AE CLI
Reading guide
Parent section: Developer Tools. Search keywords: AE CLI, command line, installation, login, authorization, Skill.
- Connect through a tool protocol instead: MCP.
- Learn about reusable task methods: Skill.
- Start interactions in the platform: Conversations.
- Set up recurring tasks: Automations.
The original instructions on this page follow below.
Let AI use AE directly to help you manage tracking, query data, review experiments, and follow up on operation tasks.
Your AI is good at analyzing problems, but if it can't see the projects, reports, and experiments in AE, it has to wait for you to copy the data out before it can give you suggestions.
After you install AE CLI, AI can use AE directly within the scope of your permissions. Just say clearly what you want to do, such as "Find out why new users dropped yesterday", "Summarize the results of this experiment", or "Check recent tracking errors", and AI finds the right capability on its own, reads the results, and continues the analysis.
You don't need to learn the command line or remember which commands AE CLI has. After a one-time installation and authorization, just talk to AI in natural language as usual.
Who it's for
Product and data analysts can have AI find existing reports, read data, compare trends, and then turn the results into conclusions.
Tracking and data teams can manage tracking plans, verify events and properties, and view collection errors, and can also have AI help generate plans and sample code.
Growth and experimentation teams can query experiment configurations, metric trends, and experiment reports, with less switching back and forth between pages.
Operations teams can view journeys, tasks, campaigns, channels, and performance data, and quickly put together execution status and review materials.
Data development teams can view tables, Flows, integration tasks, and run status to support daily troubleshooting and release checks.
Quick install
Method 1: Let AI install it for you
Send the following text to Codex, Claude Code, Cursor, WorkBuddy, Trae, or another AI agent that can operate a terminal:
Install AE CLI and the companion Skills for me. First check whether Node.js is at least version 20;
after installation, verify ae-cli --version, the login host, and auth status in order.
Do not print or copy any token. If a step fails, keep the original error, stop, and tell me. Do not skip it.
After installation, we recommend closing and reopening your AI tool before you start your first task.
Method 2: Manual installation
Before you install, check Node.js. To run both AE CLI and the Skill installer, we recommend Node.js 20 or later:
node --version
If your version is earlier than 20, upgrade or switch Node.js first. With an older version, Skill installation may fail with errors related to EBADENGINE or styleText.
npm install -g @thinkingai/ae-cli
npx -y skills add ThinkingAIAgenticEngine/ae-cli -g -y
After installation, confirm the version first, and then log in to your AE environment:
ae-cli --version
ae-cli auth login --host https://your-ae-host.example.com
ae-cli auth status
Your browser opens the authorization page. After you complete authorization, auth status should return the logged-in status and the current host. If AI still uses old commands after you install the Skills, restart the AI tool or reopen the task so that it reloads the Skills.
Install curated solution Skills
ThinkingAI has distilled years of service experience into Skills, which you can have AI install for you directly. They cover everything from LTV analysis to anomaly diagnosis.
Install https://github.com/ThinkingAIAgenticEngine/scenario-skills for me
Start your first task
The first time you use it, start with a read-only connection check instead of querying full data right away.
Send the following text to AI:
Check whether AE CLI is available: confirm the current version, login status, and host; then list the first 5 projects I can access, returning only the project name and project ID. Only view for now. Do not modify anything, and do not go on to query data. Wait until I choose a project before taking the next step.
After you choose a project, send:
In the project I just chose, list the first 5 dashboards, returning only the name, ID, and update time. If there are duplicate names, keep all candidates and let me choose; do not guess, and do not modify any content.
Finally, choose a dashboard and read its data:
Read the dashboard I just chose. First confirm whether its reports still exist and have data, then pick a report with data and preview no more than 100 rows. Tell me the actual time range, whether the data was truncated, and the most worthwhile question to ask next. Read only; do not modify anything.
What it can do
After you install AE CLI, you can ask AI in natural language to view and use the data, reports, experiments, and tasks in AE. Here are some common uses.
Understand why business data changed
When new users, active users, retention, or payments fluctuate, AI can read existing reports and compare different time periods, channels, and user groups to help you find where the change mainly comes from. You don't need to copy data out of multiple reports one by one, and you can keep asking about a specific metric or how a specific type of user performed.
Review last week's game data for me. Look at notable changes in new users, retention, and payments, and focus on the channels with the biggest changes.
From requirements to tracking acceptance
Give your product requirements to AI, and it can work out the events and properties to collect and generate a tracking plan and sample code. After the plan goes live, it can also check the events in the project and view real-time data and collection errors, helping you troubleshoot issues such as missing reports, incorrect reports, or inconsistent property types.
Based on my official website project, design a tracking plan for registration conversion, content browsing, and form inquiries. Clearly state when each event is triggered and which event properties need to be recorded.
Quickly understand experiment results
Requires the Experiment module
AI can view experiment configurations, groups, and metric results, compare the performance of different Experiment Groups, and summarize significance, trends, and risks that need attention. When you review an experiment, you can have it turn the data directly into easy-to-read conclusions.
Show me how the Tidy X dual-platform interstitial experiment is performing so far. Compare the core and secondary metrics of each group, and tell me which results are already fairly clear.
Follow up on operation tasks and campaign results
Requires the Engage module
AI can find journeys, tasks, and campaigns, and view sends, deliveries, failure reasons, and conversion results. It can also put together target audiences or campaign reviews as you request. For daily follow-up, you don't have to search back and forth across multiple pages.
Review yesterday's win-back task for me. Look at sends and deliveries, the main failure reasons, and the subsequent activity and payment performance of the won-back users.
Troubleshoot data tasks and generate SQL
Requires the DataOps Platform
AI can view the running status of tables, data integration, Flows, and production instances, and read failure logs to help locate problems. For ad hoc data pulls or data processing needs, you can also describe the business definition directly and have it generate or check SQL.
Show me which data integration tasks failed today, analyze the failure reasons, and tell me how to handle them.
Summarize community and user feedback
Requires the Omni Insights feature
AI can search posts, comments, and live stream content in the community, and summarize trending topics, user sentiment, product suggestions, and risk signals. For weekly reports or version reviews, you can have it summarize feedback over a period of time directly, with the corresponding content sources attached.
Summarize the community discussions about the new version over the past week. Sum up the issues people care about most, the positive and negative feedback, and the risks that need priority attention.
Common capabilities at a glance
| What you want to do | What AI can do for you |
|---|---|
| View data | Find reports and dashboards, read data, compare time periods, break down by dimension, export results |
| Analyze experiments | View experiment info, sample size, metric results, and trends; prepare review materials |
| Manage metadata | Find events, properties, and metrics; check naming, types, relationships, and scope of impact |
| Handle tracking | Generate and maintain tracking plans, import and export, run checks, view collection errors and real-time data |
| Analyze operations | View journeys, tasks, campaigns, channels, audiences, and performance data |
| Manage data tasks | View tables, Flows, integration tasks, SQL run results, and production instances |
| Analyze community feedback | Search posts and comments, analyze topics and sentiment, view live stream performance |
| Use knowledge bases | Create and maintain knowledge bases, search pages, read source text, answer business questions |
| Manage Agents | Manage Agents, models, MCP, Skills, attachments, and automation tasks |
| Team collaboration | Create Agent Teams, start tasks, continue conversations, view results and artifacts |
| System Admin | View members, permissions, usage, quotas, channels, and system configuration |
The capabilities available may differ by company, project, and account. AI performs operations based on the current login environment and your permissions.
How to phrase requests so AI gets them right
An easy-to-execute task usually includes five kinds of information: the target environment or project, the actual objects, the time range, how far AI is allowed to go, and what to do when there are duplicate names or no data.
In "Project Name", find existing reports related to "New users by channel". If there are multiple candidates, first return their names and IDs for me to choose. After I choose, read the last 14 complete calendar days and compare them with the previous 14 days; only read and analyze, and do not modify the reports. Finally, give me three conclusions, each with supporting data, the actual time range, and whether the data is complete. If the object does not exist or has no data, stop and explain; do not switch to a similar object.
If you only say "Take a look at the data for me", AI still needs to confirm the project, objects, time, and permission boundaries. Making this information clear the first time is usually faster.
Useful phrases to add:
- "Read the current state first. Don't modify anything."
- "First return no more than 5 candidates, with names and IDs, and let me choose."
- "Prefer existing reports and saved metrics. Don't redefine the metric definitions."
- "Confirm that the object exists and has data before you start the analysis."
- "If there are duplicate names, nothing is found, or there is no data, stop and explain. Don't guess."
- "Preview at most 100 rows. If you need the complete results, use export instead of stitching pages together."
- "First give me the plan, the scope of impact, and the rollback method, and write only after I confirm."
- "After you finish, read the results again to confirm that the changes have taken effect."
- "When you give conclusions, include the data source, object IDs, time range, and a note on completeness."
FAQ
Do I need to learn the command line?
No. After installation and authorization, you can keep describing tasks in natural language. You may only need to copy a command or two when you troubleshoot installation or login issues.
Why are login and authorization needed?
AE CLI needs to know which AE environment you are using and which projects and resources you have access to. It doesn't automatically gain permissions beyond the scope of your current account.
Will AI modify or delete data directly?
It depends on your request and your account permissions. To avoid mistakes, we recommend stating clearly in the task "View first, don't modify" or "Give me a plan first, and execute after I confirm". High-risk operations such as deletion usually also require another confirmation.
Why can't I find a capability that others can use?
Check four layers in order: the AE CLI version, the version of the Skills that AI has loaded, whether the capability is deployed on the current host, and whether your account has permission. A successful CLI installation doesn't mean the Skills have been updated. After you install or update Skills, you usually need to restart the AI tool or reopen the task. If a command doesn't exist, first keep the output of ae-cli --version, the current host, and the complete error, and don't keep retrying with different parameters.
What if queries take a long time?
When the data volume is small, AI can read the results directly. Complete details or large-range queries may require an export task. You can have AI save the task information and continue downloading and analyzing when the task finishes, so you don't have to keep waiting on the current page.
How can I make analysis results more reliable?
Tell AI the project, time range, metric definition, and comparison method. Prefer saved reports and metrics, and ask for conclusions with supporting data. Experiments and important business decisions should still be reviewed by the owner.
Can AI run tasks on a schedule?
AE CLI supports Agent automation tasks, which you can set up in your Agent. We recommend completing the task manually once first to confirm that the project, permissions, output format, and exception handling all meet expectations, and then setting up scheduled runs.
What should I provide when I run into a problem?
Tell your customer success manager which AE environment you use, the project name, the task goal, when the problem occurred, and the error message returned by AI. Don't send passwords, access tokens, or other credentials.