Gemini Scribe

Appearance

Advanced Settings Guide ​

This guide covers the advanced settings available in Gemini Scribe, including model parameter tuning, API configuration, and developer options.

Accessing Advanced Settings ​

Advanced settings are hidden by default to keep the interface clean. To access them:

  1. Open Obsidian Settings
  2. Navigate to Gemini Scribe under Community plugins
  3. In the General section, scroll to the bottom and click Show Advanced Settings

Model Parameter Controls ​

Temperature Settings ​

Temperature controls the randomness and creativity of AI responses:

  • Range: 0 to 2.0 (dynamically adjusted based on available models)
  • Default: 0.7
  • Lower values (0.0-0.5): More deterministic, consistent responses
  • Higher values (1.0-2.0): More creative, varied responses

When to adjust:

  • Creative writing: Use higher temperature (1.0-1.5)
  • Technical documentation: Use lower temperature (0.2-0.5)
  • General chat: Default (0.7) works well

Top P Settings ​

Top P controls the diversity of word choices the AI considers:

  • Range: 0 to 1.0 (always fixed range for Gemini models)
  • Default: 1.0
  • Lower values (0.1-0.5): More focused, predictable responses
  • Higher values (0.8-1.0): More diverse, exploratory responses

When to adjust:

  • Focused analysis: Use lower Top P (0.3-0.7)
  • Brainstorming: Use higher Top P (0.9-1.0)
  • Balanced responses: Use default values

Dynamic Parameter Ranges ​

Gemini Scribe automatically discovers the parameter limits for your available models:

  • Temperature ranges adapt to the maximum supported by your models
  • Model-specific limits are enforced to prevent API errors
  • Real-time validation adjusts values that exceed model capabilities
  • Informational displays show the actual ranges and default values

API Configuration ​

Custom API Endpoint ​

Route all Google API requests through a proxy or gateway instead of hitting the public endpoint directly.

  • Setting name: Custom API endpoint
  • Default: empty (uses official Google endpoint)
  • When to use:
    • Corporate networks that block generativelanguage.googleapis.com or aiplatform.googleapis.com
    • Local reverse proxies for API key management or cost tracking
    • Regional mirrors for latency or compliance requirements
  • Scope: All seven Google GenAI SDK call sites are covered — chat, streaming, web fetch, Google Search grounding, RAG embedding, deep research, and token counting. Leaving one path unproxied while routing others is not possible with this setting.
  • Validation: The value is validated on blur; invalid URLs will show a warning notice and be cleared automatically.

Retry Settings ​

Configure how the plugin handles API failures:

Maximum Retries

  • Default: 3 attempts
  • Range: 0-10 retries
  • Purpose: Handles temporary network issues or API rate limits

Initial Backoff Delay

  • Default: 1000ms (1 second)
  • Range: 100-10000ms
  • Purpose: Time to wait before first retry (uses exponential backoff)

How retry works:

  1. First attempt fails
  2. Wait initial delay (e.g., 1 second)
  3. Second attempt fails
  4. Wait double the delay (e.g., 2 seconds)
  5. Third attempt fails
  6. Wait quadruple the delay (e.g., 4 seconds)
  7. Final attempt or success

Model Discovery ​

Model discovery is automatic — no configuration is required. On startup, the plugin fetches the latest available Gemini models from GitHub and caches the result for 24 hours. If the fetch fails, the bundled static model list is used as a fallback.

Both providers expose a Refresh model list button in Settings → General:

  • Gemini — bypasses the 24-hour cache and re-fetches the remote model list immediately. You can also trigger this from the command palette with Gemini Scribe: Refresh model list (gemini-scribe-refresh-model-list). Useful when a newly-published model doesn't appear yet.
  • Ollama — re-queries the Ollama daemon for any models you've pulled since the plugin loaded (ollama pull <name>). Use this instead of restarting Obsidian.

Performance Optimization ​

Context Management ​

In v4.0+, context is manually managed through session-based file selection:

Context File Selection:

  • Use @ mentions in chat to add files as persistent context
  • Context files are included with every message in the session
  • Start with 2-3 relevant files and add more as needed
  • Remove unused context files to save token budget

AGENTS.md - Vault Context:

  • Create AGENTS.md via "Initialize Vault Context" button
  • Provides AI with overview of your vault structure
  • Enables better file discovery without adding every file as context
  • Update periodically as your vault evolves

Optimization tips:

  • Start minimal (2-3 files) and expand as needed
  • Use AGENTS.md for vault-wide awareness instead of adding many context files
  • Let agent use tools to read additional files on-demand
  • Monitor token usage in long conversations
  • Use Flash models for faster responses

Model Selection Strategy ​

For Chat (Quality focused):

  • Primary: Gemini Flash Latest (default)
  • Alternative: Gemini 2.5 Pro for harder reasoning (requires billing)

For Completions (Speed focused):

  • Primary: Gemini Flash Lite Latest (default)
  • Alternative: Gemini Flash Latest if you want richer suggestions

For Summaries (Balanced):

  • Primary: Gemini Flash Latest (default)
  • Alternative: Gemini 2.5 Pro for long or technical documents

Best Practices ​

Parameter Tuning ​

  1. Start with defaults - They work well for most use cases
  2. Make incremental changes - Adjust by 0.1-0.2 at a time
  3. Test with your content - Different content types may need different settings
  4. Document your preferences - Keep notes on what works for different tasks

API Management ​

  1. Monitor usage - Check Google AI Studio for API quota
  2. Use appropriate models - Don't use Pro models for simple tasks
  3. Adjust retry settings - More retries for unreliable connections
  4. Enable fallback models - Ensures continued functionality

Model List ​

  1. Use Refresh model list in Settings → General (or run Gemini Scribe: Refresh model list from the command palette) to pick up newly published Gemini models without waiting for the 24-hour cache to expire
  2. Use Refresh model list (Ollama provider) after pulling new models with ollama pull
  3. Check your API key if the model list looks empty or stale

Troubleshooting ​

Parameter Issues ​

Temperature/Top P not taking effect:

  • Check if model supports the parameter range
  • Verify settings are saved (restart Obsidian if needed)
  • Look for validation warnings in notices

Extreme responses:

  • Lower temperature if too random
  • Adjust Top P if responses are too narrow/broad
  • Reset to defaults if unsure

API Problems ​

Frequent failures:

  • Increase retry count
  • Extend initial backoff delay
  • Check API key permissions
  • Verify internet connection

Slow responses:

  • Reduce number of context files in session
  • Use faster models (Flash variants)
  • Start new session to clear conversation history
  • Lower retry count for quicker failures

Model List Issues ​

Models not appearing or stale:

  • For Gemini: click Refresh model list in Settings → General (or run the Gemini Scribe: Refresh model list command) to bypass the 24-hour cache; check API key validity and network connectivity if it still fails
  • For Ollama: click Refresh model list in Settings → General after pulling new models
  • If the list still looks wrong after refreshing, restart Obsidian

Security Considerations ​

API Key Protection ​

  • Secure storage - Your API key is stored using Obsidian's SecretStorage API, not in plaintext data.json
  • Never share your API key
  • Use environment variables for development
  • Rotate keys regularly as a security practice
  • Monitor usage for unauthorized access

Data Privacy ​

  • Direct API calls - Data goes only to Google
  • Local storage - Chat history stays in your vault
  • No third parties - No intermediate servers involved
  • Encryption - Consider vault encryption for sensitive data

Safe Settings ​

  • Review parameter changes - Extreme values may produce unexpected results
  • Test with non-sensitive data - Before using on important content
  • Backup regularly - Especially when experimenting with settings
  • Use version control - Track changes to your vault

Advanced Use Cases ​

Research Projects ​

Temperature: 0.3-0.5 (focused analysis)
Top P: 0.7-0.9 (balanced diversity)
Context Files: Research question, literature review, methodology notes
Model: Gemini 2.5 Pro (best quality)

Creative Writing ​

Temperature: 1.0-1.5 (high creativity)
Top P: 0.9-1.0 (maximum diversity)
Context Files: Character profiles, world building notes, plot outline
Model: Gemini 2.5 Pro (best quality)

Technical Documentation ​

Temperature: 0.2-0.4 (consistent style)
Top P: 0.5-0.8 (focused responses)
Context Files: API specs, architecture docs, style guide
Model: Gemini Flash Latest (fast, accurate)

Brainstorming Sessions ​

Temperature: 1.2-1.8 (maximum creativity)
Top P: 0.9-1.0 (diverse ideas)
Context Files: Project overview, relevant background materials
Model: Gemini 2.5 Pro (creative capability)

Support ​

For issues with advanced settings:

  1. Check the troubleshooting section above
  2. Review the main documentation for basic setup
  3. Report bugs on GitHub Issues
  4. Join the discussion in the Obsidian community

Advanced settings provide powerful control over AI behavior. Start conservative and adjust based on your specific needs and content.