Cee is a multi-agent system, meaning under the hood, it’s composed of a main AI agent and several specialized sub-agents. As a user, you will just interact with “Cee” as a single chat assistant.
Main Agent
- Main Documentation Agent (Cee): This is the primary agent you converse with and it has access to Countly documentation.
- Intent Analysis and Orchestration: Main agent can interact with user in two ways:
- Answer directly using documentation, or
- Hand off to a specialized agent if your request coincides with one of these agents' specialty.
Specialized Agents
There are 4 specialized agents as a companion to main AI agent:
- Cohorts Agent
- Funnels Agent
- Drill Agent
- Drill Insight Agent
These agents (except for the Drill Insight Agent) has these features:
- Guidance: Agents will guide you to the specific feature section they are specialized on.
- Interaction: They can ask you clarifying questions if they think it will be necessary.
- Access to metadata: They have access to your app's metadata so they can interpret your vague comments and guess the event you are talking about.
- Form filling: They can fill Cohort, Funnel or Drill forms from your descriptions.
- Giving onus to user: They will leave the execution or saving of those forms to you, so that you have full control.
- Detecting user action: They can detect manual changes you made in the forms and includes this information when you chat to them without a need for reminding.
- Offers raw data: Agents has an option to show you how they are filling those form by offering the raw data they generate.
Drill insight Agent
After any Drill query runs, you can click Drill Insights button to have Cee scan the result and post a concise insight summary in chat.
Output includes:
- Key spikes, drops, sustained trends.
- Segments or segment combos that over/underperform.
- Business-style interpretations (e.g. spike tied to release; weak cohort to improve).
- Action buttons (e.g. Create Cohort for Segment X, Build Funnel from Event Y, Drill further by Property Z).
Techniques used:
- Correlation: Finds attributes or attribute combinations strongly linked to higher/lower values.
- Trend & Anomaly Detection: Shows direction (up, down, stable), flags unusual spikes/dips or outlier days.
- Seasonality & Changepoints: Points out repeating patterns (e.g. weekends) and structural shifts (e.g. “after July 15 values moved to a higher band”).