Avenue launches virtual teammate powered by ChatGPT

Paul Noble
March 13, 2023

Avenue helps operations teams do their best work. That’s why we created Avenue copilot, your all-in-one ops, analytics, and customer service teammate! Powered by OpenAI’s ChatGPT, copilot can help you brainstorm ways to solve ops issues, write sql, and answer questions about Avenue.

Copilot is your all-in-one teammate for:

  • Ops: generate ideas around the types of issues your organization should track. Get help brainstorming monitor ideas, develop existing ones, and rounding out playbooks.

  • Analytics: write SQL queries and answer SQL questions

  • Customer Service: answer questions about Avenue

Tips and Tricks

  • Try to give copilot as many details as possible when providing it context to get the best recommendations.

  • If you need help writing SQL queries, make sure to provide copilot information about your table schema, when necessary. copilot has no context of the structure of your database and will otherwise hallucinate table and columns names unless you provide them.

  • If you don’t like the monitor recommendations that copilot outputs, or you want more options, ask for more! “copilot could you generate more recommendations?”

  • If you want more information on a recommendation, or it’s unclear what the monitor should be for, you can ask copilot to expand on an option. “copilot expand on the first option please.”

  • If you already have an idea for a new monitor and want help creating it, start by telling copilot your monitor idea and try collaborating with copilot. “Copilot I had an idea for a fraud pattern alert monitor, could you help me out?” “Can you help me with writing a SQL query for a monitor idea I had?”


How do I get started using copilot?

You can start interacting with copilot on our app by visiting https://app.useavenue.com/co-pilot or by clicking the “copilot” link in the apps sidebar. No set up required!

What is being released today?

We are releasing our AI assistant copilot, which can answer a variety of questions you may have and collaborate with you on creating monitors.

How do I write SQL with copilot?

Copilot has no context on your database schema or the specific SQL server you are using. To get effective support for this, make sure to provide Copilot with detailed instructions and consider providing specific table and column names that you know you’ll need for the query. It may also be helpful to explain any foreign key relationships that may be relevant. If you don’t provide this information, copilot will attempt to provide you with a decent starting point for your query, but you will have to fill in your table and column names.

Are conversations with copilot moderated?

Yes. Conversations with copilot are sent to OpenAI’s API and thus are subject to their usage guidelines. Your messages to copilot are not stored by OpenAI but they will be preprocessed for content that does not align with their guidelines. In general, think of copilot as your coworker, don’t say anything to copilot you wouldn’t say to a coworker.

What happens to my conversation data when interacting with copilot?

Messages sent to our assistant copilot are sent to the OpenAI api for processing, but they are not stored by OpenAI or used in any future training sets by OpenAI. See their GPT-3.5 FAQ or their data usage policy for more details. Avenue does save encrypted versions of your messages that we may use for refining copilot in the future.

Can copilot create monitors for me?

copilot can create monitor drafts with titles and descriptions that the assistant suggests to you. copilot can also help you create SQL queries for your monitors, but you’ll need to copy and paste them for now. Look out for future updates to copilot's monitor editing capabilities.

Can I talk to copilot in Slack?

Not yet. But look out for more updates to copilot, including new ways of interacting with the assistant!

Is copilot always correct?

copilot is very capable and can help you with many tasks. However, current LLMs like GPT-3.5 are not perfect and are subject to hallucination. This means that if you ask copilot a question that it cannot answer based on the context it has, it will try and make up an answer. We’ve tried to encourage the model to communicate when it is unable to answer a particular question, but it can still happen. While copilot is certainly a useful tool, for business critical tasks it is always a good idea to verify the information copilott is giving you.

Interested in making an AI assistant for your platform? See our engineering blog Designing an LLM Chatbot for Ops Teams

Software Engineer