contextual.ai

•

API Key

About Contextual AI

Contextual AI is an enterprise AI platform specializing in retrieval-augmented generation (RAG) for complex document-heavy industries including engineering, finance, legal, and manufacturing. It provides a multi-stage document understanding pipeline, a state-of-the-art instruction-following reranker, and an LLM engineered to minimize hallucinations. Responses include clear source attributions, which is critical for regulated enterprise use cases. The platform is available on AWS Marketplace as a single-tenant deployment.

AI Agent Use Cases

- Multi-stage document understanding pipeline for unstructured content ingestion

- Instruction-following reranker for high-precision retrieval over enterprise data

- Hallucination-minimizing LLM with cited, grounded responses

- SOC 2 certified with single-tenant deployment on AWS Marketplace

- Component APIs for usage-based integration into existing RAG architectures

- Targeted at engineering, finance, legal, and manufacturing enterprise teams

Available Actions

These are the specific actions that AI agents can perform with this tool

Create Agent

9 inputs

Create a new agent by providing its name, instructions, configurations, and which datastores to connect.

Inputs

name

Name of the agent to be created.

system_prompt

Instructions that your agent references when generating responses.

no_retrieval_system_prompt

Instructions for the agent when no relevant retrievals are found.

filter_prompt

Prompt for filtering out irrelevant retrieved chunks.

suggested_queries

Sample queries shown to users in the UI.

agent_configs

Advanced configurations for retrieval, filtering, response generation, etc.

description

Description of the agent.

datastore_ids

IDs of the datastores to associate with this agent.

api_key

API key to authenticate requests using Bearer authentication in the request header.

Create Datastore

1 input

Create a new datastore with a specified name.

Inputs

name

The name to give to the new datastore.

List Agents

3 inputs

List all agents currently available in the account, with optional pagination.

Inputs

limit

Maximum number of agents to return in the response.

cursor

Pagination cursor returned in a previous response, if retrieving next set of agent results.

Authorization

Bearer authentication token used to authorize this API call.

List Datastores

4 inputs

List all datastores, optionally filtered by agent and paginated.

Inputs

agent_id

ID of the agent used to filter datastores. Only datastores linked to this agent will be returned if specified.

limit

Maximum number of datastores to return. Must be between 1 and 1000.

cursor

Cursor from the previous call for pagination. Used to retrieve the next set of results.

Authorization

Bearer authentication header in the form 'Bearer <token>', where <token> is your API key. Must be passed as a request header.

Query Agent Conversation

9 inputs

Ask an agent a question or continue a conversation and retrieve a structured response including relevant document content and metadata.

Inputs

agent_id

Agent ID of the agent to query.

messages

Messages sent so far in the conversation, with the latest message from the user.

stream

If true, receive a streamed response.

conversation_id

Optional conversation identifier to continue an existing conversation.

llm_model_id

Specific model ID to use for the query. Defaults to the base model if not supplied.

structured_output

If provided, customizes the structure and schema of the output JSON.

documents_filters

Custom metadata filters for documents, as a composable JSON expression.

retrievals_only

If true, only retrieves content and metadata, omits agent response generation.

include_retrieval_content_text

If true, also includes the full text of retrieved content. Ignored if 'retrievals_only' is true.