AI Settings

The AI Settings tab lets you customize how the AI behaves — from the words it uses in responses to how it searches your documents. Most users won't need to change these settings, but they're available if you want to fine-tune the experience.

System Prompt

The system prompt is a set of instructions that tells the AI how to behave. The default prompt works well for general document questions, but you can customize it for your use case.

For example, you might change it to:

• "You are a legal assistant. Always cite specific clauses when answering."
• "You are a study helper. Explain concepts simply and give examples."
• "Answer in bullet points. Keep responses under 200 words."

To edit the system prompt, type your instructions in the text area and click Save System Prompt. If you want to go back to the default, click Reset to Default.

Model Parameters

These settings control how the AI generates responses. Changes to most of these require a model reload, which takes a few seconds. You'll see a yellow banner when a reload is needed.

LLM Context Length

This controls how much text the AI can consider at once — both your question and the document passages it retrieves. Options range from 4,096 to 32,768 tokens.

• Lower values use less memory and respond faster
• Higher values let the AI consider more document passages at once

The default of 8,192 is a good balance for most users.

Temperature

This controls how creative or predictable the AI's responses are, on a scale from 0 to 2.

• Lower values (0–0.3) — More focused and consistent answers. Good for factual lookups.
• Higher values (0.5+) — More varied and creative responses. Better for brainstorming or summaries.

The default of 0.1 keeps answers closely tied to your documents, which is usually what you want.

Embedding Context Length

This controls the maximum size of passages that can be processed during upload. You generally don't need to change this unless you're using very large chunk sizes. It does not require a model reload.

Search Settings

These settings control how the AI finds relevant passages in your documents when you ask a question.

Max Chunks to Retrieve

How many passages the AI pulls from your database when answering a question. The default is 5.

• Fewer passages — Faster responses, but the AI might miss relevant information.
• More passages — More thorough answers, but responses take longer and may include less relevant material.

Similarity Threshold

How closely a passage must match your question to be included, expressed as a percentage. The default is 80%.

• Higher threshold — Only very relevant passages are used. Answers are precise but may miss related information.
• Lower threshold — More passages qualify. Answers are broader but may include less relevant material.

Filtering Mode

Controls what happens when passages don't meet the similarity threshold:

• Adaptive (Recommended) — Always returns at least some passages, even if none meet the threshold perfectly. This ensures the AI has something to work with, even for unusual questions.
• Strict — Only returns passages that meet the threshold. Answers are more reliable but may be empty if nothing matches well enough.

Show Citations

When enabled, the AI will show which document passages it used to generate each response. This lets you verify answers and read the source material yourself. Enabled by default.

Switching AI Models

The model dropdown shows which AI models are available. The free version comes with a capable default model. If you have a paid license, you can upload additional models in GGUF format for different use cases — for example, a model specialized for medical or legal language.

To switch models, select one from the dropdown and click Activate Selected Model. The model will reload automatically.

Next Steps

• Licensing — Unlock the ability to upload custom models
• Ask questions — Try out your adjusted settings