🔧 BaseTool — The Contract
BaseTool is the interface that makes the tool system extensible. Four Required Attributes and the LLM already knows how to use the tool.
📌 BaseTool Interface
- •name: str — unique identifier ('google_calendar')
- •description: str — for the LLM to decide when to use it
- •parameters: dict — JSON Schema for the arguments
- •async execute(**kwargs) -> str — implementation
- •registry.py automatically discovers tools in tools/
💡 Practical Tip
Write the description as if it were for an intelligent human. The LLM reads exactly this to decide when to call the tool.
🔍 Registry and Automatic Discovery
registry.py uses Python reflection to automatically discover all classes that extend BaseTool. No manual registration code.
📌 How Auto-Discovery Works
- •Scans all files in tools/*.py
- •Import each module with importlib.import_module()
- •Inspects classes with inspect.getmembers()
- •Filter for BaseTool subclasses with issubclass()
- •Instantiates and registers in the tool_registry dict
💡 Practical Tip
Put a tool in tools/ and restart Jarvis. It becomes available automatically. No need to edit Agent or manually edit the registry.
📅 Google Workspace
Google Workspace integration connects Jarvis to the world's most common productivity infrastructure. Calendar, Drive, Gmail, and Sheets in native tools.
📌 Google Workspace Tools
- •GoogleCalendarTool: create, list, update, and cancel events
- •GoogleDriveTool: upload, download, search, and share files
- •GmailTool: read, send, archive, and categorize emails
- •GoogleSheetsTool: read and write spreadsheets
- •Authentication: OAuth2 with automatic token refresh
💡 Practical Tip
Use service accounts for automation without human interaction. For personal use, OAuth2 with a token saved in .secrets is simpler.
🐙 GitHub Integration
The GitHubTool turns Jarvis into a development collaborator. Issues, PRs, commits, and code search via API.
📌 GitHub Capabilities
- •List and create issues with labels and assignees
- •Review PRs: view diffs, comment, approve, or request changes
- •Search code by default in a specific repository
- •Make simple commits (e.g., update a config file)
- •Check CI/CD status on the latest commit
💡 Practical Tip
Configure GITHUB_TOKEN with the minimum required scope. For read-only access, use only 'repo:read'. For write access, add the full 'repo' scope.
🌐 BrowserTool — Web Scraping
BrowserTool uses headless Playwright to give Jarvis real-time access to the web. Search, extract, and browse with user approval.
📌 Browser Capabilities
- •web_search(): searches DuckDuckGo or Google with structured results
- •fetch_page(): extracts text from any URL (with injection detection)
- •take_screenshot(): visual capture of pages for analysis
- •fill_form(): fills out forms with explicit approval
- •All actions go through the level 2 approval gate
💡 Practical Tip
Always run web page content through the injection scanner before adding it to the context. Malicious pages may contain indirect injection.
⚡ Creating Custom Tools
The INTELECTO extension model is deliberately simple. Fewer than 50 lines to turn any API into a Jarvis tool.
📌 Tool Template in 5 Steps
- •1. Create tools/minha_tool.py
- •2. from intelecto.tools.base import BaseTool
- •3. class MinhaTool(BaseTool):
- •4. Define name, description, parameters (JSON Schema)
- •5. async def execute(**kwargs) -> str: # your logic
💡 Practical Tip
The description parameter is the most important. Write it like a function docstring: what it does, when to use it, and what it returns. The LLM reads it literally.
✅ Module 4.2 Summary
Next:
4.3 Automation and Webhooks