Experts
An expert is an AI assistant backed by your curated knowledge. You control what it knows by assigning topics, and your team interacts with it through the AI tools they already use.
What is an expert?
An expert is the thing your team talks to. It has a name, a description, and access to one or more topics. When someone asks a question, the expert answers from the knowledge in those topics and nothing else.
You might create a "Sales Enablement" expert that knows about deal attribution, partner processes, and competitor positioning. Or an "Engineering Onboarding" expert that knows your architecture, conventions, and setup guides. Each expert is scoped to exactly the knowledge you assign to it.
Creating an expert
- Go to the Experts page in the Obris dashboard.
- Click Create expert.
- Give it a name that tells your team what it helps with (e.g. "Sales Enablement," "Engineering Onboarding").
- Optionally add a description to remind yourself what the expert covers.
- Select topics the expert should have access to. The expert can only answer from the knowledge in these topics.
- Click Create expert.
Topic scoping
Experts only see the topics you assign to them. This is intentional.
Why scoping matters:
- A customer-facing expert should not have access to internal pricing discussions.
- An onboarding expert does not need your sales playbook.
- A support expert for Product A should not answer questions about Product B.
You control the boundary. If an expert says something like "I don't have information about that," it means the knowledge is not in any of its assigned topics. Either add it to an existing topic or create a new one and assign it.
Changing scope: You can add or remove topics from an expert at any time from the expert's settings page. Changes take effect in the next conversation.
Deploying an expert
Once you create an expert, you deploy it by connecting it to an AI tool. There are three ways:
Remote MCP - Connect to Claude, ChatGPT, or any MCP compatible client using OAuth. Your team authorizes once and the expert is available in their conversations. See Remote MCP Connectors.
Custom GPTs - Build a ChatGPT Custom GPT backed by your expert's knowledge. Good for sharing a focused expert with your team or customers. See Custom GPTs.
Local MCP - Run the Obris MCP server on your own machine using an API key. Useful if you want to use the expert yourself without sharing access. See Local MCP Server.
One expert can be deployed to multiple AI tools at the same time. Your sales team can use it in ChatGPT while your engineering team uses it in Claude Code.
The feedback loop
After each conversation, the expert asks for a rating and captures anonymous feedback. Obris logs every session and surfaces:
- Conversation summaries - What the conversation was about and how it went.
- Friction areas - Where the user got stuck or where knowledge was missing.
- Ratings - How well the expert answered, on a scale of 1-10.
Low ratings prompt an optional anonymous form for more detail. This tells you what to fix without requiring anyone to file a ticket or write a report. Your expert gets better because you can see exactly where it falls short.
Tips for effective experts
One expert per domain, not one expert for everything. A focused expert gives better answers than a generalist. "Partner Deal Attribution" is better than "Everything Sales."
Start small, then expand. Create an expert with 2-3 topics. Let your team use it for a week. Check the feedback to see what is missing. Add knowledge, adjust scope, repeat. The feedback loop is the product.
Review the feedback regularly. Low-rated sessions and friction areas tell you exactly where to improve. Each gap is a piece of knowledge you should add.