0g-aura
Decentralized AI coding agent powered by 0G Compute. Describe what you want to build, edit, or fix — aura streams responses from verifiable GPU providers and applies file changes directly to your project.
Install
npm install -g 0g-aura
Or with npx (no install):
npx 0g-aura
Or clone and link locally:
git clone <repo-url>
cd 0g-aura
npm install
npm run build
npm link
Setup
aura init
This walks you through wallet setup interactively:
- Create a new wallet or import an existing private key
- Choose where to save (
./.envor~/.aura/.envfor global) - Checks your A0GI balance automatically
You need A0GI tokens to pay for decentralized compute. Get them at portal.0g.ai.
Usage
# Start in current directory with default model
aura
# Choose a model
aura -m deepseek-chat-v3-0324
# Target a specific project directory
aura -d ./my-project
# Skip apply confirmations
aura --no-confirm
Inside the TUI, type a prompt and press Enter. Aura will:
- Read your project files for context
- Stream a response from a 0G compute provider in real-time
- Show a diff of proposed changes
- Ask for confirmation before writing files
CLI Flags
| Flag | Description |
|---|---|
-m, --model <name> | Model to use (default: GLM-5-FP8) |
-d, --dir <path> | Project directory (default: cwd) |
--no-confirm | Apply changes without confirmation |
-V, --version | Show version |
-h, --help | Show help |
Slash Commands
| Command | Description |
|---|---|
/help | Show all commands |
/model <name> | Switch AI model |
/add <path> | Add a file to prompt context |
/remove <path|all> | Remove file(s) from context |
/files | Show detected project tree |
/status | Check 0G Compute connectivity |
/balance | Check wallet A0GI balance |
/undo | Revert the last applied file changes |
/export [file] | Save conversation to markdown |
/cost | Show session token usage |
/clear | Clear conversation history |
/quit | Exit |
Keyboard Shortcuts
| Key | Action |
|---|---|
Enter | Send prompt / confirm apply |
Esc | Clear input / reject apply |
Up/Down | Recall previous prompts |
Shift+Up/Down | Scroll conversation |
PageUp/PageDown | Scroll fast |
Home/End | Jump to oldest/latest |
Ctrl+C | Exit |
Models
| Model | Description |
|---|---|
GLM-5-FP8 | Default, fast general-purpose |
deepseek-chat-v3-0324 | DeepSeek v3 |
gpt-oss-120b | 120B open-source |
qwen3-vl-30b-a3b-instruct | Qwen3 multimodal |
Configuration
| Environment Variable | Default | Description |
|---|---|---|
ZG_PRIVATE_KEY | (required) | 0G wallet private key |
ZG_RPC_URL | https://evmrpc.0g.ai | 0G RPC endpoint |
ZG_AUTO_DISCOVER_PROVIDER | true | Auto-discover compute providers |
ZG_PROVIDER_ADDRESS | — | Pin a specific provider (skips discovery) |
Config is loaded from ./.env first, then ~/.aura/.env as fallback. Run aura init to create one.
How It Works
Aura connects to the 0G decentralized compute network through the @0glabs/0g-serving-broker SDK. When you send a prompt:
- Context assembly — scans your project tree and reads key files (with automatic token budget management)
- Provider discovery — finds an available GPU provider serving your chosen model (or uses a pinned address)
- Streaming inference — sends your conversation + project context to the provider and streams the response in real-time
- Parsing — extracts file changes and shell commands from the model output
- Apply — shows a diff preview, then writes files and runs commands on confirmation (with
/undosupport)
All inference runs on decentralized, verifiable GPU providers — no centralized API keys needed, just an 0G wallet with A0GI tokens.
License
MIT