v0.10.0

Added

• Custom Endpoint URLs for All AI Providers — every provider now has an "Endpoint URL" field in AI Settings, enabling use of custom/self-hosted API endpoints (e.g. LM Studio, vLLM, Azure ML Studio, LiteLLM). Previously only Ollama supported a custom host. The field shows the provider's default URL as placeholder text. Model listing for custom endpoints uses the OpenAI-compatible /models format with automatic fallback to Ollama format. See features request #2 for details.
• API Key Visibility Toggle — added an eye icon button next to the API key field to toggle between hidden (default) and visible text.
• Automatic Error Recovery — when AI-generated code causes a compilation error (e.g. "Parse error: unexpected token"), the error is automatically sent back to the AI with a request to fix the issue. The AI then produces corrected code, creating a self-healing feedback loop.

Changed

• Default Temperature — changed the default temperature from 0.0 to 0.1 for new installations to encourage more creative responses while still being mostly deterministic. Existing users keep their current temperature setting.
• API Key Optional for Custom URLs — the "Required" marker is no longer shown on the API key field when a custom endpoint URL is configured, since many local inference servers don't require authentication.
• Model Selection Optional for Custom URLs — when using a custom endpoint URL, the model dropdown selection is no longer mandatory. Some inference servers (e.g. LM Studio with Claude compatibility) handle model routing internally.

Fixed

• Compile Button Requires Double Click After Edit — fixed update scheduling so editor changes are always applied before compile triggering. A single click on Compile now reliably compiles the latest code without needing a second click.
• Custom Ollama Host Without API Key — fixed a bug where setting a custom Ollama host was silently ignored when no API key was configured, falling back to the default localhost:11434 endpoint.
• Adapter Kind Override — the user-selected provider now always determines the API format used (e.g. OpenAI's /chat/completions vs Ollama's /api/chat), regardless of what the genai library infers from the model name. This fixes issues when using non-standard model names with providers like LM Studio.
• Model List Empty on Startup — the model selector was empty after app restart because the model fetch system's run_if guard prevented it from running on the first frame. Models are now fetched automatically on startup.
• Model List Empty After Provider Switch — switching the AI provider via the chat header selector now triggers a model list reload instead of showing an empty model selector.
• Chat History Overwrites Draft — pressing Arrow Up/Down in the chat input no longer wraps cyclically and discards unsent text. The current draft is preserved and restored when pressing Arrow Down past the newest history entry.
• "Select a model" Warning with Custom URL — the warning "Select a model in ⚙" no longer appears when using a custom endpoint URL, since model selection may be handled server-side.