Example Use Cases

• "Find the newest invoice PDF in Downloads, summarize it, and save the summary."
• "Search this repository for the desktop automation implementation and explain the flow."
• "Open the browser, fill this form, submit it, and tell me what happened."
• "Check unread client emails, draft a professional reply, and leave it for review."
• "Take a screenshot and explain what is visible on the screen."
• "Remind me tomorrow morning to continue this release task."
• "Create a reusable Python tool for this file-processing workflow."
• "Search current web results, read the best source, and summarize it with a link."
• "Remember that I prefer concise Hebrew answers for this project."

Model Support

Smarti supports multiple providers from the settings screen:

• Google Gemini
• OpenAI
• Anthropic
• OpenRouter
• Groq
• NVIDIA NIM
• Cerebras
• Hugging Face
• DeepSeek
• Alibaba Qwen / DashScope
• Zhipu GLM
• Moonshot Kimi
• Mistral AI
• Together AI
• Perplexity
• xAI
• Local OpenAI-compatible servers, such as LM Studio

API keys can be stored through Keyring or Windows DPAPI. Local OpenAI-compatible servers can be configured without a cloud API key.

Safety and Control

Smarti is built around explicit control rather than blind automation. Sensitive capabilities are governed by settings, policy checks, and approval prompts. You can enable or disable tool groups, choose autonomy level, require approval for shell commands and extension installs, configure sandbox roots, redact logs, and control cloud-upload behavior.

Actions such as writing files, running commands, controlling the desktop, reading or sending email, using screenshots, installing external tools, or uploading local content to a cloud model are treated as sensitive and may require user approval.

Quick Start

Requirements for source runs: Windows, Python 3.10 or newer, and Git if you want to use Git tools.

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt
python smarti_core.pyw

To run without a console window:

pythonw smarti_core.pyw

Optional Capabilities

• Browser automation requires Google Chrome.
• MCP support requires Node.js and npx when running from source. Packaged releases include a private Node.js runtime.
• Voice input requires a working microphone, SpeechRecognition, keyboard, and PyAudio.
• OCR requires pytesseract, Pillow, and a system Tesseract OCR installation available in PATH.
• Usage-cost estimates are richer when litellm is installed.
• Live Visual Canvas is experimental and off by default. The WebEngine dependency is included in requirements.txt; enable קנבס חזותי מתקדם under Settings → Tools & Communication → Advanced. It uses an isolated local WebEngine profile; network, external files, downloads, popups, and media permissions are blocked by default. The separate אפשר תמונות HTTPS מהרשת בתוך קנבס setting permits HTTPS image resources only; navigation and every other external request remain blocked.
• Packaged releases include a private Python runtime used for custom Python tools, MCP support, and Skill dependency installs.

Building a Windows Release

The repeatable release recipe lives in scripts/ and packaging/. Build artifacts, downloaded runtimes, and generated installers are ignored by Git.

For the current release:

.\scripts\build_release.ps1 -Version 0.81.1

The build script:

• verifies that APP_VERSION matches the requested release version;
• creates or updates the build virtual environment;
• installs requirements.txt and requirements-build.txt;
• prepares private Python and Node runtimes;
• runs pip check for the build environment and private runtime;
• packages the app with PyInstaller;
• force-resynchronizes assets\ into the bundled _internal\assets folder and verifies the copied files by SHA-256;
• validates the release layout and installer path lengths;
• creates a portable ZIP and, when Inno Setup 6 is installed, a setup EXE.

For a fully clean rebuild:

.\scripts\build_release.ps1 -Version 0.81.1 -Clean -ForceRuntime

Release outputs are written to release\:

• SmartiAI-Agent-for-Windows-0.81.1-Setup.exe
• SmartiAI-Agent-for-Windows-0.81.1-win-x64-portable.zip

Automatic Updates

Smarti checks GitHub Releases, compares the latest release tag against APP_VERSION, and downloads the correct Windows asset for the current install type.

For V0.81.1, publish a GitHub release tagged V0.81.1 or 0.81.1 and attach:

• SmartiAI-Agent-for-Windows-0.81.1-Setup.exe
• SmartiAI-Agent-for-Windows-0.81.1-win-x64-portable.zip

Installed copies prefer a setup EXE. Portable copies prefer a portable ZIP. When GitHub provides a release asset digest, Smarti uses it for SHA-256 verification.

Project Structure

smarti_core.pyw          Source launcher and main entry point
smarti/                  Modular application code
  app.py                 Startup, splash screen, legal gate, and main window setup
  core.py                Compatibility facade that composes SmartiCore from domain mixins
  agent/                 SmartiCore domain modules: lifecycle, policy, tool calls, automation, email, web, memory, TTS
  runtime.py             Source vs packaged runtime and private toolchain resolution
  config.py              Tool schemas, categories, and default settings
  chat.py                Chat UI, tray behavior, voice input, updates, notifications, and background-task UI
  ui_pages.py            Settings, tools, task center, usage, developer logs, and about pages
  managers.py            Settings migration, memory, policy, audit, MCP, Skills, and tool registries
  workers.py             QThreads for agent work, voice, TTS, model loading, and validation
docs/                    Architecture notes and maintainer-facing documentation
assets/                  Logo, icons, fonts, sounds, and UI assets
packaging/               PyInstaller spec, Inno Setup script, and runtime version recipe
scripts/                 Release build and private runtime preparation scripts

Local Data

On Windows, new installs store runtime data under %APPDATA%\SmartiAI by default. Generated output files go to Documents\Smarti_Outputs when the Documents folder exists. Legacy project-root runtime files are copied forward on first run and left in place.

Common local files include settings, usage data, memory, chat history, logs, custom tools, MCP tools, Skills, browser session data, and attachment cache data. Sensitive and generated paths are excluded from Git.

License

Smarti is no longer licensed under the MIT License for new versions.

Starting from version V0.69.0, Smarti is licensed under the:

PolyForm Noncommercial License 1.0.0

This means:

• Smarti may be used for personal, educational, research, experimental, hobby, or other noncommercial purposes.
• Smarti may not be used for any commercial purpose without a separate written commercial license from the copyright holder.
• Commercial use includes use by a company, business, startup, contractor, consultant, service provider, paid product, SaaS service, commercial product integration, internal business use, resale, commercial distribution, repackaging, white-labeling, or any use intended to generate revenue or commercial advantage.
• Anyone who wants to use Smarti commercially must obtain prior written permission and a separate commercial license.

Earlier versions of Smarti that were previously published under the MIT License remain available under the MIT License for those specific versions only. This license change applies only to new versions published after this change.

The software is provided as is, without any warranty of any kind. Use of Smarti is entirely at the user's own risk. The copyright holder is not responsible or liable for any damage, data loss, file deletion, malfunction, privacy issue, security incident, financial loss, legal violation, incorrect action by an AI model, incorrect action by a tool, command, script, plugin, Skill, integration, third-party service, or any other harm caused directly or indirectly by installing, using, modifying, distributing, integrating, or operating the software.

Smarti may include or enable autonomous actions, computer control, command execution, file access, network access, third-party tools, artificial intelligence models, and automation of sensitive operations. The user is responsible for reviewing, supervising, limiting permissions, creating backups, using a sandbox or test environment when appropriate, and approving the software's actions.

For the full lic