# aura-code

> Terminal coding agent that remembers past sessions and works with 31 model providers

- Repository: https://github.com/DusanCar-sudo/aura-code
- Category: Coding agents & dev tools
- Stars: 47 · Forks: 5 · Language: TypeScript · License: MIT
- Experience: Some experience
- Page: https://ir.adyog.com/repos/dusancar-sudo/aura-code/

Aura Code is a command-line coding agent. You describe a task in plain English. It reads your code, plans, makes changes, runs commands to verify them, and reports back. It is model-agnostic: you pick the provider, or point it at a local model through Ollama or LM Studio.

It keeps lessons between sessions in ~/.aura/memory/ on your computer, such as what broke last time and what your build needs. An optional layer called Archimedes tries each task with a small local model first. It tracks how often that model succeeds on similar tasks, and a larger cloud model takes over when the local one is unproven or fails a check.

The README contrasts Aura with fixed coding-agent products: skills, plugins and MCP servers are ordinary files you can edit. Besides the terminal UI it offers a web studio, a Telegram bot, and a lean chat mode called Gazelle that skips tool schemas to save context. The project is young, and the token-saving figures come from the authors' own measurements.

## What it does

- **Autonomous coding loop**: It reads files, edits code, runs shell commands, verifies the result and retries. You can type while it works, and your note is folded into the next turn.
- **Persistent memory**: Lessons and project context are saved locally and read back in later sessions. The README says this memory never leaves your machine.
- **Local-first routing (Archimedes)**: A small Ollama model attempts tasks first. A cloud model steps in when the local one has a weak track record or fails a verification check. The default local model is qwen2.5-coder:1.5b.
- **31 routing targets, 4 transports**: Works with DeepSeek, Claude, GPT, Gemini, GLM, Ollama, OpenRouter and more. Most providers share one OpenAI-compatible transport, so adding one needs a base URL, a key variable and a model-id rewrite.
- **Gazelle chat mode**: A conversational path with no tool schemas or project context. The README measures 356 tokens per turn against 9,367 for the full coder mode.

## What you could build

- Refactor a module by describing the change in plain English and letting the agent edit and verify it.
- Run a coding agent against a local Ollama model so your code stays on your machine.
- Compare cloud models on the same task by switching providers through one tool.
- Collect your sessions as structured training examples with the :mine command, for a later fine-tune.
- Write your own skills, plugins or MCP servers as plain files and connect them to the agent.

## Who it's for

Developers comfortable with the terminal, Node 18+ and API keys, who want to choose or run their own models.

## Get started

Install the package globally with npm, set an API key for your chosen provider (the example uses DeepSeek), then run aura with a task in quotes. Node 18 or newer is required.

```
npm install -g aura-code
```
```
export DEEPSEEK_API_KEY=sk-...
```
```
aura 'refactor the auth module to use JWT'
```

Tags: coding-agent, cli, tui, local-llm, memory