Phase 2a: Camera pipeline — capture frame → Ollama vision → response

[chatde]

Parent PRD

See plans/vector-gemma4-roadmap.md Phase 2

What to build

The minimal end-to-end path: Vector camera frame → base64 JPEG → Ollama gemma4:e4b vision endpoint → text response returned. This is the core pipe that everything else in Phase 2 depends on. No TTS, no dedup, no animations yet — just verify the data flows correctly.

Acceptance criteria

• capture_frame() returns a non-None base64 JPEG string when Vector camera is active
• ask_gemma(prompt, image_b64=frame) hits Ollama /api/chat with the images field populated
• Response is a non-empty string (not an error message) for a live camera frame
• Works with gemma4:e4b pulled to local Ollama
• python scripts/test_connection.py passes all 3 checks (Ollama, vision ping, Vector)

Blocked by

None — can start immediately (requires Vector awake + ollama pull gemma4:e4b complete)

User stories addressed

• Vector sees through its camera and Gemma 4 processes what it sees locally