root.

# View current resolved config
dcode config show

# Change settings
dcode config set provider anthropic
dcode config set model claude-sonnet-4-5
dcode config set theme dracula

| Key | Default | Description | |---|---|---| | provider | anthropic | Default AI provider | | model | (provider default) | Model ID | | default_agent | coder | Agent to use (coder, planner, explorer) | | max_tokens | 8192 | Max output tokens | | streaming | true | Stream responses | | theme | dark | TUI colour theme | | snapshot | true | Enable git snapshots for undo |

TUI Key Bindings

| Key | Action | |---|---| | Enter | Send message | | Ctrl+C | Cancel running request | | Ctrl+Z | Undo last AI action | | Ctrl+Shift+Z | Redo | | Ctrl+Y + digit | Copy Nth code block | | @ | Open file picker (attach file or image) | | Ctrl+L | Clear screen | | Ctrl+M | Toggle mouse mode | | Tab | Cycle through sessions / autocomplete | | ↑ / ↓ | Scroll history | | /help | List slash commands | | /undo, /redo | Undo / redo | | /copy N | Copy Nth code block | | /clear | Clear conversation | | /model | Switch model interactively | | /agent | Switch agent |

Available Commands

dcode                    # Start the TUI (default)
dcode run <prompt>       # Non-interactive single prompt
dcode auth               # Manage credentials
dcode config             # View/edit configuration
dcode models             # List all models by provider
dcode agent list         # List available agents
dcode session list       # List saved sessions
dcode session delete <id>
dcode export <session-id>
dcode import <file>
dcode share create <session-id>
dcode mcp list           # List MCP servers
dcode mcp add            # Add an MCP server
dcode worktree create <name>
dcode stats              # Usage statistics
dcode debug config       # Dump resolved config JSON
dcode version            # Show version and commit
dcode upgrade            # Self-update via go install
dcode completion bash    # Shell completion script

Project Structure

dcode/
├── cmd/dcode/              # CLI entry point (Cobra commands)
├── internal/
│   ├── agent/              # Agent definitions and system prompts
│   ├── config/             # Config loading, credentials, provider registry
│   ├── earlyinit/          # Lipgloss dark-background init (WSL2 fix)
│   ├── provider/           # AI provider implementations (20+)
│   ├── server/             # HTTP API server (headless mode)
│   ├── session/            # Session store, prompt engine, compaction
│   ├── share/              # Session sharing client
│   ├── tool/               # Tool registry and implementations
│   ├── tui/                # Bubbletea TUI model and components
│   └── worktree/           # Git worktree management
├── .github/workflows/      # CI + release (GoReleaser)
├── .goreleaser.yml         # Cross-platform release config
├── Makefile                # Build targets with ldflags
├── go.mod
└── LICENSE

Development

Prerequisites

• Go 1.24.2+
• golangci-lint (optional, for make lint)
• goreleaser (optional, for make snapshot)

Common tasks

make build      # build with version/commit injected
make test       # go test -race ./...
make cover      # test + HTML coverage report
make fmt        # gofmt -w .
make lint       # golangci-lint run
make cross      # cross-compile Linux/macOS/Windows
make snapshot   # local goreleaser snapshot

Adding a new AI provider

Implement the Provider interface in internal/provider/:

type Provider interface {
    Name() string
    Models() []ModelInfo
    CreateMessage(ctx context.Context, req *MessageRequest) (*MessageResponse, error)
    StreamMessage(ctx context.Context, req *MessageRequest, cb func(*StreamChunk) error) error
}

Register it in internal/provider/provider.go → CreateProvider.

Adding a new tool

Implement tool.Tool in internal/tool/ and register it in tool.GetRegistry().

License

MIT — see LICENSE.

Acknowledgments

• Inspired by opencode
• TUI powered by Bubbletea, Bubbles, Lipgloss, Glamour
• Syntax highlighting by Chroma
• CLI by Cobra