HomepageContextual AI Components
State-of-the-art component APIs to enhance every step of your RAG pipeline
CORE RAG BUILDING BLOCKS
Optimize your RAG pipeline end-to-end with APIs for document understanding, reranking, generation, and evaluation
STATE-OF-THE-ART PERFORMANCE
Ground agents in your enterprise knowledge with RAG components that outperform competing tools on leading benchmarks
SIMPLE, FLEXIBLE INTEGRATION
Incorporate components into your existing RAG pipeline without needing to overhaul your entire architecture
Better RAG performance in minutes
Powerful, modular RAG components for when accuracy and simplicity matter
Our multi-stage document understanding pipeline for converting unstructured content into AI-ready formats
The first instruction-following reranker, providing greater control over how retrieved knowledge is prioritized
The most grounded large language model in the world, engineered specifically to minimize hallucinations
Our evaluation-optimized model for preference, direct scoring, and natural language unit test evaluation
Achieve even greater performance
with the Contextual AI Platform
Each component API provides state-of-the-art performance. Our end-to-end platform delivers even greater performance and simplicity by jointly optimizing and orchestrating the components as a single, unified system.
Flexible, modular components
PARSE
Extract complex multimodal content from any document
- Convert unstructured documents into structured output optimized for your RAG pipeline
- Easily process text, charts, tables, code, and other complex modalities
- Infer document hierarchy and add positional metadata to each chunk, enabling agents to connect information across hundreds of pages
RERANK
Resolve knowledge conflicts with the only instruction-following reranker
- Steer how your reranker prioritizes information with natural language instructions
- Serve the most relevant content to your model’s context—with or without instructions
- Drop into your existing RAG pipeline with just a few lines of code
GENERATE
Maximize accuracy with the world’s most grounded language model
- Generate answers strongly grounded in your retrieved documents
- Minimize hallucinations with response tags to delineate between facts and model commentary
- Provide in-line attributions to source documents for end-users to verify responses
LMUnit
Evaluate LLM responses with fine-grained natural language unit tests
- Assess responses for your defined criteria, like conciseness, technical precision, and more
- Test for accuracy with out-of-the-box evaluation tests for equivalence and groundedness
- Reduce your dependence on end-user testing and get to production faster
Simple APIs to improve any RAG pipeline
# Convert unstructured content into structured output
import os
from contextual import ContextualAI
client = ContextualAI(api_key=os.environ.get("CONTEXTUAL_API_KEY"))
with open("your filepath", "rb") as fp:
response = client.parse.create(
raw_file=fp,
parse_mode="standard",
figure_caption_mode="concise",
enable_document_hierarchy=True,
page_range="0-5",
)
job_id = response.job_id
results = client.parse.job_results(job_id, output_types=['markdown-per-page'])# Rank retrieved documents using instructions
import os
from contextual import ContextualAI
client = ContextualAI(api_key=os.environ.get("CONTEXTUAL_API_KEY"))
rerank_response = client.rerank.create(
query = "your query",
instruction = "your instructions",
documents = ["your documents"],
metadata = ["your metadata"],
model = "ctxl-rerank-en-v1-instruct"
)
print(rerank_response.to_dict())# Generate highly grounded responses
import os
from contextual import ContextualAI
client = ContextualAI(api_key=os.environ.get("CONTEXTUAL_API_KEY"))
generate_response = client.generate.create(
model="v1",
messages=[{
"content": "content",
"role": "user",
}],
knowledge=["your knowledge"],
avoid_commentary=False
)
print(generate_response.response)# Evaluate responses with user-defined unit tests
import os
from contextual import ContextualAI
client = ContextualAI(api_key=os.environ.get("CONTEXTUAL_API_KEY"))
response = client.lmunit.create(
query="your query",
response="your response",
unit_test="your unit test"
)
print(response)
“At ClaimWise, we work with complex patents and scientific articles. Meeting the quality expectations of our users has always been one of our main technical challenges. The Contextual AI reranker’s instruction-following capability is great; we have some nuanced requirements about how to prioritize different types of paragraphs for different fields of invention, and the reranker follows those instructions well.”