With Alkemi, you can query data, create visualizations, generate detailed reports, and build MCP tools that enable data querying from other chat clients.
:bolt:
Quickstart
Get an overview of the interface and try out your first query!
# Overview
With Alkemi, you can query data, create visualizations, generate detailed reports, and build MCP tools that enable data querying from other chat clients. Watch a demo of Alkemi below.
{% @arcade/embed flowId="za5rO3O7lUCdIsB3fxHu" url="" %}
# Quickstart
Get an overview of the DataLab interface and ask your first questions to the agent.
{% embed url="" %}
This guide outlines the steps to upload a CSV file and analyze data using the Alkemi DataLab interface. You can download the CSV we use in this guide from data.gov [here](https://ers.usda.gov/sites/default/files/_laserfiche/DataFiles/50673/CPIHistoricalForecast.csv?v=85900) or utilize any of the data products that are offered for free with your account.
### Key Steps
The steps below outline the quickstart video above with links to where each step is discussed in the video.
{% stepper %}
{% step %}
## **Accessing Alkemi DataLab** [**0:00**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=0)
* Log In to [Alkemi DataLab.](https://datalab.alkemi.ai/?request_access=true)
* Familiarize yourself with the layout:
* Left sidebar: Chat history and API endpoints.
* Main chat input area for queries and commands.
* Right sidebar: Credit balance, integrations, and current data slices.
{% endstep %}
{% step %}
## **Preparing to Upload a CSV File** [**1:11**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=71)
* Ensure you have a CSV file ready for upload (e.g., consumer price index historical data).
{% endstep %}
{% step %}
## **Uploading the CSV File** [**1:22**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=82)
* Use the upload functionality in Alkemi DataLab to upload your CSV file.
* Review the columns in your CSV file, noting important attributes.
{% endstep %}
{% step %}
## **Creating a New Table from the CSV Data** [**2:11**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=131)
* Construct a query to filter and group data:
* Filter by the attribute column (e.g., midpoint of prediction interval).
* Group by consumer price index item.
* Calculate a new column for forecast percent change.
{% endstep %}
{% step %}
## **Querying Results** [**2:50**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=170)
* Send the query to the Alkemi agent.
* Wait for the results to be assembled and saved as a data asset.
{% endstep %}
{% step %}
## **Accessing and Reviewing the Results** [**3:06**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=186)
* Use the mentioning functionality to reference the new table in chat threads.
* Click to expand and view the data, or download it as a CSV.
{% endstep %}
{% step %}
## **Cleaning and Sorting the Data** [**4:00**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=240)
* Remove any irrelevant categories (e.g., 'all food').
* Sort the results by total increase in descending order.
{% endstep %}
{% step %}
## **Visualizing the Data** [**4:37**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=277)
* Request a visualization of the data (e.g., as a bar chart or pie chart).
* Interact with the chart to view details and download it for presentations.
{% endstep %}
{% step %}
## **Next Steps** [**5:16**](https://loom.com/share/4d49ebbc73364a2bb2ffb7172e367f9a?t=316)
* Consider connecting integrations like Snowflake for enhanced data analysis.
{% endstep %}
{% endstepper %}
# Connect Alkemi to Claude
Securely querying Alkemi data from Claude is easy thanks to Alkemi's OAuth MCP features.
Use our remote MCP server URL when adding the Alkemi connector to Claude `https://api.alkemi.ai/mcp`
For Clients that support OAuth, simply input this URL in the connection details and authorize the connection through the chat client.
If you only want to connect to your Data Product, Alkemi automatically includes a unique remote MCP Server for each Data Product. This allows AI agents to securely query your data product when you authorize them to do so. You can find a Data Product's MCP Server URL in the Data Product's info.
{% hint style="info" %}
The first time you use an Alkemi tool in Claude you will need to Allow Claude to use it
{% endhint %}
Before Claude uses an Alkemi tool it will ask you to allow it
## How to Add Alkemi to Claude:
{% stepper %}
{% step %}
### Add a new custom connector in Claude
In your Claude account, click **Customize** in the side nav. Click **Connectors** then **+** and select **Add Custom Connector**
{% endstep %}
{% step %}
### Add the Alkemi MCP Server
Input Alkemi into the Name field and input the following URL into the Remote MCP Server URL field `https://api.alkemi.ai/mcp` . Click **Add**.
{% endstep %}
{% step %}
### Connect
Click **Connect** to securely connect to the Alkemi MCP server via OAuth.
{% endstep %}
{% step %}
### Begin using Alkemi in Claude
Create a new chat and try a few queries or skills. You can use the `load_skill` tool to call any skill avaiable in Alkemi. You can also ask Claude which tools you have available. Here's a few example prompts to try in Claude:
> Load the the amazon best seller snapshot skill from Alkemi and run it on the gardening tools category
> Whats the price history on ASIN B0BZYF1C1V?
> {% endstep %}
> {% endstepper %}
## Get a full overview of using Alkemi in Claude:
{% @arcade/embed flowId="FFMsQrr6S2avkL5jVe6w" url="" %}
# Connect Alkemi to ChatGPT
Securely querying Alkemi data from ChatGPT is easy thanks to Alkemi's OAuth MCP features.
Use our remote MCP server URL when adding the Alkemi connector to ChatGPT `https://api.alkemi.ai/mcp`
For Clients that support OAuth, simply input this URL in the connection details and authorize the connection through the chat client.
If you only want to connect to your Data Product, Alkemi automatically includes a unique remote MCP Server for each Data Product. This allows AI agents to securely query your data product when you authorize them to do so. You can find a Data Product's MCP Server URL in the Data Product's info.
{% @arcade/embed flowId="9mdaCEe5Y6wGzoR9aGUY" url="" %}
Use the following URL when adding the Alkemi connector to ChatGPT `https://api.alkemi.ai/mcp`
# Searching and Scraping the Web
Alkemi DataLab comes with tools that enable you to easily search the web and scrape content.
{% embed url="" %}
# Searching the Web
Alkemi's web search tool enables you to search for terms and pull in the results to DataLab
{% stepper %}
{% step %}
### Select the Web Search tool from the tool menu
{% endstep %}
{% step %}
### Enter your query
The web search tool takes two parameter, your search term and an optional number of results to pull.
You can click the web\_search tool to view more info about the tool and accepted parameters.
{% endstep %}
{% step %}
### Submit your query!
Once your search is completed your results will be saved as data asset for future use.
{% endstep %}
{% endstepper %}
# Scraping a Website
Alkemi's web scrape tool enables you to scrape content from a URL pull in the results to DataLab
{% stepper %}
{% step %}
### Select the web\_scrape tool
Type **"@web\_scrape"** into the chat input to select the web\_scrape tool.
{% endstep %}
{% step %}
### Paste your URL
Paste the URL you want to scrape into the chat input
{% endstep %}
{% step %}
### Submit your message
Submit your message and once complete, your scraped content will be available in a .txt data asset file.
{% endstep %}
{% endstepper %}
# Researching a Company
Alkemi's company research tool enables you to research information and news about a company and pull in the results to DataLab
{% stepper %}
{% step %}
### Select the company\_research tool
{% endstep %}
{% step %}
### Type in the name of the company
Optionally, you can add the number of results you want as well.
{% endstep %}
{% step %}
### Submit your message
Submit your message and once complete, your company info will be available in a .txt data asset file.
{% endstep %}
{% step %}
### (Optional) Follow-up and ask the agent for a report on the company
{% endstep %}
{% endstepper %}
# Connect your data
Cover image
:file-circle-plus:
Uploading Files
Work with CSVs, spreadsheets, parquet, croissant files, and more within Alkemi DataLab
# Uploading Files
{% stepper %}
{% step %}
## **Click the add button in the chat input**
Click the add button and select **Upload from computer**
{% endstep %}
{% step %}
## **Let the file upload**
After a few seconds an alert will notify you that your file is ready to use.
The file will be auto-inserted into your chat input.
Click the mentioned asset or file preview link beneath the the input to view the file contents.
{% endstep %}
{% step %}
### **Use the uploaded asset!**
Start chatting with the agent to extract data from your CSV, transform it into a new dataset, or to visualize it. Type “**@”** and begin typing the file name to mention your asset and bring in your data into any chat thread.
{% endstep %}
{% endstepper %}
# Transforming a CSV
Transforming an uploaded CSV into a new data asset within DataLab is easy. Simply ask the AI agent to transform the data and you can create a completely new data asset tailored to your needs.
{% stepper %}
{% step %}
## **Mention your uploaded CSV**
Mention your CSV with a request to transform the data into a new asset and Alkemi’s agent will generate a new, transformed data asset for you.
{% endstep %}
{% step %}
## **Use the transformed asset**
Any asset created by the Alkemi agent be utilized by mentioning it. Your new asset is given a title that you can mention to do a number of tasks with this data, even to transform it again into a new asset.
However, you don’t always have to mention the asset as long as your are referencing it from the thread in which it was generated.
{% endstep %}
{% endstepper %}
# Exporting to CSV
Exporting to CSV in DataLab is simple.
{% stepper %}
{% step %}
## **Click the download button**
From any table table within your chat thread you will have a button to download the asset. Click this and select a location on your computer to download the asset as a CSV.
{% endstep %}
{% endstepper %}
# Connecting Databricks
Connect your Databricks workspace to the Alkemi platform. Once connected, you can create and manage Data Products within Alkemi using your Databricks warehouse as the data source.
{% stepper %}
{% step %}
## Click +Add under integrations
Click the add button and select Databricks
{% endstep %}
{% step %}
## Input a Connection Name and your Databricks Workspace URL and HTTP Path
Input a name for your connection. Then In another tab, go to your account in Databricks and to obtain your Workspace URL and HTTP Path.
To find these, in Databricks navigate to **SQL Warehouses**, select the warehouse you want to connect to DataLab, and click on the **Connection Details** tab.
Copy Server hostname to your clipboard and paste into the Workspace URL field in DataLab. Then copy your HTTP Path for the warehouse and past into the HTTP Path field in DataLab.
{% endstep %}
{% step %}
## Create a Service Principal User
In another tab, go to your account in Databricks and create a Service Principal User. To do this, click on your user name in the upper right and select Settings.
Select **Identity and Access** and then click Manage under **Service principals.** Select **Add service principal** and click **Add new** in the modal. Assign your service principal a name and click **Add**.
You now have a new service principal user. Click into the new service principal to access connection details and to create a secret.
Click on the **Secrets** tab and then **Generate secret**. Assign a lifetime value for the secret and click **Generate**.
Copy the secret to your clipboard, navigate back over to DataLab and paste the value into the Client Secret field in the Databricks integration modal. Now do the same for the Client ID.
{% endstep %}
{% step %}
## Grant Permissions to your Service Principal User
For this next step you can connect to Databricks using your preferred SQL client or use the SQL Editor within databricks.
Run the following SQL commands to grant your service principal user the necessary permissions. Replace `` , `` , and `` with your actual values:
```sql
GRANT USE CATALOG ON CATALOG TO ``;
GRANT USE SCHEMA ON SCHEMA . TO ``;
GRANT SELECT ON SCHEMA . TO ``;
GRANT CREATE TABLE ON SCHEMA . TO ``;
```
{% endstep %}
{% step %}
## Test and Save your connection
That’s it. Now you can create Data Products within Alkemi using your connect Databricks warehouse!
{% endstep %}
{% endstepper %}
# Connecting Snowflake
Connect your Snowflake workspace to the Alkemi platform. Once connected, you can create and manage Data Products within Alkemi using your Snowflake database as the data source.
{% stepper %}
{% step %}
## Click +Add under integrations
Click the **Add** button and select Snowflake.
{% endstep %}
{% step %}
## Input a Connection Name and your Snowflake Account ID
Input a name for your connection. Then In another tab, go to your account in Snowflake to obtain your Account ID.
To find your account ID, in Snowflake click your username in the lower left and navigate to **Account > View Account Details**.
Copy the **Account identifier** to your clipboard and paste into the **Account ID** field in the integration form in DataLab.
{% endstep %}
{% step %}
### Create a new user for Alkemi in Snowflake (recommended)
For added security, it's recommended that you create a service user in Snowflake for your DataLab connection. In the form we provide a guide with SQL that you can run in a Snowflake workspace to create the user and roles. Expand the guide to view the generated SQL.
In Snowflake, navigate to **Projects > Workspaces** and add a new SQL file.
Copy and paste the SQL from the guide into this new file. This includes a unique public key that DataLab has generated for this connection. Replace database\_name and schema\_name in the SQL with your database and schema details. Now click the **Run** button to create the user.
{% endstep %}
{% step %}
### Enter your Snowflake Username, Role, and Warehouse
Make sure these match what you created in Step #3.
{% endstep %}
{% step %}
### Test and Save your connection
Alkemi will let you know if it encounters any connection issues.
{% endstep %}
{% step %}
### Start creating data products!
Begin creating data products in DataLab for text-to-SQL on your Snowflake data, and enable MCP connections to your data in chat clients such as Claude, Cursor, and ChatGPT. For more details, see "Creating a Data Product."
{% endstep %}
{% endstepper %}
# Connecting BigQuery
Learn how to connect your Google BigQuery Dataset to the Alkemi platform. Once connected, you can create and manage Data Products within Alkemi using your Google BigQuery Dataset as the data source.
{% stepper %}
{% step %}
### Click +Add under integrations
Click the **Add** button and select BigQuery.
{% endstep %}
{% step %}
### Create a Google Cloud Service Account
Go to the IAM & Admin section of Google Cloud
And go to the Service Users sub-section
Create a new Service Account
Add a name that will tell you later what it is and click "Done"
Add Permissions
And create and download a JSON key
{% endstep %}
{% step %}
### Gather BigQuery DataSet Information
Go to the BigQuery Console
Select your DataSet and copy the ID of it
The ID is in the format of \[PROJECT\_ID].\[DATASET\_ID]
{% endstep %}
{% step %}
### Add Information to Alkemi
Use the first part of the DataSet ID, separated by a dot, as the Project ID. The second part is the Dataset ID. And copy in the contents of the JSON file for the secret key.
{% endstep %}
{% step %}
### Test and Save your connection
Click Test Connection and resolve any issues before clicking Save Connection
{% endstep %}
{% step %}
### Start creating data products!
Explore the creation of data products in DataLab for generating text-to-SQL queries on your BigQuery datasets. Enable MCP connections to integrate your data with chat clients like Claude, Cursor, and ChatGPT. For further information, refer to "Creating a Data Product.
{% endstep %}
{% endstepper %}
[ ](https://open-2v.gitbook.com/url/preview/site_ybgqm/documentation/~/revisions/qNEJ8ULA5Br3aoLGqIDl/getting-started/connect-your-data/connecting-databricks)
# Connecting Linear
See how to securely connect your Linear workspace to Alkemi and get insights into your issues and product status
{% @arcade/embed flowId="mT0SAxJPCBd4gMmlaf6V" url="" %}
#### 1. Open the Settings menu.
Start by opening the Settings menu to configure your workspace
#### 2. Navigate to Integrations
Navigate to Integrations to explore available app connections.
#### 3. Select the Linear integration
Select the Linear integration to connect your issue tracking system.
#### 4. Click Connect Linear
Click Connect Linear to start linking your Linear workspace with DataLab.
#### 5. Authorize Alkemi to securely access your Linear workspace
Authorize Alkemi to securely access your Linear workspace for seamless data integration.
#### 6. Start a new Chat test the connection
Start a new Chat to interact with your connected Linear workspace and generate reports.
#### 7. Run a report on completed issues
Now let's run a detailed launch report, looking at the completed issues
#### 8. Click the expand icon to view your Weekly Engineering Report
Click the expand icon to view your Weekly Engineering Report in greater detail.
# Querying your data
To query your data, simply upload and mention a file with your questions, and the Alkemi agent will execute your queries. For complex queries, such as joining data across multiple files or connecting to a data source like Databricks, Snowflake, or BigQuery, create a Data Product. Refer to the "Creating a Data Product" section to get started.
{% content-ref url="/pages/5X1HQxwsFDKQK7GySXTf" %}
[Creating a Data Product](/documentation/basics/data-products/creating-a-data-product)
{% endcontent-ref %}
# Data Products
Data Products provide a way for you to create curated data assets from your connections with automatic AI functionality like text-to-SQL and MCP servers.
## What is a Data Product?
An Alkemi Data Product is a comprehensive package that allows seamless querying and utilization of your data within AI applications. Connect your data source—such as Snowflake, BigQuery, or Databricks—and create multiple Data Products from these connections.
**Easy Creation**
To create a Data Product, select the tables you want to include. Once your product is created, Alkemi's agent will analyze your data to ensure efficient querying by generating a dedicated MCP server equipped with text-to-SQL functionality.
**Enhanced Performance**
Enhance your Data Product's performance using our automatic text-to-SQL training features and by providing specific instructions to the Alkemi agent.
**Privacy and Publicity Options**
By default, all Data Products are kept internal to your organization. If you choose to make your Data Product public, additional functionalities for listing, pricing, and controlling access become available.
# Creating a Data Product
{% @arcade/embed flowId="rVam69xh9gImvdEBnTa2" url="" %}
{% stepper %}
{% step %}
### Select "Data Products" from your account menu
Go to the Data Products management page by choosing **Data Products** from the drop-down menu in the top-right corner of the DataLab interface.
{% endstep %}
{% step %}
### Begin creating your Data Product
1. Click the **+Data Product** button.
2. Enter your data product details. By default, your data product is internal and only visible within your organization. To make it public, adjust the privacy setting during creation or you can edit it later.
3. Select a Data Provider for your data product. Commonly, this is your organization's name. To add a new provider, go to the Data Providers section.
4. Name your data product and provide a description.
{% endstep %}
{% step %}
## Selecting a Data Source and Tables
1. **Connect Your Data Source**: Follow our guides in [Connect your data](/documentation/getting-started/connect-your-data) if you need help connecting a data source.
2. **Select a Data Source**: Use the dropdown menu to choose your data source. This action will display a list of available tables.
3. **Add Tables**: Click the **+ button** beside each table you want to include in your data product.
**Note**: You're limited to 5 tables per Data Product. For more extensive data sets, create multiple Data Products to ensure better performance and efficient querying by the Alkemi agent.
**Need Extra Tables?** Contact our team at if you require additional tables.
{% endstep %}
{% step %}
### Select Create Product
Once your new Data Product appears on the list, you can begin training it. This ensures Alkemi's agent efficiently answers complex questions by translating them into precise queries.
{% endstep %}
{% endstepper %}
# Training your Data Product
This guide covers how to create, manage, and optimize your training data through the Text to SQL Training interface.
Our AI Agent serves as your personal data scientist, proficient in interpreting natural language queries and converting them into SQL to analyze your connected data sources. The effectiveness of the Agent's responses relies heavily on high-quality examples of user prompts paired with their corresponding SQL queries.
## Creating Training Data
The Text to SQL Training feature consists of two primary approaches:
{% content-ref url="/pages/Y1ieVIb12fr6u0jbJAo5" %}
[Creating Manual Training Data](/documentation/basics/data-products/training-your-data-product/creating-manual-training-data)
{% endcontent-ref %}
{% content-ref url="/pages/MpqFlKhq9sQAWI8RrjOn" %}
[Creating Automated Training Data](/documentation/basics/data-products/training-your-data-product/creating-automated-training-data)
{% endcontent-ref %}
Both methods enhance the training dataset, improving the Alkemi Agent's capacity to understand and respond to questions about your specific data.
## Reviewing and Managing Training Data
Quality control is crucial for ensuring an efficient Agent. The system supports various review workflows tailored to the data source and confidence levels.
{% content-ref url="/pages/EN3vQKxIfwQ1YtKfqPrN" %}
[Reviewing Pending Queries](/documentation/basics/data-products/training-your-data-product/reviewing-pending-queries)
{% endcontent-ref %}
{% content-ref url="/pages/rnTjB84c8FV4C5aPnU8d" %}
[Review Questionable Training Examples](/documentation/basics/data-products/training-your-data-product/review-questionable-training-examples)
{% endcontent-ref %}
## Helpful Tips
{% hint style="success" %}
## Our Recommended Approach
1. Begin with manually crafting essential queries.
2. Transition to automated generation with manual oversight.
3. Regularly assess queries that have low certainty.
4. Continuously monitor agent performance and refine training data.
5. Gradually increase automation as quality enhances.
{% endhint %}
{% hint style="info" %}
Instruct: Quality training data enhances Agent learning. Investing time in this process leads to better performance and increased user satisfaction.
{% endhint %}
{% content-ref url="/pages/bc7U92eG9xsKgXHw57Ey" %}
[Best Practices](/documentation/basics/data-products/training-your-data-product/best-practices)
{% endcontent-ref %}
{% content-ref url="/pages/V5PqoC2SURUFeNAXjQUZ" %}
[Troubleshooting](/documentation/basics/data-products/training-your-data-product/troubleshooting)
{% endcontent-ref %}
# Creating Manual Training Data
Creating training examples by hand with direct control.
In the Manual tab, you maintain full control over training examples to ensure precise handling of essential queries. You have two options for creating these examples manually:
{% hint style="success" %}
## Best Practices for Manual Training Data
* **Be specific**: Include examples that reflect how your users actually ask questions
* **Cover variations**: Create multiple ways of asking for the same information
* **Include context**: Use column names and business terms specific to your data
* **Test queries**: Ensure all SQL queries are valid and return expected results
{% endhint %}
## Option 1: CSV Upload
Upload a CSV file with the following structure:
* **text** column: Contains the natural language prompt.
* **query** column: Contains the corresponding SQL query.
**Example CSV format:**
```
text,query
"Show me total sales by month","SELECT DATE_TRUNC('month', sale_date) as month, SUM(amount) as total_sales FROM sales GROUP BY month ORDER BY month"
"Which products had the highest revenue last quarter","SELECT product_name, SUM(revenue) as total_revenue FROM products WHERE sale_date >= DATE_TRUNC('quarter', CURRENT_DATE - INTERVAL '3 months') GROUP BY product_name ORDER BY total_revenue DESC LIMIT 10"
```
{% embed url="" %}
## Option 2: Create New Query
Use the "Create New Query" modal to add training examples one at a time:
{% stepper %}
{% step %}
Click the **"Create New Query"** button
{% endstep %}
{% step %}
Enter a natural language prompt in the text field
{% endstep %}
{% step %}
Write the corresponding SQL query
{% endstep %}
{% step %}
Run the query to test it (optional, but recommended)
{% endstep %}
{% step %}
Save the training pair
{% endstep %}
{% endstepper %}
{% embed url="" %}
# Creating Automated Training Data
Create synthetic training examples automatically using the Alkemi Agent
The Automated tab leverages AI to generate synthetic training examples, accelerating the training process while maintaining quality through review mechanisms.
## Ways to generate training data
### 1. Quick Generation (Synchronous)
Instantly creates up to 10 prompt/query pairs. Navigate to the Text to SQL Training tab in your Data Product and Click the **"Generate"** button.
{% hint style="success" %}
#### **What synchronous generation is best for**
Quick testing or when you need a few examples immediately.
{% endhint %}
{% hint style="warning" %}
#### **Trade-offs**
Uses faster but lower-quality models due to speed requirements
{% endhint %}
{% embed url="" %}
### 2. Configured Generation (Asynchronous) - Recommended
Configures ongoing generation of synthetic queries.
* **Quantity**: Specify how many synthetic queries to generate in total
* This number includes existing rows. If you have 5 rows already and set the
* **Auto-approval**: Choose whether queries should be:
* Automatically approved and used immediately
* Held in "pending" status for manual review
#### **Configuration Options**
When setting up automated generation, consider:
* **Volume**: Start with smaller batches (between 5 and 20) to assess quality
* **Review requirements**: Initially, either enable manual review or set a high minimum certainty (between 80% and 95%) to ensure the quality of generated examples
* **Iteration**: Adjust configuration based on the quality of generated examples
{% hint style="success" %}
#### **Why asynchronous generation is preferred**
* Uses higher-quality AI models
* Includes additional validation steps
* Produces more accurate and relevant training examples
* Runs in the background without interrupting your workflow
{% endhint %}
{% embed url="" %}
# Reviewing Pending Queries
How to review and approve synthetic training examples
Depending on your configuration (see [Creating Automated Training Data](/documentation/basics/data-products/training-your-data-product/creating-automated-training-data)), automated training data may be placed in "pending" status, requiring manual review in order for the Alkemi Agent to start using it. If there are any queries in “pending” status, you'll see a "Review Pending Queries" button.
{% embed url="" %}
### Individual Review
1. Click "Review Pending Queries" to open the review modal
2. For each query, you can:
* **Accept**: Approve the prompt/query pair as-is
* **Edit**: Modify the prompt or query before saving
* **Delete**: Remove the training example entirely
### Batch Review
1. Select queries using checkboxes in the table
2. Choose to either:
* **Accept**: Approve selected queries for training
* **Delete**: Remove unsuitable queries
# Review Questionable Training Examples
How to review questionable training examples
The system automatically analyzes all training queries and assigns a certainty rating between 0% and 100%. Queries with certainty below 70% should generally be reviewed.
### Understanding Certainty Ratings
* **70-100%**: High confidence, likely correct
* **Below 70%**: Low confidence, manual review recommended
* The "Review Low Certainty" button appears when low-certainty queries exist
{% embed url="" %}
### Low Certainty Review Process
1. Click "Review Low Certainty" to open the review modal
2. For each low-certainty query, you'll see:
* The original prompt and query
* Certainty rating and analysis explaining the rating
* Available actions
#### Review Actions
* **Delete**: Remove the problematic query
* **Edit**: Manually correct the prompt or query
* **Use AI Fix**:
* If available: Click "View Fix" to see suggested corrections
* If not available: Click "Generate Fix" to request AI assistance
#### AI Fix Review
When viewing an AI-suggested fix, you'll see:
* Original prompt
* Original query with identified issues
* Suggested corrected query
* Explanation of improvements
* Option to accept or reject the fix
{% embed url="" %}
# Best Practices
Make it easy to succeed
### Building Effective Training Data
1. **Start with common queries**: Focus on the questions users ask most frequently
2. **Include edge cases**: Add examples for complex or unusual queries
3. **Maintain diversity**: Cover different aspects of your data schema
4. **Regular updates**: Add new examples as user needs evolve
### Quality Over Quantity
* 20 high-quality examples are more valuable than 200 poor ones
* Focus on accuracy and relevance to your specific use cases
* Regularly review and refine existing training data
### Monitoring and Improvement
1. **Track Agent performance**: Note when the Agent struggles with certain types of queries
2. **Add missing examples**: Create training data for queries the Agent couldn't handle
3. **Review certainty ratings**: Regularly check and fix low-certainty queries
4. **Iterate**: Training is an ongoing process—continuously improve your dataset
### Common Pitfalls to Avoid
* **Ambiguous prompts**: Ensure prompts clearly indicate the desired outcome
* **Outdated examples**: Remove training data that references deprecated tables or columns
* **Duplicate concepts**: Avoid too many similar examples that don't add value
# Troubleshooting
Common issues and solutions
### Agent not understanding queries?
* Check if you have training examples similar to the problematic query
* Review your training data for accuracy
* Add more diverse examples covering the missing use case
### Low certainty ratings on many queries?
* Review the analysis provided for each low-certainty query
* Look for patterns in what's causing low confidence
* Consider adjusting your prompt writing style for clarity
### Automated generation producing poor results?
* Switch to asynchronous generation for better quality
* Review and edit generated examples
* Provide more manual examples to guide the Alkemi Agent’s automatic generation
# Organizations
Creating an organization in Alkemi allows you to collaborate seamlessly with your co-workers on data projects and reports. By setting up an organization, you can invite team members, assign roles, and securely collaborate with data within a centralized workspace.
## Inviting Members to Your Organization
{% stepper %}
{% step %}
### Navigate to your organization invitations page
Go to to invite your team members
{% hint style="info" %}
You can also get here by click on the Members item in settings menu from DataLab then selecting the Invitations tab.
{% endhint %}
{% endstep %}
{% step %}
### Click the “+Invitation” button
{% endstep %}
{% step %}
### Fill in your team members email address and select their role
{% endstep %}
{% step %}
### Click “Send Invitation”
{% endstep %}
{% endstepper %}
# Credits
This page provides detailed information on credit usage for Alkemi's data and models.
## How Credits Work
DataLab charges credits for any message sent in a chat and additional credits for paid data or tool use - like using a data product that's not owned by another organization or doing a web search.
If your question requires two paid data products or tools, Alkemi charges credits for the initial message and for querying the two paid data products.
Here's a break down of an example prompt to better understand how credits are charged:
> Search the web for technology news and identify businesses, their website domains and stock tickers for the businesses found **(1 credit)**, then for each of those stock tickers, run a query against the @(stock data) product for the performance YTD **(6 credits)** and then run another query against the @(clickstream) product for who the top 5 referral domains are to their websites **(6 credits)**
## How Credits are Priced
Each new account comes with free credits. Additional credits can be purchased à la carte and the cost of each credit is determined by your plan.
# Model Packages
This page provides detailed information on the models currently in use in Alkemi's model packages
## Why Does Alkemi Use Model Packages?
In Alkemi, any question typically undergoes several steps, with different LLMs excelling at various tasks. Therefore, Alkemi employs a dynamic approach, assigning each task to the most suitable LLM. For example, a single request might require the Alkemi agent to: create a plan, continuously reassess it, generate SQL, query data, analyze responses for accuracy, suggest alternatives, transform data, and offer next-step recommendations. This strategy ensures optimal handling of diverse tasks.
We continuously evaluate leading LLMs for each task, selecting the most suitable one based on a mix of speed, cost, and accuracy, tailored to your package choice. Each LLM is tested for accuracy, bias, relevancy, and answer correctness using industry defined best practices and standards for evaluating LLMs.
## What LLMs Are Available?
The table below outlines a list of models that are currently in use in each package.
{% hint style="warning" %}
This list is likely change frequently as new LLMs are benchmarked and made activated in Alkemi often.
{% endhint %}
Provider
Model
Available Packages
Google
gemini-2.0-flash
Bronze
Google
gemini-2.0-flash-lite
Bronze
OpenAI
gpt-5-mini
Bronze
OpenAI
o3-mini
Bronze, Silver
OpenAI
gpt-4.1-mini
Silver
OpenAI
o3
Gold
Anthropic
claude-sonnet-4
Silver
Anthropic
claude-opus-4
Gold
Depending on your activated package, different criteria are prioritized. The Bronze package focuses on speed and cost, maintaining accuracy as a key priority. The Silver and Bronze packages allow the use of more advanced LLMs, which may be slower and more costly but offer higher quality responses. Silver and Gold packages are only available on Team plans.
You can use different model packages by selecting the package dropdown in the right hand side bar
## Which LLMs Does Alkemi Evaluate for Use?
Not all language models (LLMs) are currently active. Models are only activated if they meet our benchmark criteria. Below is the list of LLMs evaluated for potential use.
| Provider | Model |
| ----------- | -------------------------- |
| Google | gemini-2.0-flash |
| Google | gemini-2.0-flash-lite |
| Google | gemini-2.5-flash |
| Google | gemini-2.5-flash-lite |
| Google | gemini-2.5-pro |
| OpenAI | gpt-4o |
| OpenAI | gpt-4o-mini |
| OpenAI | o3 |
| OpenAI | o3-mini |
| OpenAI | o3-mini-high |
| OpenAI | gpt-4.1 |
| OpenAI | gpt-4.1-mini |
| OpenAI | gpt-4.1-nano |
| OpenAI | o4-mini |
| OpenAI | o4-mini-high |
| OpenAI | gpt-5 |
| OpenAI | gpt-5-mini |
| OpenAI | gpt-5-nano |
| Anthropic | claude-3.7-sonnet |
| Anthropic | claude-3.7-sonnet:thinking |
| Anthropic | claude-sonnet-4 |
| Anthropic | claude-opus-4 |
| DeepSeek | deepseek-r1 |
| DeepSeek | deepseek-chat-v3-0324 |
| Moonshot AI | kimi-k2 |
# Alkemi Changelog
New features, updates, and improvements to Alkemi.
## August 2025 - 🎉 DataLab is Now Available
We're thrilled to kick off our changelog with some massive news: **DataLab is now officially open for signups!**
After months of development and testing, we're excited to put our AI-powered data analysis platform in your hands. DataLab transforms how you interact with your data - no more wrestling with complex queries or waiting days for insights.
[**Sign up for free at datalab.alkemi.ai**](https://datalab.alkemi.ai/?request_access=true)
Every new user gets **200 free credits** to explore everything DataLab has to offer. No credit card required.
{% columns %}
{% column %}
{% endcolumn %}
{% column %}
#### **Data Integration Made Simple**
* **File uploads**: Easily import and transform your spreadsheets, CSV files, and text documents
* **Real-time connections**: Connect directly to BigQuery, Databricks, and Snowflake
* **Zero-code MCP servers**: Launch your server and build custom MCP tools without writing a single line of code
[Read the documentation](https://docs.alkemi.ai/alkemi/documentation/getting-started/connect-your-data)
#### AI-Powered Analysis
* **High-performance text-to-SQL**: Ask questions in plain English, get precise SQL queries instantly
* **Custom data products**: Train your own text-to-SQL models on your specific data and business logic
* **Smart visualizations**: Automatically generate charts and graphs that actually make sense
[Read the documentation](https://docs.alkemi.ai/alkemi/documentation/getting-started/querying-your-data)
#### Intelligent Reporting
* **Automated report writing**: Let our AI agents craft comprehensive reports from your data
* **Transparent reasoning**: See exactly how our agents discovered insights and reached conclusions
* **Web search integration**: Enrich your analysis with real-time web data and scraping capabilities
[Read the documentation](https://docs.alkemi.ai/)
### 🎯 What This Means for You
Whether you're a data analyst tired of writing the same queries over and over, a business leader who wants faster insights, or a developer building data products, DataLab meets you where you are. Ask questions in natural language, get answers in seconds, and understand the reasoning behind every insight.
### 🔮 What's Next
This is just the beginning. We're already working on exciting new features and integrations that will make DataLab even more powerful. In the coming weeks we'll be rolling out our automated training functionality and a dedicated section and new agent capabilities for generating presentation-ready reports. Stay tuned to this changelog for regular updates on new capabilities, improvements, and integrations.
{% endcolumn %}
{% endcolumns %}