Skip to main content
AI Assist is the menu option you use to open KAI, Kameleoon’s AI-powered assistant. KAI helps you interact with the Kameleoon platform in two ways: it answers questions about the platform and your A/B test data, and it takes direct action on your account, such as creating an experiment or a feature flag. Both capabilities are native to KAI: any user with a Kameleoon account gets them without installing or connecting anything.

​
How to access KAI

Open KAI from two places in the platform:
  • Click the sparkle icon in the top-right corner > Ask a question.
  • From an experiment’s Results page, click the three-dots menu > Analyze with AI Assist to open KAI with that experiment’s data already loaded.

​
Answer questions about the platform and your data

KAI answers questions using two sources.
  • Documentation: KAI retrieves the most relevant developer and user documentation to answer questions about how the platform works.
  • Experiment data: When a question depends on your account, such as a request to summarize an experiment’s results, KAI makes a secure API call with JSON Web Token (JWT) authentication to fetch only the data your account has permission to view:
    • Experiment metadata: Name, website, duration, objectives, and targeted segment.
    • Goal metadata: Names and types of the experiment’s goals.
    • Segment metadata: The conditions that define who the experiment targets.
    • Experiment results: Conversion rates, significance, confidence intervals, and health checks such as Sample Ratio Mismatch (SRM).
KAI generates responses with a large language model (LLM) from a third-party provider, using only the documentation and data it retrieves for your query, which keeps its answers scoped to your question instead of pulling in unrelated information.

​
Take action through the MCP server

KAI carries out actions on your Kameleoon account through the Kameleoon MCP server, the same server that powers IDE integrations. KAI includes this access by default, so ask it to:
  • Create, duplicate, or run the full lifecycle of an experiment.
  • Create and manage feature flags, including delivery rules and typed variables.
  • Create goals and audience segments, and bind segments to experiments.
  • Retrieve an experiment’s raw variation code and results.
KAI uses the same tools documented in the Kameleoon MCP server reference, so any tool listed there is available to KAI. Invoke a tool in plain language, for example, “Create a feature flag called new_search for site d1alzzxd7k,” or type a slash command with the tool’s exact name, such as /experiment_create, to run it directly.
For an automated summary instead of a conversation, use the Learnings tab’s AI Summary on an experiment’s Results page. It generates insights and recommendations automatically, without a prompt.

​
Data privacy and security

Kameleoon protects your data whether KAI is answering a question or taking an action through the MCP server.
  • Documentation access: KAI only uses Kameleoon’s publicly available documentation to answer questions. It does not access external data sources.
  • User data access: KAI’s secure API calls use JWT authentication, so KAI can only access data tied to your authenticated account. Kameleoon logs and monitors each data request to enforce compliance with its security policies.
  • Data minimization: KAI accesses only the minimum data it needs to answer your question or complete an action.
  • MCP actions: Actions that KAI takes through the MCP server run under your account’s own permissions. KAI cannot perform an action your account isn’t authorized to perform.

​
Limitations

  • Scope of responses: KAI answers questions related to the Kameleoon product. It cannot answer questions unrelated to Kameleoon’s services, tools, or your experiment data.
  • Accuracy of information: KAI’s answers depend on the available documentation and data. Kameleoon continuously updates its documentation to improve response accuracy.

​
Common concerns

​
Will KAI access my data without permission?

No. KAI only accesses data your account has permission to view. Its data requests use JWT-authenticated API calls, and Kameleoon logs and monitors every request.

​
Can KAI access data from other users?

No. JWT authentication limits KAI to data associated with your account. KAI cannot view or retrieve other users’ data.

​
Is my experiment data safe when I use KAI?

Yes. Experiment data shared with the LLM provider for takeaways consists of aggregated data with no personal identifiers linked to your visitors.

​
Can I turn off KAI?

Yes. A super admin can disable the AI-powered reporting setting under Organization > Settings > Data privacy. Turning off this setting stops Kameleoon from sharing experiment data with the LLM provider. See Organization page for details.

​
Use cases

Use KAI to run a basic analysis of your test results, identify an experiment’s key metrics, explain unexpected results, get a recommended next step, or create and manage experiments and feature flags without leaving the chat. Example prompts:
  • Summarize the key findings of this experiment.
  • Which of the attached goals should we prioritize in evaluating the experiment’s effectiveness?
  • How do you explain the test’s observed results?
  • What could be a follow-up experiment?
  • Create a feature flag called new_search for site d1alzzxd7k.
  • Enable feature flag new_search in production.
For more prompt ideas, see Kameleoon’s blog post on writing AI prompts for A/B testing teams.

​
Contact us

For further questions about KAI, contact Kameleoon support at support@kameleoon.com.