For SpinxAI Client 0.7. Status labels and limitations distinguish available features from what is still on the roadmap.
Introduction
SpinxAI 0.7 is a desktop application for delegating work to AI assistants, following progress, and using the results. Work data is stored on your computer. This guide distinguishes features available today from work in progress.
It is designed for office workers, researchers, customer support teams, and small business owners who want to automate repetitive work without programming.
What you can do:
- Have multi-turn conversations with AI and refine your requests.
- Work with PDF, Word, Excel, and PowerPoint documents; export results as Word, Excel, HTML, or Markdown.
- Create role-based assistants with their own instructions and permissions.
- Schedule daily, weekly, one-off, or interval-based tasks.
- Let AI read websites, fill forms, and transfer files under your supervision.
- Delegate through Telegram, personal Zalo, and Discord.
- Use two-way Live Voice, locally or through a cloud provider.
- Install additional skills from the SpinxAI Library.
Status labels:
| Label | Meaning |
|---|---|
| Available | Implemented and usable in version 0.7 |
| Partial | Core behavior exists; the remaining limitations are listed |
| Interactive design preview | A sample screen without a working backend; data disappears on reload |
| Planned | Not available in the app yet |
Roadmap items are directional, not promised release dates. The local AI feature is marked Experimental.
Installation and system requirements
Version 0.7.0 has installers for the following platforms. Choose the matching build from the SpinxAI home page.
| Platform | Installer | Notes |
|---|---|---|
| macOS Apple Silicon (M1 or newer) | .dmg | Signed and notarized. Includes local voice and on-device AI |
| macOS Intel | .dmg | No local Live Voice or on-device AI |
| Windows 10/11 x64 | SpinxAI_0.7.0_windows_x64-setup.exe (~152 MB) | Not code-signed; SmartScreen displays a warning. For the official installer, select More info → Run anyway. No on-device AI yet |
| Linux x86_64 (glibc, Ubuntu 22.04+) | .AppImage (~258 MB) | Requires FUSE 2 (sudo apt install libfuse2). GNOME Keyring or KWallet is needed to remember API keys. Compare the file with the accompanying SHA256SUMS.txt |
Before you start:
- Prepare an AI service compatible with OpenAI Chat Completions: base URL, model name, and API key. Presets are available for OpenAI, OpenRouter, and Ollama/LM Studio; other services can be entered manually. Models must support function calling to use tools.
- Without an API key, you can try the experience mode without AI.
- Install Google Chrome or Microsoft Edge for browser automation.
- A Google account is required to install skills from the Library.
Data and keys: tasks, selected documents, and results live in the app's data directory under your OS account. This data is not encrypted at the application level. If you choose to remember a key, it is stored in the OS secure store: Keychain, Credential Manager, or Secret Service.
Quick start
On first launch, SpinxAI opens a six-step setup guide.
- Choose your needs — select the type of work closest to yours. This sets the default audience for your results.
- Check your computer — the app checks required components. If anything fails, open Gateway → Diagnostics.
- Connect AI — choose OpenAI, OpenRouter, Ollama/LM Studio, or Other service. Enter the URL and key, select Test connection, choose a model, and Save connection. Without a key, choose Try locally.
- On-device AI (optional, macOS Apple Silicon) — choose a model suited to your RAM. Download runs in the background and can be stopped.
- Workspace and extras — set the workspace name, theme, and font size; choose tool permissions for your Personal Assistant; optionally configure Chrome Extension or Live Voice.
- Review and start — review the summary, including key storage. Try sample documents if you wish, then start.
You can select Later at any step. Reopen the guide from Settings → General → Open getting started guide.
Your first task: open Workspace (Không gian làm việc), enter a request such as “Plan my work for this week,” attach documents if needed, and select Start task. Continue in the same conversation: “Shorten this to five actions.” Find all delegated work in Tasks (Công việc).
Costs and data: AI mode sends requests, conversation history, and selected documents to your configured provider, which may charge for usage. Experience mode does not send data over the network and clearly labels results as not analyzed by AI.
Available features
Each section explains the workflow and current limitations. Bold screen names refer to navigation items in the app. Screenshots show the Vietnamese interface.
1. Tasks and conversationsAvailable
- Enter a request in Workspace, with or without documents, or choose a task template.
- Continue with follow-up questions in the same conversation. Stop this turn cancels a running turn; retry the last turn after an error.
- Requests for your input or approval appear directly in the conversation.
- Limit: 50 turns per conversation. Start a new conversation when needed. Editing sent messages and queueing several messages while AI is working are not supported yet.
2. Tasks and resultsAvailable
- Tasks groups each conversation with its latest result and earlier results.
- Filter by All, Needs you (questions, approvals, errors), In progress, Has results, or Archived. Search accepts Vietnamese without accents.
- Edit a result directly. Each save creates a new version; view and restore earlier versions.
- Export Word, Excel, HTML, or Markdown. Excel export puts each table on its own sheet.
- PDF and PowerPoint export from the result screen are not available yet.
View full image3. Input documentsAvailable
- Supported: text-based PDF, DOCX, XLSX, PPTX, TXT, MD, and UTF-8 CSV. Preview extracted text with page, sheet, or slide references before delegating.
- Limits: 10 files per task, 4 MB per original file, and 6 MB of extracted text in total.
- Scanned PDFs and images are not supported yet; OCR is not available.
4. AI connectionsAvailable
- Open AI connections → Add connection. Enter a name, API URL, model, and key. Test connection only reads the model list; it does not send documents.
- Enable Remember API key in secure storage to keep it across app restarts.
- Save up to 50 connection profiles and switch with Use this connection. Forget API key removes the stored key.
- Only OpenAI Chat Completions-compatible services are supported. Account sign-in through OAuth is not available yet. A run lasts at most 15 minutes.
5. AI assistantsAvailable
- Open AI assistants → Create assistant. Start from an office, document research, or customer support role, or write your own instructions. Choose an avatar and test before saving.
- Select a connection and model per assistant, with a fallback model if the primary service is overloaded.
- Attach skills, default documents, and tool permissions described in plain language.
- Under Folders & safety, grant read or read/write access to a folder. Each read/write still requires approval. Run code in a separate container uses Docker or Podman for isolated Node.js/Python execution without networking.
- Transfer an assistant using a
.spinx-agentfile or directly over LAN/VPN. API keys, folder grants, and channel sessions are excluded. Imported assistants stay paused until reviewed.
View full image6. Knowledge sourcesPartial
- Save reusable documents in collections. Search names and content without Vietnamese accents, and select documents when creating a task or schedule.
- Semantic search, automatic retrieval of relevant documents, folder/web synchronization, and long-term memory are not available yet.
7. Scheduled tasksAvailable
- Open Schedules → Create schedule. Choose an assistant, description, documents, and timing: once, daily, weekly, or at an interval, with a time zone. Use Run now to test.
- Enable, disable, skip, or catch up once. Run history and results remain after deleting the schedule.
- The app must remain open. Documents are snapshots from schedule creation; changes to original files do not update them automatically.
8. Browser automationAvailable
- AI uses a separate Chrome/Edge session, isolated from your personal profile. It can read pages and iframes, fill forms, click, select options, and upload/download files.
- Open Browser → Live view to follow the page. Take over at any time, for example to sign in yourself.
- Create a dedicated Sign-in profile to retain website sessions between launches.
- Website confirm/prompt dialogs appear in SpinxAI for your response. Grant or revoke notification permissions per website.
- Control of personal Chrome tabs and automatic profile unlocking after an unexpected shutdown are not available yet.
View full image9. Chrome Companion extensionAvailable
- Send the current page or selected text from Chrome to SpinxAI as a task awaiting review.
- Open Settings → Chrome Extension → Export package and unzip it. At
chrome://extensions, enable Developer mode and select Load unpacked. Generate a pairing code in SpinxAI and paste it into the extension's Options. - The extension is not on the Chrome Web Store yet. Pair it again whenever SpinxAI or Chrome restarts.
10. Telegram, Zalo, and Discord channelsPartial
- Open Channels → Add channel. Enter a Telegram/Discord bot token or scan a personal Zalo QR code, then assign an assistant.
- New senders receive a six-character pairing code. Approve them in Pairing. Group replies require an @mention and are disabled by default.
- Only the computer owner can use assistant tools; other approved users can chat. Supported files: PDF, DOCX, XLSX, PPTX, TXT, MD, and CSV, up to 4 MB.
- Channels require the app to stay open. Image/audio input and delivery of result files are not available yet; replies are text.
- Personal Zalo uses an unofficial API and the account may be restricted; use a secondary account. All three channels are still being validated with real accounts.
11. Live VoiceAvailable
- Have two-way conversations, interrupt the assistant, and read transcripts. Open Live Voice at the top of the app; configure it under Settings → Voice.
- Local: Whisper speech recognition and Vietnamese VieNeu speech synthesis run on your computer, without a separate voice key. Not available on macOS Intel.
- Cloud: OpenAI Realtime or xAI Grok Voice requires a separate API key with realtime access. This key stays in RAM only.
12. On-device AI (offline)Experimental
- Available only on macOS Apple Silicon. Under Settings → On-device AI, download and load Gemma or Qwen, then choose it in Model for this turn.
- No API key is needed, and processing does not send data off the device.
- The app checks RAM before loading. Reload the model after each app restart. Quality is lower than cloud models; use it for simple tasks.
13. Skill libraryAvailable
- Open Skills & extensions → Discover. Sign in with Google, review details and permissions, then install a signed skill. Skills are currently free.
- Choose a skill when delegating. Permissions are confirmed before execution. Each turn uses one skill.
View full image14. Notifications, settings, and backupsAvailable
- Notifications shows requests for input, new results, errors, and skipped schedules. OS notifications are not available yet.
- Settings controls the workspace name, opening screen, light/dark theme, larger text, and reduced motion.
- Under Settings → Data & backup, export JSON without keys. Restore adds data without overwriting existing work. Imported schedules are disabled and tool permissions are reduced to read-only. Backups are unencrypted, with an 8 MB limit.
- Gateway → Diagnostics shows runtime status and exports a JSON report without work content for support.
Quick reference
Channel commands
| Command | Who can use it | Action |
|---|---|---|
/start, /trogiup | Approved users | Show help |
/id | Approved users | Show user/group IDs for access setup |
/moi | Approved users | Start a new conversation |
/trangthai | Approved users | Show running or queued requests |
/huy | Approved users | Stop the current request |
/viec <description> | Computer owner only | Create a separate task with tools |
Limits to know
| Item | Limit |
|---|---|
| Documents per task | 10 files; 4 MB per original; 6 MB total extracted text |
| Turns per conversation | 50 |
| AI run duration | 15 minutes; 3 minutes per model call |
| Output per model call | 6,000 tokens |
| Channel attachments | 4 MB; PDF, DOCX, XLSX, PPTX, TXT, MD, CSV |
| AI connection profiles | 50 |
| Assistants | 200; each with up to 12 folders, 20 skills, and 10 default documents |
| JSON backup | 8 MB |
Does it cost anything?
SpinxAI and Library skills are currently free. Your connected AI provider charges directly for AI usage, if applicable. Experience mode, on-device AI, and local Live Voice do not incur provider charges.
Features in development and roadmap
The following groups describe readiness, not promised release dates. A feature is considered ready only after clean-install validation, usable cancellation, actionable errors, and testing on all advertised operating systems.
Coming soon — implemented, awaiting installer inclusion
- 26 built-in skills for Vietnamese users, without Library sign-in: document summaries, meeting minutes, Vietnamese administrative writing following Decree 30/2020, emails, contract comparisons, Word/PDF reports, slides, Vietnamese editing, Vietnamese–English translation, marketing copy, spreadsheet analysis, data normalization, lesson plans, tests, planning, and checklists.
- Create real Office files: write Word, typed Excel cells, 16:9 PowerPoint, and PDF into an authorized folder. Every write requires approval.
- Email and calendar: Gmail, Yahoo, and Zoho through app passwords; CalDAV/Google Calendar through iCal. Read mail, draft, send with per-message approval, find free time, and create meetings.
- Daily briefings: summarize email and calendar for delivery to your Telegram/Zalo/Discord. You can preapprove read-only permissions for unattended schedules; external email sending and file writes can never be preapproved.
In development — priorities for everyday use
- System tray/background operation, startup with the OS, and OS notifications so schedules and channels continue when the window is closed.
- A 24/7 channel inbox: the server holds Telegram messages while your computer is off and delivers them when it returns. It does not call AI or hold your API keys.
- Scheduled external delivery through channels/email, advanced cron, file/webhook triggers, and quiet hours.
- Richer channels: image input, result files, quick replies, channel-based approvals, and the official Zalo Bot.
- OCR for scans/images, PDF/PowerPoint export from results, and multiple result files per task.
- Semantic knowledge search, automatic retrieval with page/sheet references, and user-editable long-term memory.
- Multi-step workflows with branches, approval gates, and resume points; currently an interactive design preview.
- Usage/cost statistics and budgets with automatic spending limits.
- In-app automatic updates with backup and recovery; code signing for Windows.
- Explicitly shared personal Chrome tabs and recovery of sign-in profiles after a crash.
- More reliable voice, more local models, and additional OS support.
Long-term plans
- AI applications with their own interfaces and interactive outputs, installed like plugins.
- Image creation/editing, followed by video.
- Teams and remote access across devices with member-level limits.
- Agent API and LLM Proxy with scoped keys and signed webhooks.
- Additional channels as needed, such as WhatsApp, Slack, Teams, and Facebook Page; only announced after real-account acceptance testing.
- Paid Library skills and skills with scripts in a controlled execution environment.
- Outlook/Microsoft 365 email through OAuth.
Troubleshooting and frequently asked questions
| Situation | What to do |
|---|---|
| The app asks for the API key after restart | Open AI connections → Edit, enable secure key storage. Linux requires GNOME Keyring or KWallet |
| Connection test lists models but a task fails | A model list does not prove remaining credit or generation access. Check the provider's balance and key permissions |
| AI errors when using tools | Use a model with function calling, or set the assistant to No tools |
| A schedule does not run | Keep SpinxAI open with a usable key. Check Schedules → History |
| Telegram/Discord bot does not reply | Check the app is open, channel status, and sender pairing. Enable Message Content Intent for Discord |
| A task is marked interrupted | The app closed during execution. Open Tasks → Needs you and retry. Tasks are not restarted automatically to avoid repeating actions |
| A PDF has no extracted text | It may be a scan. Use a text-based PDF or convert it to Word; OCR is not available yet |
| Windows SmartScreen warning | The Windows installer is not signed yet. For the official installer, select More info → Run anyway |
| Linux AppImage does not open | Install FUSE 2 (sudo apt install libfuse2), then make the file executable |
| On-device AI cannot load | Reduce context length, close other apps, or choose a smaller model to free RAM |
| Chrome Companion disconnects | Generate a fresh pairing code under Settings → Chrome Extension and paste it into Options after restarting either app |
| A screen says Interactive design preview | It is a future-feature preview; its data is not saved. Do not enter real keys or passwords |
Need help? Open Gateway → Diagnostics and export the JSON report, which excludes work content and keys. Send it with a description of the issue to SpinxAI support. If the app cannot start, provide runtime.log:
- macOS:
~/Library/Logs/net.spinxai.desktop/runtime.log - Windows:
%LOCALAPPDATA%\net.spinxai.desktop\logs\runtime.log - Linux:
~/.local/share/net.spinxai.desktop/logs/runtime.log
Ready to delegate your first task?
Start small, review the result, and build your own way of working with AI.