AI Configuration

AcaTrove's AI features are powered by large language models from OpenAI and Anthropic. The AI configuration page lets administrators control which models are used, set parameters, configure rate limits, and manage API keys. Admin role is required.

Accessing AI Configuration

Navigate to /admin/ai-config from the admin sidebar.

AI configuration page with model settings and parameter controls

Model Selection

AcaTrove supports multiple AI model providers. For each AI feature, you can select the model to use:

• Default model -- The model used for general AI features (Paper Chat, Q&A, Writing Assistant). Options include GPT-4, GPT-4o, Claude 3.5, and Claude 3 Opus.
• Embedding model -- The model used for vector embeddings and semantic search. AcaTrove uses Sentence-Transformers (all-MiniLM-L6-v2, 384 dimensions) by default.
• Specialized models -- Some features allow overriding the default model. For example, you might use a faster model for quick suggestions and a more capable model for detailed analyses.

Parameter Settings

Adjust AI behavior through parameter controls:

• Temperature -- Controls response creativity. Lower values (0.1-0.3) produce more focused, deterministic responses. Higher values (0.7-1.0) produce more varied output. The default is 0.3 for research-oriented features.
• Max tokens -- The maximum length of AI responses. Increase for features that generate long-form content (outlines, syntheses). Decrease to encourage concise responses.
• Context window -- How much surrounding context is included when sending documents to the AI. Larger windows produce more informed responses but consume more tokens.

Rate Limits

Configure usage limits to manage costs:

• Per-user daily limit -- The maximum number of AI requests a single user can make per day. Set to 0 for unlimited.
• Workspace monthly limit -- The total number of AI requests across all users per billing month.
• Alert threshold -- The percentage of the monthly limit at which administrators receive a warning notification.

When a user hits their daily limit, they receive a clear message explaining the restriction. When the workspace approaches its monthly limit, administrators are notified.

API Keys

If your institution has its own OpenAI or Anthropic API keys and prefers to use them:

1. Navigate to the API Keys section of the AI configuration page.
2. Enter your API key for the desired provider.
3. Select which features should use your key versus AcaTrove's default.
4. Click Save.

API keys are encrypted at rest and never displayed in full after saving.

Feature-Specific Settings

Some AI features have additional configuration options:

• Paper Chat -- Memory duration (default: 6 hours), maximum conversation length, and annotation types enabled.
• Outline Generation -- Default template set, citation style preference, and export format.
• Multilingual Writing -- Default source and target languages, formality level.

Tips

• Start with default settings and adjust based on your team's feedback and usage patterns.
• Lower the temperature for grant writing features where precision matters; raise it slightly for brainstorming tools.
• Monitor the RAG Analytics page to understand how AI performance relates to your configuration choices.
• Set rate limits conservatively at first and increase them as you understand your workspace's usage patterns.