Phase 3: Agentic Function Calling (weather, timer, time)

[chatde]

What to build

Full tool-call pipeline: Gemma 4 returns a tool_calls block → bridge parses it → dispatches to Python function → result injected back into conversation → Vector speaks the answer.

Built-in tools: get_weather (wttr.in, no API key), get_time (system clock), set_timer (background thread + spoken alert).

Acceptance criteria

• tools/weather.py:get_weather(location) returns current conditions via wttr.in JSON (no API key)
• tools/timer.py:set_timer(seconds, label) starts background thread; Vector speaks when timer fires
• tools/time.py:get_time() returns formatted current date/time string
• bridge.py:_handle_tool_calls() dispatches to all three tools correctly
• End-to-end: "What's the weather in London?" → Gemma calls tool → Vector speaks answer
• Tool errors handled gracefully (Vector says "I couldn't get that" rather than crashing)
• ENABLE_TOOLS=true in .env activates tool calling; false skips cleanly

Blocked by

• Blocked by Phase 2: Live Vision + Conversation #1 (Phase 2 — live camera connection validated)

User stories addressed

• Vector can tell you the weather, set timers, answer real-world questions using tools