mrkeyoor.com_
Fri 16 Oct 18:51 UTC
AI Toolsevaluationupdated 14 Oct 2026

AionUi review

AionUi is a desktop workspace where AI agents read and change files, generate documents, and show their output beside the conversation. It includes a built-in agent and can bring tools such as Claude Code into one interface, with browser access and scheduled tasks for work away from the desktop.

+43 / 1dstars / 7d
Verdict

Our AionUi run passed 4,980 tests but failed its build after 81 seconds while preparing bundled AionCore resources, so we cannot recommend this commit as ready for source packaging. Try the desktop workspace when file previews and multiple agents solve a daily problem, after checking your provider and permission path. Wait if Windows Antigravity approvals or unverified custom vision input are essential to your work.

We ran it

Lab card: what happened when we ran AionUiScreenshot of AionUi (www.aionui.com)
Install✓ · 29s1603 packages · 1490 MB
Build✗ · 81s
Tests✓ · 309s4980 passed · 0 failed · 5 skipped of 4985 (vitest)
Repo2341 files~298,097 lines of source · 494.1 MB · 13 CI workflows · Dockerfile · tests dir

Answers from our run

Does AionUi build from source?

Dependencies installed in 29 seconds (1603 packages), and the build failed. We cloned commit c140550 into a clean Debian container with 3 CPUs and no project-specific setup.

Do AionUi's tests pass?

Yes: 4980 of 4985 passed when we ran the project's own test command (vitest). Some failures need services or credentials a bare container does not have.

Who should not use AionUi?

Windows users who need Antigravity with approval prompts: open issue 4317 reports failed tool hooks in v2.2.2, with Full Auto bypassing approval as the workaround.

What are the alternatives to AionUi?

Goose, Cherry Studio. Our AionUi run passed 4,980 tests but failed its build after 81 seconds while preparing bundled AionCore resources, so we cannot recommend this commit as ready for source packaging.

Setup2/529-second install, 1,490 MB footprint, then a failed resource build
Docs3/5English guides explain the backend, but the installer link is stale
Community4/5October 14 push, 33,503 stars, and October 11 issue activity
Maturity3/54,980 tests passed, with packaging and permission paths unresolved

Who it’s for

Developers using several CLI agents who want shared file previews and one place to manage MCP tools.
People who regularly generate Word, Excel, or PowerPoint files and need to inspect the output beside the conversation.
Technical users willing to configure a model provider and verify their chosen agent's approval behavior before enabling remote or scheduled work.

Who it’s NOT for

Windows users who need Antigravity with approval prompts: open issue 4317 reports failed tool hooks in v2.2.2, with Full Auto bypassing approval as the workaround.
Teams depending on custom NVIDIA NIM vision input without request inspection: open issue 4315 reports a validated configuration sending text without attached images.
Developers expecting one JavaScript checkout to supply the whole application: the development guide requires a separate Rust AionCore backend discoverable through PATH.
Organizations that require desktop installers to come from GitHub releases: v2.2.2 says installers now come from the official website.

Setup reality

Our sandbox run of commit c140550 installed 1,603 packages in 29 seconds and used 1,490 MB on disk. The build failed with exit code 1 after 81 seconds. Vitest succeeded in 309 seconds: 4,980 passed, 0 failed, and 5 skipped out of 4,985. We used a fresh Debian container with 3 CPUs, 8 GB RAM, image lab-node:22, no secrets, and no elevated privileges.

The built-in agent needs a configured provider, using an API credential or a local model endpoint such as Ollama or LM Studio. External agents need their own authentication, and messaging integrations need bot configuration. The latest release sends installer downloads to the official website, despite the README's GitHub download link.

Local development needs a separate AionCore repository, Rust/Cargo, Node.js, Bun, and Python for native compilation. Our renderer built before preparation tried to execute downloaded AionCore v0.2.2 under linux-arm64 and failed with shell errors. The log does not prove the cause. Approval behavior also varies by agent, with a current Windows Antigravity hook failure reported upstream.

AionUi puts file operations and previews in one desktop

AionUi lets an agent rename files in your workspace, so its permission controls matter as much as its chat window. AionUi puts that kind of work inside an Electron app: a built-in agent can read and write files, while a preview panel displays the results. You can also bring installed agents such as Claude Code into the same interface.

The appeal is having your work and its output together. Document previews cover PDF, Word, Excel, and PowerPoint. Our checkout occupied 494.1 MB across 2,341 files, with about 298,097 source lines.

What happened when we ran it

Our run installed 1,603 packages in 29 seconds and used 1,490 MB on disk. We cloned commit c140550 on October 14, 2026, into a fresh Debian sandbox with 3 CPUs and 8 GB of RAM, using image lab-node:22. The container had no secrets or elevated privileges. Installation succeeded, but the build exited with code 1 after 81 seconds.

Vitest succeeded in 309 seconds: 4,980 tests passed, none failed, and 5 were skipped out of 4,985. The repository scan found 13 CI workflow files, a Dockerfile, a tests directory, and workspace configuration. The passing suite gives us confidence in the checks that actually ran, while the failed build leaves the packaging path unresolved. We measured no agent accuracy or document quality.

The renderer finished building in 59.21 seconds before resource preparation failed. The log shows AionCore v0.2.2 downloading into a linux-arm64 directory, followed by an attempt to run that binary with prepare-managed-resources --bundle. The shell printed an ELF-related not found message and Syntax error: newline unexpected. The excerpt does not establish the host architecture or prove a missing system package caused the failure.

Installers moved off GitHub, and development needs AionCore

The v2.2.2 release, published September 9, says installers are no longer published on GitHub and directs users to the official website. The README still sends its download button to GitHub releases. That's a frustrating detour for someone following the quick start. Our source build failure does not establish whether the website's installers work, because that distribution path was outside the supplied measurements.

The development guide requires a separate AionCore checkout, Rust and Cargo, plus Node.js and Bun. Python is also listed for native compilation. AionUi discovers the backend through the terminal's PATH and launches it itself. The 1,603-package installation therefore describes the measured repository environment, not a complete account of everything needed for local development. Plan for both repositories if your adoption depends on changing the application.

Agent permissions vary, and Windows Antigravity has a reported blocker

AionUi manages MCP tools and coordinates agents through the Agent Client Protocol, which connects external agents to the interface. Team mode gives a leader and teammates a shared folder, mailbox, and task board. Each agent has its own approval dialog. The README explicitly says unattended modes and permission behavior depend on the selected agent. Sharing a window does not make different agents enforce identical permissions.

Open issue 4317 reports broken Antigravity tool hooks on Windows in v2.2.2, with Full Auto offered as a workaround that bypasses approval. I would wait for a verified fix if that combination is essential. Our 4,980 passing tests cannot settle this reported integration failure. Start with a disposable workspace and confirm a rejected operation stays rejected before allowing the agent near files you need.

Local conversation storage still requires a deliberate model choice

The README says conversations live locally in SQLite and lists hosted providers alongside Ollama and LM Studio. Local storage describes where the application keeps its records. Selecting a hosted model still means choosing an external service to process requests. The built-in agent needs provider configuration, such as an API credential or a local endpoint. External CLI agents bring their own authentication, and Telegram or other channels require separate bot configuration.

A passing connection check also needs scrutiny. Open issue 4315 reports a custom NVIDIA NIM configuration accepting a vision setting while sending text without the attached images. That's a user report, not an outcome from our 309-second test run. If image input matters, ask about a known image and inspect the actual request. A green settings indicator alone is insufficient evidence that your chosen model receives the attachment.

Goose suits one agent, while Cherry Studio suits model chat

Goose has a desktop app and CLI with MCP extensions. Pick it when you want one general-purpose agent rather than AionUi's interface for coordinating several external agents. Cherry Studio supports multiple providers, document conversations, and local models. It makes more sense when comparing answers and working with documents are your main activities, without needing AionUi's team coordination.

Our 1,490 MB installed footprint makes the extra interface worth questioning if you already work comfortably in a terminal. AionUi earns that space when its previews and shared workspace reduce your daily switching between tools. Try a representative task before migrating your habits: have your chosen agent produce a document, open it in the preview panel, and verify an approval prompt. We have no comparative measurements for these alternatives.

An October 14 push shows activity alongside unresolved reports

The repository's last push was October 14, 2026. GitHub listed 33,503 stars and 862 open issues and PRs, and the Windows and vision reports received October 11 updates. Those dates show ongoing activity alongside unresolved problems. The September release date alone says little about maintenance. Apache-2.0 licensing and an English development guide make inspection and modification accessible, though maintaining the separate backend adds work.

Choose your adoption check around the path you'll actually use. For a source-based deployment, reproduce the resource-preparation step that stopped our build after 81 seconds before investing in customization. For Windows Antigravity, require working approval hooks. For document work, verify the generated file opens correctly in the software that will consume it. Keep your existing workflow until AionUi completes that specific job with the permissions you expect.

Alternatives

ProjectWhat it isPick it when
Goose gh↗A general-purpose local AI agent with a desktop app, CLI, and MCP extensions.pick this instead when you want one agent across desktop and terminal workflows rather than AionUi's coordination of several external agents.
Cherry Studio gh↗A desktop client for multiple model providers, document conversations, and local model endpoints.pick this instead when comparing model answers and discussing documents matter more than AionUi's shared agent workspace and team coordination.

What people are saying

  1. [velocity-scout] iOfficeAI/AionUi

Sources

  1. AionUi README
  2. AionUi repository facts and activity
  3. AionUi v2.2.2 release and installer location
  4. AionUi development guide
  5. Windows Antigravity permission hook report, issue 4317
  6. Custom NVIDIA NIM vision input report, issue 4315
  7. Goose README
  8. Cherry Studio README

More ai tools reviews

universal-modder · DeepEP-Ascend · DeepGEMM-Ascend · hypoarena · jeeves · faster-whisper · the whole board →