Knowledge Base Settings

The knowledge base chat has a small set of admin-only settings that control how the agent answers questions across your workspace. You'll find them at Knowledge Base → Settings.

The settings are intentionally minimal. The retrieval engine is hands-off by design — Opisense picks chunking, embedding, and search behavior so you don't have to. What you can configure is how the agent talks.

Default model

The AI model used to answer questions. Members can override this per-conversation with the model selector in the chat input, but the default applies until they change it.

Pick the model that matches your needs:

• Higher-capability models (Opus, GPT-class large) are better at reasoning across many sources, complex synthesis, and nuanced answers
• Lower-cost models (Haiku, Sonnet, smaller GPT-class) are faster and cheaper for straightforward lookups

If you're unsure, start with a balanced default and adjust based on what kinds of questions your team asks.

System prompt

A free-form prompt that's prepended to every conversation. Use it to shape how the agent should behave for your workspace.

Common things to include:

• Tone — "Always answer in a professional, concise tone."
• Scope — "Only answer using documents in this workspace; never speculate beyond what's cited."
• Format preferences — "For policy questions, always quote the source verbatim and link back to the section number."
• Audience — "Our team includes non-technical staff; explain technical concepts in plain language."

The system prompt is shared across the workspace, so it should reflect your team's general needs rather than personal preferences.

Welcome message

The first message the agent shows when someone opens a fresh KB chat conversation. Use it to:

• Set expectations ("I can answer questions about anything in our knowledge base — try asking about policies, processes, or product specs.")
• Highlight new content ("I just learned about the Q4 launch playbook — feel free to ask.")
• Suggest entry points ("Some good starter questions: 'What's our refund policy?', 'Summarize last month's all-hands.'")

Keep it short — one or two sentences. The welcome message is the first thing users see, so a clear pitch helps.

What you can't configure

The knowledge base intentionally hides a lot of retrieval knobs that older versions exposed (search depth, rerank, recency bias, agentic effort levels, breadth vs. depth modes). These are now managed automatically — the agent picks the right strategy based on the question.

If you need behavior the system prompt can't shape, contact support. The retrieval engine is configurable on the backend; some changes can be made by your account team.

Permissions

Settings are admin-only. Members can:

• Use the chat
• Pick a different model in the model selector for their own conversations

Members can't:

• Change the default model, system prompt, or welcome message
• See the system prompt's full text (it runs invisibly)

If you need to delegate KB administration, promote a teammate to admin from the workspace member list.