Contextual provides a platform for creating enterprise-grade AI agents, grounded in your documents and data. See a demo of our agents here.

Our APIs provide powerful—yet simple—interfaces for ingesting data, creating agents, and interacting with our state-of-the-art Contextual RAG 2.0 system.

Follow this guide to create your first agent! See https://github.com/ContextualAI/examples for easy-to-follow Jupyter notebooks for using our APIs.

Get your API Key

If you do not have access to the platform, you can request an API key via the Request Access button in the header of the documentation page or on the Contextual website.

Contextual uses API keys to authenticate requests. Only admins within a tenant can create API keys. To create a key:

Log into your tenant at app.contextual.ai
Click on API Keys in the sidebar
Click the Create API Key button in the upper right and follow the instructions
Save the generated key in a secure location

Access the API via SDK/curl commands

Step 1: Create a datastore

A datastore contains all of the files that your agent(s) will be able to access. Each agent must be associated with at least one datastore. You can create a datastore using the /datastores endpoint with the following command:

from contextual import Contextual

# Initialize the client with your API key
contextual = Contextual(api_key="API_KEY")

# Create a datastore
datastore = contextual.datastores.create(name="Test Datastore")

curl --request POST \
     --url https://api.contextual.ai/v1/datastores \
     --header 'accept: application/json' \
     --header 'authorization: Bearer $API_KEY' \
     --header 'content-type: application/json' \
     --data '{"name":"Test Datastore"}'

Note: Remember to replace $API_KEY with your key. You can rename the datastore if you want.

If the request is successful, the id of the newly created datastore will be returned to you. Be sure to save this id as you will need it in subsequent steps!

Step 2: Add documents into your datastore

Now that you've created a datastore, you can add documents to it. All documents are stored securely in the Contextual platform, and are processed in ways that we have optimized for RAG agents.

If you don't have your own documents handy, feel free to use our Beginner's Guide test documents, found here
For the best results, use renderable PDFs, i.e., documents that have text that can be copied and pasted.

You can upload a single document using the following command:

from contextual import Contextual

# Initialize the client with your API key
contextual = Contextual(api_key="API_KEY")

# Upload a document
with open('file.pdf', 'rb') as f:
    ingestion_result = contextual.datastores.documents.ingest(datastore_id, file=f)
    document_id = ingestion_result.id
    print(f"Successfully uploaded document_id: {document_id} to datastore_id: {datastore_id}")

curl --request POST \
     --url https://api.contextual.ai/v1/datastores/{datastore_id}/documents \
     --header 'accept: application/json' \
     --header 'authorization: Bearer $API_KEY' \
     --header 'content-type: multipart/form-data' \
     --form file=@'${file_path}'

Note: Remember to:

Replace {datastore_id} in the url path with the datastore id from the previous step
Replace $API_KEY with your API key
Replace {file_path} with the path to the document on your machine

If the request is successful, the id of the uploaded document will be returned to you. The time required to upload documents depends partly on their length and features. Some documents may require a few minutes to fully process after upload.

To check the status of documents uploaded into the datastore, use this command:

from contextual import Contextual

# Initialize the client with your API key
contextual = Contextual(api_key="API_KEY")

# Get the status of documents in the datastore
metadata = contextual.datastores.documents.metadata(datastore_id=datastore_id, document_id=document_id)
print("Document metadata:", metadata)

curl --request GET \
     --url https://api.contextual.ai/v1/datastores/{datastore_id}/documents \
     --header 'accept: application/json' \
     --header 'authorization: Bearer $API_KEY'

Note: Remember to:

Replace {datastore_id} in the url path with the id from the previous step
Replace $API_KEY with your API key

You should see the document you uploaded in the list, along with its ingestion_job_status.

Step 3: Create an agent

Now that you have a datastore with some files, you can use the /agents endpoint to create your first agent.

from contextual import Contextual

# Initialize the client with your API key
contextual = Contextual(api_key="API_KEY")

# Create an agent
agent = contextual.agents.create(name="Test Agent", description="Test Agent", datastore_ids=[datastore_id])

curl --request POST \
     --url https://api.contextual.ai/v1/agents \
     --header 'accept: application/json' \
     --header 'authorization: Bearer $API_KEY' \
     --header 'content-type: application/json' \
     --data '
{
  "name": "Test",
  "description": "Test Agent",
  "datastore_ids": []
}
'

Note: Remember to:

Replace $API_KEY with your API key
Populate the datastore_ids list field with the datastore id from above

If the request is successful, the agent_id of the newly created agent will be returned to you. You'll need this to query your agent in the next step.

Step 4: Query your agent

Now that you've set up an agent and uploaded documents for use with it to use, you can use the /query endpoint to send messages:

from contextual import Contextual

# Initialize the client with your API key
contextual = Contextual(api_key="API_KEY")

# Query the agent
response = contextual.agents.query.create(
  agent_id=agent_id,
  messages=[
    {
      "role": "user",
      "content": "What is the revenue of Apple?"
    }]
)

curl --request POST \
     --url https://api.contextual.ai/v1/agents/{agent_id}/query \
     --header 'accept: application/json' \
     --header 'authorization: Bearer $API_KEY' \
     --header 'content-type: application/json' \
     --data '
{
  "stream": false,
  "messages": [
    {
      "role": "user",
      "content": "What is the revenue of Apple?"
    }
  ]
}
'

Note: Remember to:

Replace {agent_id} in the url path with the agent_id from the previous step
Replace $API_KEY with your API key
Replace the content field with a question that is relevant to the document(s) you uploaded

If the request is successful, you'll receive a response back that will contain:

The body of the response
The sources retrieved from the datastore that are relevant to the response
Attributions/citations of sources to specific text spans in the response

Note: You can only query your agent once at least one document in the datastore has been processed. You can check the status of uploaded documents by following the instructions in the previous step.

🙌 Congratulations! You've now created a basic agent in the Contextual platform.

Via Contextual API Docs

As an alternative to shell commands, you can also use the interactive areas of our documentation to interact with our APIs.

Step 1: Create a Datastore

To create a datastore via the API docs:

Expand the /datastores section in the sidebar
Select Create Datastore
Enter the name of your datastore in the Body Params field in the central panel
Input your API key in the Bearer field in the top of the right panel
Click the Try It! button
Save the returned datastore id for future reference

Step 2: Upload documents into your datastore

To upload a document:

Expand the /datastores section in the sidebar
Select Ingest Document
Enter the datastore id from the previous step in the datastore_id field
Click the upload button in the file field to select a file from your local machine for upload
Input your API key in the Bearer field in the top of the right panel
Click the Try It! button

Keep in mind that processing the document after upload can take a few minutes. To check the status of your uploaded document(s):

Expand the Datastores section in the sidebar
Select List Document
Enter the datastore id in the datastore_id field
Input your API key in the Bearer field in the top of the right panel
Click the Try It! button

Step 3: Create an agent

To create an agent linked to the datastore you created:

Expand the /agents section in the sidebar
Select Create Agent
Input a name and description in the respective fields in the central panel
Click Add String in the datastore_ids field and input the datastore_id from earlier
Input your API key in the Bearer field in the top of the right panel
Click the Try It! button

If successful, you will see a 200 response and the agent should also appear in your tenant. Save the returned agent id for the query step.

Step 4: Query your agent

To send a message to your agent:

Expand the /agents/{id}/query section in the sidebar
Select Query
Input the id from the prior step in the agent_id field
Click Add Object in the messages field and type a question that is relevant to the given document(s) you uploaded in the content field
Input your API key in the Bearer field in the top of the right panel
Click the Try It! button

If successful, you'll see a 200 response, the body of which will contain:

The body of the response
The sources retrieved from the datastore that are relevant to the response
Attributions/citations of sources to specific text spans in the response

WHAT'S NEXT

Now that you've got a basic working agent, you can explore our advanced features, including evaluation and tuning options.

Evaluation & Tuning Guide