Download OpenLive v0.1.9
| Platform | Download |
|---|---|
| 🍎 macOS (Apple Silicon + Intel) | v0.1.9 — Mac (.dmg) |
| 🪟 Windows (64-bit) | v0.1.9 — Windows (.exe) |
On first launch macOS may take a few seconds to verify the app — that's normal.
[0.1.9] - 2026-07-11
Added
- A dozen more model providers. Alongside Anthropic, OpenAI, and MiniMax,
OpenLive now speaks to Google Gemini, xAI Grok, DeepSeek, Mistral, Groq, Cerebras,
Together, Fireworks, OpenRouter, Perplexity, and Ollama (local and cloud). Paste a
key and the model list loads live from the provider — nothing hardcoded. - Separate vision model (optional). If your live model can't see, point OpenLive
at a dedicated vision model under Settings → Vision model. Camera and screen frames
are described by that model and handed to your live model, so a fast text-only
model can still watch your screen. Leave it off and the live model sees for itself. - Real vision capability in the picker. The
visionbadge now comes from actual
provider / models.dev metadata rather than a name guess, and the picker warns when
the selected model can't accept images.
Changed
- A third wire adapter (OpenAI Chat Completions) joins the Anthropic and OpenAI
Responses adapters, so most hosted providers work through one code path. - Snapshot model defaults refreshed to current IDs (e.g. DeepSeek V4, Grok 4.5),
preferring fast vision-capable models for the voice + camera loop.