Aura IDE
A desktop AI coding workspace for repository-aware changes, validation, reusable agent teams, and controlled diff approval.
- Source repo
- CarpseDeam/Aura-IDE
- Stars
- ★ 57
- Last updated
- 1mo ago
- License
- MIT
- Primary language
- Python
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer making a focused change in an existing repository can ask Aura to identify an under-tested function, add a test, run it, and explain the outcome.
- Not a fit if
- macOS and Linux users who want a one-click installer instead of a source build
- Users who want the phone app to run tasks without the desktop staying on
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Aura IDE is an open-source desktop coding application that places conversation, project files, source code, and execution results in one workspace. It reads and searches repositories, follows references, and uses project tools and the terminal to make and validate changes. A task can run through Aura directly or be divided among Agents with their own instructions, model choices, and permissions. Agents can be connected into reusable Teams whose assignments and branches are visible in a graph editor. The application supports hosted DeepSeek, OpenAI, Anthropic, Gemini, and OpenRouter connections as well as Ollama, LM Studio, llama.cpp, and compatible OpenAI-style local servers. It is a practical fit for developers who want multi-model flexibility while retaining diff review, read-only or planning controls, and isolated worktrees for Agent changes.
The user opens a project folder and either adds a hosted provider key under Settings → API Keys or connects a running local server under Settings → Models. Aura reads and searches the repository, follows code references, and works through project tools, Git, and terminal tools; its workspace exposes tool activity, terminal output, checks, failures, and a final report. With Agents enabled, Aura can assemble a team or invoke a saved Team, while each Agent can inherit Aura's current model or use another configured model and thinking level. A Team connects Agent assignments into a graph that can be arranged, edited, extended with branches or helpers, and run again for a new task. Writable Agent work takes place in isolated worktrees, after which retained changes are reviewed and applied through Aura; manual approval can also require inspection of proposed diffs before writes. Aura can connect to MCP servers, while Aura Companion lets a phone browse projects, send messages, and follow work through a local or hosted relay as long as the paired desktop remains running.
- A developer making a focused change in an existing repository can ask Aura to identify an under-tested function, add a test, run it, and explain the outcome.
- A team that repeatedly follows an implementation-test-review sequence can save those responsibilities as a Team and rerun the workflow with new tasks.
- A maintainer working in a sensitive repository can use Read Only, Plan review, or manual pre-write diff approval to restrict how changes are made.
- A developer evaluating model options can mix hosted and local models in one Team and select a separate configured model for each Agent.
- A user supervising a long-running desktop task can pair Aura Companion, browse the project from a phone, and monitor progress while the desktop stays online.
How do you install or deploy this agent?
On Windows, download and run the latest .exe from GitHub Releases. Python is bundled, installation is per user, and subsequent updates are available inside Aura.
On macOS and Linux, Aura must be run from source with Python 3.10 or newer. The supplied material does not provide the exact repository-cloning, dependency-installation, or launch commands, so a verified command-line installation sequence cannot be reproduced here.
For first use, also provide either an API key for a supported hosted provider or a running local server such as Ollama, LM Studio, or llama.cpp. Local coding requires a model or server with tool-calling support and enough context for the task.
How do you use this agent?
- Open a project folder.
- For a hosted model, add its key in Settings → API Keys. For a local model, connect its running server in Settings → Models.
- Submit a small first task such as: Find a function that needs a test, add one, run it, and explain the result.
- Inspect the changed files, proposed or retained diffs, and reported validation results.
- For multi-Agent work, enable Agents and let Aura assemble a team or select a saved Team.
- To define a reusable process, ask Aura to create a workflow that implements a requested change, tests it, and reviews it. Creation saves the workflow without executing it; inspect or revise its graph, then select Run and provide a task.
- Review writable Agent changes from their isolated worktree and apply the changes you want to retain through Aura.
What are this agent's strengths and limitations?
- Supports five named hosted options and several local-server paths, with hosted and local models usable together in one Team.
- Agents and Teams can be saved, refined through conversation, and inspected in a graph editor with explicit assignments and branches.
- Manual pre-write diff approval, Read Only, Plan review, and isolated Agent worktrees provide several concrete layers of change control.
- Project files, chat, tool activity, terminal output, validation results, failures, and the final report are presented in one desktop workspace.
- The Windows installer bundles Python and supports in-app updates, reducing runtime setup for Windows users.
- macOS and Linux require a source installation, but the supplied material does not include complete copyable installation or launch commands.
- Hosted providers charge separately for API use and require the relevant credentials.
- Local operation depends on a separately running Ollama, LM Studio, llama.cpp, or compatible server, plus a model capable of tool calling and adequate context.
- Aura Companion is not an independent execution host; the paired desktop must remain running.
- The documented completed example is a small Python project, leaving performance and reliability on large repositories, other languages, or complex Agent graphs unproven.
How does this agent compare with similar options?
Unlike a hosted-API-only setup, Aura can connect to Ollama, LM Studio, llama.cpp, and compatible OpenAI-style servers, allowing model execution to remain local; the tradeoff is that the user must operate the server and choose a model with tool calling and sufficient context. Its hosted choices—DeepSeek, OpenAI, Anthropic, Gemini, and OpenRouter—offer a more direct connection path but incur provider API charges. An individual Agent suits a reusable standalone role, while a Team is intended for saved, visual workflows that connect responsibilities such as implementation, testing, and review.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Aura IDE This agent | 52 · Major gaps | Desktop appFree + model costs | ★ 57 | 1mo ago | Python | OpenAI API · Claude API |
| Pi Dynamic Workflows | 72 · Some gaps | Agent plugin / skillFree + model costs | ★ 591 | 12d ago | TypeScript | — |
| Fusion Software Factory | 67 · Some gaps | CLIFree + model costs | ★ 1.3k | 13d ago | TypeScript | Codex · OpenAI API · Claude API |
| Synara | 57 · Major gaps | Desktop appFree + model costs | ★ 2.2k | 1d ago | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README and configuration guide describe read-only mode, plan review, diff approval, and isolated Agent worktrees; test samples also cover permission freezing at submission, path validation, and rollback on persistence failure. These support least privilege, confirmation, and rollback in part. The guide advises removing keys and private project content before sharing logs, but the excerpts do not show key-storage protections or a complete data-flow account. Dependencies have lower-bound versions but no visible security audit. MCP, terminal, Git, model services, and the Companion relay can have external effects, while the excerpts give limited detail on their scope and safeguards. The MIT license and repository identity do not establish authorship or maintenance responsibility.
Startup code logs failures and presents a dialog or console message, and the setup guide includes troubleshooting, so failure handling has some evidence. The README describes Windows downloads and in-app updates while source installation covers multiple platforms; the configuration guide says ordinary coding uses one continuous production model while the README also describes Agents/Teams and per-Agent model choices, leaving some product boundaries unclear. Dependencies have version floors, but no lockfile or availability policy is shown.
The materials target desktop coding users and cover hosted and local models, multiple operating systems, Agents/Teams, and Companion use. Read-only, plan, approval, sandbox modes, and local-server requirements give some capability boundaries; however, WASM is reserved for the future and relay pairing endpoints are explicitly unimplemented stubs. Agent activation is largely user-driven through a toggle or request, and precise automatic team-selection behavior is not established.
The README, setup guide, configuration guide, and links to tool and safety documentation provide a clear entry structure, installation steps, example tasks, troubleshooting, and some limitations. The MIT license text is complete and earns full marks. A version is present, but the supplied material has no changelog; support and issue-reporting channels exist, but do not establish specific maintenance ownership or update commitments. Naming varies somewhat among Aura, Aura IDE, and Aura-IDE.
The desktop workspace combines repository context, editing, terminal use, validation, and Agent workflows; diff reporting, approvals, and isolated worktrees provide evidence of usable outputs. Reusable teams and local-model connections offer practical value, but the described sample run and feature claims do not establish general task correctness. Hosted API usage is billed separately, and the supplied configuration does not provide a complete cost-benefit account for defaults such as temperature and tool-round limits.
README claims about review controls and worktrees have related support in the configuration guide and Agent test samples; installation, version, and license claims also have file-level support. Test files contain concrete assertions for the behaviors they cover, but static material cannot establish that tests were executed or that they cover the whole product. Documentation distinguishes some features, limitations, and costs, which constrains inference; some behavior remains documented assertion, and the supplied sample is not the full repository.
- This is a low-confidence static review; the material does not establish that tests were run or cover the full repository.
- Before use, check model/API data transmission, key storage, and the actual permissions and configuration of enabled terminal, MCP, and Companion relay features.
- The documentation says the WASM sandbox is reserved for the future and relay pairing endpoints are unimplemented stubs; do not assume those capabilities are available.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/aura-ide)<a href="https://followagents.com/en/agents/aura-ide"><img src="https://followagents.com/badge/aura-ide.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2.3.04 # Aura v2.3.04
-
v2.3.03 # Aura v2.3.03
-
v2.3.02 # Aura v2.3.02
Release notes excerpt
* Talk to Aura while she works: Send corrections or additional instructions during an active task. Updates reach Agents, Teams, and nested helpers, and remain in conversation history. * Steer now or queue next: Choose Send update to guide the current task or Queue next task to line up another request. Completed operations keep their results; unstarted tools from the previous plan are marked as skipped. …View on GitHub ↗
FAQ
Does Aura itself cost money?
Will Agents modify my main working directory directly?
Can it run entirely with local models?
What is visible when a task or validation step fails?
Can phone access continue after the desktop is shut down?
Thanks — we use this to improve the page.