Maia AI Agents features🔗
This guide describes the capabilities available through Maia AI Agents in Maia. It is intended for Matillion ETL users who want to understand what Maia AI Agents offers before upgrading to Maia.
Maia AI Agents are the agentic AI capability built into Maia, operating inside Designer, where you can use natural language prompts to build pipelines, explore data, troubleshoot errors, and manage your project files.
Pipeline development🔗
Maia AI Agents build, run, and manage both transformation and orchestration pipelines from the Maia AI Agents chat interface in Designer. For more information, read Building data pipelines with Maia AI Agents.
Transformation pipelines🔗
Describe a data transformation objective and Maia AI Agents create the pipeline for you. They add the appropriate components—such as Table Input, Calculator, Filter, and Aggregate—connect them, and configure each one based on your description.
Example prompt: "Get sales data from the orders table, calculate revenue by region, and store the results in a new table called regional_revenue."
You can iterate on the pipeline within the same session using follow-up prompts such as "Add a margin column" or "Filter for US customers only."
Orchestration pipelines🔗
Maia AI Agents build orchestration workflows from high-level descriptions. Supported pipeline elements include connectors, actions, DDL operations, control logic (conditions and loops), iterators, and scalar variables. Maia AI Agents prompt you for required details such as secrets or credentials as they build.
Example prompt: "Extract sales data from Salesforce, transform it using the clean_sales pipeline, then load it into Snowflake."
Multi-pipeline workflows🔗
Maia AI Agents understand cross-pipeline dependencies and can set up multiple pipelines to run in sequence. You can build and chain pipelines step by step from a single session, adding conditions, alerts, or post-processing steps as you go.
Running pipelines🔗
You can ask Maia AI Agents to run any orchestration or transformation pipeline directly from the chat interface. Maia AI Agents request explicit permission before executing, then monitor the run and report live status updates—including rows processed and any component errors.
Sampling data🔗
When building transformation pipelines, Maia AI Agents can sample existing tables in your data warehouse to detect column names, data types, and structure. This helps Maia AI Agents configure components more accurately. Only 10–20 rows are sampled, and sampled data is stored for 30 days. Maia AI Agents always ask for permission before sampling.
Committing and pushing changes🔗
Maia AI Agents can generate a commit message, commit your pipeline changes, and push them to your current branch in Designer. You must approve the action before they proceed.
Data exploration and visualization🔗
Maia AI Agents can query your data warehouse to explore tables, views, and schemas. They can also visualize sampled data directly in the chat interface as charts and graphs, using 10–20 sampled rows. These visualizations help you identify trends and patterns without running a full pipeline.
Project and file management🔗
Maia AI Agents can search through all files in your project to find specific content, configurations, or pipeline patterns. You can also ask Maia AI Agents to move, rename, copy, or delete files to maintain a clean project structure.
Pipeline analysis and optimization🔗
Maia AI Agents can analyze and explain existing pipelines—their components, data flow, and logic—without making any changes. You can also ask Maia AI Agents to identify configuration problems and suggest fixes, or recommend optimizations such as parallelization and component selection improvements.
Slash commands provide a standardized way to run common analysis tasks. Type / in the Maia AI Agents chat interface to see the available commands.
| Command | Task |
|---|---|
/document |
Generate a Markdown file with information about a specified pipeline, file, or directory. |
/explain |
Generate a brief explanation of a specified pipeline or concept. |
/fix |
Identify issues in a pipeline and present options to fix them. Maia AI Agents do not apply changes until you approve. |
/multi-table-load |
Load multiple tables from a data source using incremental or full load. |
/notes |
Add notes to a pipeline on the Designer canvas. |
/optimize |
Suggest performance improvements. Maia AI Agents do not apply changes until you approve. |
Custom connector creation🔗
Maia AI Agents automate the creation of custom connectors for any API that has publicly viewable documentation. This process requires no coding and replaces the manual API profile configuration in Matillion ETL.
You can provide Maia AI Agents with either:
- A link to the API documentation page
- A PDF, YAML, or JSON file (5 MB limit)
Maia AI Agents scan the documentation, identify available endpoints, configure authentication, apply pagination, and send test requests. You can review and edit the result in the Custom Connector workbench before saving.
For a step-by-step guide, read Create a custom connector with Maia AI Agents.
Context files🔗
Context files are Markdown (.md) files stored in your project at .matillion/maia/rules/. Maia AI Agents read all context files with every prompt and use them to apply your project's naming conventions, data standards, business rules, and connection information automatically—without requiring you to repeat these in every prompt.
Key constraints:
- There is a 12,000-character limit across all context files in the
.matillion/maia/rules/directory. - You can store additional, more detailed context files elsewhere in your project and reference them from your main context file. These do not count toward the character limit.
You can create and edit context files using the Add context or Edit context buttons in the Maia AI Agents chat interface, or via the Files panel in Designer.
For more information, read Customize Maia AI Agents with context files.
Skills🔗
Skills are reusable instruction sets stored at .matillion/maia/skills/[skill name]/SKILL.md in your project. Unlike context files—which Maia AI Agents always read—Maia AI Agents activate skills selectively based on how they interpret your prompt.
Skills can reference additional files such as template pipelines or SQL scripts stored in the same folder. You can create skills from the Skills drop-down in the Maia AI Agents chat interface.
Once a skill is committed and pushed to your remote repository, other users in the same project can merge it into their branch.
For more information, read Maia AI Agents skills.
Sessions and plan mode🔗
A session is a structured, ongoing interaction with Maia AI Agents that persists across your current branch. Maia AI Agents build on prior messages within the session, providing more coherent responses across multiple prompts. Sessions are private to each user and are saved in conversation history when you switch branches.
Plan mode🔗
Plan mode lets you review Maia AI Agents' proposed actions before they make any changes to your pipeline. Enable the Plan mode toggle at the bottom of the Maia AI Agents chat interface. Maia AI Agents will present a step-by-step plan for your review and approval before acting.
Maia AI Agents can also suggest switching to plan mode automatically if they determine your prompt is complex enough to require planning.
For more information, read Sessions and tools in Maia AI Agents.
Troubleshooting and monitoring🔗
Root cause analysis🔗
When a pipeline fails, Maia AI Agents analyze the error and provide a plain-language summary of what went wrong, identify the exact location of the failure, and suggest fixes. For scheduled or API-triggered pipeline failures, a Fix with Maia button appears in the Troubleshoot with Maia banner on the Pipeline run details page. Clicking it starts a guided recovery process on a new branch.
Anomaly detection🔗
Maia AI Agents monitor your scheduled pipeline runs and flag significant deviations from expected execution times using statistical modeling. Anomalies appear as red arrows next to duration metrics in the Pipeline Runs dashboard, with tooltips describing the deviation.
For more information, read Troubleshoot pipelines with Maia AI Agents.