Welcome to the Klu documentation. This site includes resources for using the API and SDK, guides to common Klu platform use cases, and best practices for deploying LLM Apps.
Last updated: December 5, 2023
Klu.ai LLM App Platform
Key Klu Concepts
Learn key concepts in Klu, from Actions and Context to Workspaces.
Your First App
Build your first app in Klu Studio and deploy it to production.
Integrate Existing App
Connect an existing OpenAI app to Klu in 5 minutes.
Key LLM concepts when just getting started with GenAI.
Klu.ai is an LLM App Platform that streamlines prototyping, deploying multiple models, evaluating performance, and optimizing LLM-powered applications and features. Klu integrates with your preferred Large Language Models, incorporating data from varied sources, giving your applications unique context.
Klu helps engineering teams rapidly build and iterate on LLM-powered applications. It provides a unified API access to LLMs like Anthropic Claude 2 and OpenAI GPT-4, allowing developers to quickly test prompt engineering and performance.
Klu supports all leading LLM providers. Klu comes with a global deployment of OpenAI models built on Azure, enabling all platform functionality to work immediately. Usage of the Klu models is included in the subscription, all other usage is handled by the connected platform. Leading AI Teams connect to additional models using their private API tokens. This ensures data privacy and enables fine-tuned models to live in your account or infrastructure.
|Klu (Azure GPT-4 Turbo, GPT-4, and GPT-3.5)||All Plans|
|Klu (Azure GPT-4 32k)||Enterprise|
|Azure OpenAI||All Plans|
|Google Vertex||All Plans|
|AWS Bedrock||All Plans|
|Cloudflare AI||All Plans|
Build & Evaluate
AI Engineers use Klu's SDKs to build LLM Apps directly into their applications and gather usage data on LLM performance. This allows efficient A/B testing of different prompts and models to optimize the end-user experience. Klu facilitates LLM evaluation through built-in support for logging, monitoring, and analysis. Developers can easily see how different prompts and models perform with real user input. These observations enable data-driven decisions around model selection, prompt engineering, and fine-tuning.
For storage and retrieval of knowledge, Klu has built-in support to index and query embeddings, supporting a range of file types, datagbases, and integrations. This enables retrieval augmented generation, reduction in hallucinations, semantic search, and other vector similarity applications out-of-the-box.
Klu supports a variety of file types across different categories. Below you'll find a table listing all supported file extensions, which you can refer to when uploading content to the platform. If your file type is not listed, please contact the Klu team for assistance. Please note: old versions of Office files and scanned documents in PDFs are known for lower performance.
|Category||File Type Extensions|
|Audio / Video||.mp3, .mp4|
|Documents||.pdf, .rtf, .txt|
|Email Files||.eml, .msg|
|Markup/Structured Text||.md, .html, .rst, .org, .xml|
|Office Documents||.doc, .docx, .xlsx, .xls, .ppt, .pptx, .odt|
|Structured Data||.csv, .tsv|
Klu integrates with an array of platforms for use as Context. Below is a detailed list of all the supported integrations, categorized by their use case. If you require integration with a platform not listed here, please reach out to the Klu team for further assistance.
|Collaboration||Google, MS Teams, Slack, Zoom|
|Customer Platforms||Intercom, Salesforce, Zendesk|
|Projects||Airtable, Asana, Atlassian, Github, Notion|
|Websites (Crawling)||HTML, Sitemap|
|SQL Database||MySQL, PostgreSQL, SQLite, Oracle, SQL Server|
|Redis||All data types (string, list, set, zset, hash)|
|Elastic||All data types|
|Snowflake||All data types|
|Youtube||All video formats|
Teams using Klu establish a defensive moat by making AI capabilities harder to reproduce. By accelerating the build-measure-learn loop, Klu empowers AI Teams to quickly deliver AI-powered capabilities that users love and trust. Klu provides the tooling for rapid experimentation and evaluation, while engineering teams focus on their domain expertise to delight their users.
The more an AI Team iterates, the more fine-tuned the AI becomes to their use case and users' preferences.