DGC v0.47.2Edit on GitHub ↗
Docs/Overview
Everything DGC does, grouped. Start at the top, or jump straight to the part you need.
DGC documentation
#DGC is a local-first coding agent for your terminal and for VS Code. Native local and API routes run DGC's agentic loop — planning, reading files, running tools, and editing code — against a model you choose. Supported subscription routes delegate the model-and-tool turn to the vendor CLI. The selected model receives the conversation context it needs; optional remote integrations receive the requests you direct to them.
New to DGC?
Install it, launch it, and connect a model — about five minutes.
Getting started
#Getting started
Install, first launch, connect a model.
Keyboard shortcuts
Essential key bindings in the composer + transcript.
Slash commands
The full / command reference.
Command line
Flags, subcommands, and one-shot runs.
Using DGC
#Permission modes
Native default · acceptEdits · plan · auto.
Plan mode
Read-only planning, then one-tap approve.
Sessions & rewind
Resume, jump, and undo whole turns.
Checkpoints & rewind
What a recovery point holds, /rewind, the editor's Undo, and what cannot come back.
Context notes
What the project already learned, across context windows.
Memory
DGC.md, #fact, and /memory — guidance that loads every session.
Standing goals
Persistent objectives with an explicit lifecycle.
Files pane
A file explorer under the transcript while the agent works.
Diff pane
Every changed file, its counts and its diff, live under the transcript.
Git changes
Git_diff, Changes in this chat, and Workspace changes.
SDK
Embed DGC in an application or CI job.
In your editor
The VS Code and Cursor panel, and what a finished turn gives you.
Pasted text
A tall paste, a pasted link, and characters you cannot see.
Questions
When the model needs you to pick, not guess.
Turn ETA & notifications
How long the turn still needs, and a ping when it is done.
Background monitors
Watch a long-running command and hear about each line it prints.
Sandbox
Confine native shell commands to the project.
Worktrees
Fleet, named, and delegated checkouts.
Providers & models
#Connect your model
Point DGC at Ollama, llama.cpp, vLLM, or a cloud host.
Reconnecting
What DGC does when a model request fails, is refused as busy, or is cut off.
Subscriptions
Bring your own Claude / Codex / Qwen / Kimi / Copilot plan.
Thinking & reasoning
Off · low · medium · high · xhigh, and preserving it.
Token usage
Tokens and requests counted on this machine, by model and day.
Web search
DuckDuckGo by default; Brave, Tavily, or SearXNG if you set them.
Features
#Artifacts
Preview what the agent builds on a localhost URL.
Produced files
The agent shows you a file it made, as a chip you can click.
Looking at a page
Drive a real browser to see a deployed or local site.
Viewed images
Screenshots and image files the model looked at, kept with the step that looked.
Plugins, apps and MCP
Install a plugin, connect an app, and what a connector brokers.
MCP servers
Connect external tools over MCP.
Lifecycle hooks
Run observable commands at agent boundaries.
Skills
Reusable instruction packages.
Multiple agents
Run a fleet of agents at once + the dashboard.
Sub-agents
Every sub-agent this chat started, and what each one is doing.
Handoff
A continuation document from this session, for the next one.
Code intelligence
Symbols, definitions, references, and a repo map.
Built-in tools
What the native agent can call.
Training export
Your real sessions → scrubbed fine-tuning JSONL.
Python code-action (power mode)
A persistent Python interpreter for token-efficient work.