engineering / agent
Aura Code
A coding agent for the terminal with persistent memory, a web studio and a Telegram bot, using the model provider you choose.
At a glance
- Purpose
- A coding agent for the terminal with persistent memory, a web studio and a Telegram bot, using the model provider you choose.
- Runs in
- Local agent with shell, Web service
- Operating systems
- Linux, Windows
- Needs
- Runtime: Node.js 18+; Needs internet: Yes; Needs accounts or API keys beyond your agent: Not stated
- Files and data
- Reads files: Yes; Writes files: Yes; Sends data to other services: Yes
- Tool type
- Agent
- Delivery
- From GitHub
- Version
- Latest on the master branch
- Access
- Code on GitHub
- Compatibility
- The README's Node badge specifies Node 18 or later, and its installation command uses npm. The README names providers including DeepSeek, Claude, GPT, Gemini, GLM, MiMo, Ollama, LM Studio, OpenRouter, Groq and xAI, and describes support for OpenAI-compatible endpoints. The README links a Windows guide. The README says --sandboxed requires Linux with bubblewrap and describes the computer tool for Linux. The README also lists a Telegram bot and MCP support.
- External costs
- The README's quick start sets a DeepSeek API key. It names providers including DeepSeek, Claude, GPT, Gemini, GLM, MiMo, OpenRouter, Groq and xAI, and says a local model through Ollama needs no key. Its cost controls section gives example session costs and describes turn and token ceilings you can set. Your coding agent or model provider may charge for usage.
- Permissions
- The README's setup installs Aura globally with npm install -g aura-code and configures provider API keys. The README describes memory writes under ~/.aura or AURA_HOME, project .aura files, training-data output and saved configuration. The README describes file reads, edits, shell commands, configurable y/N approvals and auto-approval controls. The README lists :delete for saved-session deletion, :clear-history to wipe conversation history and --reset-setup to wipe saved configuration and rerun setup. The README names email, calendar, Telegram, WhatsApp, cron and browser tools, plus Linux screen and pointer control via :compon after disclosure. The README describes :designx scraping references and writing design artifacts. The repository's top-level entries include scripts/, whose contents the README does not describe. The README says its Telegram bot offers voice notes, PC control, file transfer and webcam snapshots.
- Usage terms
- MIT License, as stated in the repository. See its LICENSE file for the exact terms.
Code on GitHub
DusanCar-sudo/aura-code
MITAura — AI coding agent with persistent memory and TUI
Public GitHub data fetched · 0 days old
View on GitHub ↗What it does
Aura Code is a terminal coding agent. The README describes natural-language tasks and a flow of reading the codebase, planning, executing, checking results and reporting. The README describes model-provider choice, OpenAI-compatible endpoints and local Ollama models.The README lists a terminal interface, persistent memory, Telegram control, MCP support, a web studio started with aura serve and a conversational Gazelle mode. The README describes Archimedes routing between a local model and a cloud model.
The README lists :coder and :gazelle mode controls, :lessons for stored lessons and :mine for session-derived material. The README describes --readonly, --auto and --sandboxed as controls over permitted actions.
Who it is for
Developers who want a terminal coding agent they can point at many model providers or at a local model.Example tasks
Run aura with a plain-language task such as refactoring the auth module to use JWT.Run aura serve to open the web studio.
Run aura --readonly with an analysis task to inspect code without file writes or shell commands.
Type a plain line while a task runs to steer it.
Run :lessons in the terminal interface to list what Aura has learned.
Getting started
The README's quick start runs npm install -g aura-code, then export DEEPSEEK_API_KEY=sk-..., then aura with a task in quotes, for example aura 'refactor the auth module to use JWT'. The README also mentions a first-run setup wizard, which --no-setup skips, and a local profile, --profile local, that uses Ollama defaults with no API key. The README has the exact commands. The README says aura setup telegram creates a user systemd service and shows enabling it to run in the background.Limitations
The README describes Aura as a framework rather than a turnkey product. The README warns that reaching the --max-turns cap should be expected during real project work. The README's badge gives Node 18 or later, and setup uses npm. The README describes cloud-provider API keys and a local Ollama profile without a key. The README says Archimedes escalates when Ollama is unreachable and warns that an AMD iGPU's Vulkan prefill can cause long hangs, suggesting CPU-only Ollama for that case.The README says --sandboxed requires Linux and bubblewrap, refuses on other platforms and cannot be combined with --computer. The README describes the computer tool for Linux and links a Windows guide. The README says Gazelle cannot read files, search code or execute tasks. The README says :designx PDF output is rendered through headless Chrome when it is installed. The README says custom service units should use compiled JavaScript, with npm run build required after source changes, and warns that restart policies do not fix bot failures that do not exit. The repository can change at any time; check its README before you install.
Data practices
The README says memory is stored under ~/.aura or AURA_HOME, with repository lessons in the project's .aura folder. The README says lessons are injected into the system prompt and describes both cloud providers and local Ollama models; local storage therefore does not establish that lesson content stays on the machine.The README says :mine derives material from sessions, describes training-data output, and says /cost reads .aura/token-log.jsonl. The README lists --no-session to disable conversation-history persistence. The README also describes Telegram, other communication tools, browser access and reference scraping. The README does not describe telemetry.