Prototype, evaluate, and optimize LLM-powered apps

AI Teams use Klu.ai to collaborate seamlessly, confidently evaluate performance, and optimize LLM-powered applications.

Zero to production in under 10 minutes.

Fully tested with every iteration.

HomeDocsChatRSS

Copyright © Klu, Inc. 2024

Platform Updates

Collaboration and usability improvements to dashboard, apps, and actions.

Collaboration Real-time comments, presence, and notifications

Dashboard Simplified dashboard layout

Apps Updated App overview layout

Actions Updated Action overview with versions

Insights 2.0 Merged Observe and Insights functionality

Studio Advanced ouput parsing only available when activated in settings

Platform Updates

CLI klu init to start a new project

Evals Evaluation results data table improvements

Insights Associate External UID with generations

SDK Feedback labels are human-readable (positive, negative)

Platform Updates

Data Import LLM logs via UI (JSONL or CSV)

Evals Configure real-time monitoring notifications

Studio Templates for Completion actions

SDK Developer improvements for using Datasets

Studio Updates

Studio Load previous versions into studio

Studio Autocomplete variables

Studio Upload images for use with Google Gemini or OpenAI GPT-4 Vision

LLM Provider Updates

Added support for PPLX online models, which outperform GPT-3.5 and Llama 2 across freshness and factuality dimenions due to their real-time search abilities.

LLM PPLX Online LLM pplx-7b-online (Based on Mistral 7b)

LLM PPLX Online LLM pplx-70b-online (Based on Llama 2 70b)

Action & Context Improvements

Klu Public Beta: improved search, performance, SDK, and Studio

Actions Deploy & Versions Beta

Context Segmented search for infinite token queries

Context Improved JSON chunking

Studio Improved readability for prompts and responses

Studio Search across Actions and Apps when opening Actions

New SDK improvements

SDK Used cached responses

SDK Retrieve all Feedback for a Session

SDK Search and prompt Context

App performance enhancements

Performance Imagen-app Context browsing caching

Onboarding and Platform Updates

Klu Public Beta: improved Onboarding, Studio, Context, and Data

PLG New login and onboarding

PLG Self-serve plans and usage

Studio New navigation for Connections and Context

Studio Simplified App creation

Studio Improved help and empty states

Context Improved pagination and search

Context Improved file management and incremental indexing of new files

Context Improved document management with pagination and in-app search

Data Added debug log to Data UI for failed generations

Optimize Full logging for fine-tune failure

Enterprise LLM Upgrades

LLM Klu LLM for Enterprise plans

LLM Multi-region proxy for Enterprise plans

Platform performance improvements

Performance 5x faster app page loads

Infrastructure Migrated Klu Vector DB to HNSW

Context and Engine Improvements

Platform improvements for the Klu Public Beta

Data Quickly page through data points

Context Import Youtube videos for Context

Context Web crawler now supports sitemaps

Context Web crawler now supports sitemaps

Engine Transcription at scale (10+ GB of audio/video)

Engine Performance caching across platform

Klu Studio Beta & Platform Improvements

All-new Prompt Engineering Editor for Actions now in Beta

Studio Prompt Engineering for Actions

Studio Resizable UI panels

Studio Dark Mode Beta

New Context library management experience

Context New Overview for libaries

Context Icons for Context library types

Context See which Actions are using a library

Context Edit Documents in a library

Context Indexing progress indicator and document count

Workflow improvements

Workflows View & Export Run Data

Workflows 1-Click Deploy Workflows to Public UI

Expanded Feedback management

Feedback Expanded feedback collection for generations

Feedback Improved data feedback UI

Platform improvements

Engine Specific messaging when LLMs respond with errors

Engine Refactored streaming for performance

LLM Klu LLM available in all Workspaces

SDK LLM provider and model management

New OpenAI Model

OpenAI announced the GPT-3.5-turbo-instruct model, the successor to the Davinci family and a better option for non-Chat Klu Actions.

LLM OpenAI GPT-3.5-turbo-instruct

Klu Optimize GPT 3.5 Fine-tuning

OpenAI's fine-tuning capabilities for GPT-3.5-turbo gpt-3.5-turbo-0603 launched yesterday. Klu Optimize now supports fine-tuning models based on your data. Klu also supports the new Davinci and Babbage variations davinci-002 and babbage-002 for fine-tuning – these replace the legacy versions of the previous models slated for deprecation on 2024-01-04.

Optimize GPT-3.5-turbo fine-tuning

Optimize Davicini-002 fine-tuning

Optimize Babbage-002 fine-tuning

Optimize Demo: ChatJRE, fine-tuned on Joe Rogan content

Klu Observe

Klu Observe now aggregates all workspace usage and generative data analysis into one view across the entire Workspace.

Observe Usage Costs

Observe Workspace Data and Usage

Observe Top Models and Data Sources

Platform Improvements

Actions Input Token Estimations Displayed

Context Zendesk

Deploy Action-specific deployment instructions for API and SDK

Deploy Optional Caching for API and SDK generations

Playground Warning when about to lose changes

Platform Improvements

Deploy Action variables now validated

Members Pending member invites now display in Workspace

Optimize Test experiments in Klu Studio

New Skills

Skills Describe Image or URL, Request Raw HTML

Skills Answer questions from structured data with Pandas

Skills Run Terminal commands

Skills Query SQL in Natural Language

UX Improvements

Data Filter by feedback and labels

Integrations Disconnect and reconnect to a new account

LLM Set default model for all providers in Workspace

Optimize Fine-tune with custom data frames

Optimize View more Action details in Experiments

Data Backfill

SDK Data Import

Backfill your existing prompt and completion data to Klu through the SDK for optimization with RLHF and model fine-tuning.

Optimize with Experiments

Optimize Manage Experiments

API Experiments

You can now optimize actions by testing head-to-head performance in your real-world applications. Test different providers, model configurations, or prompts while tracking feedback and model performance.

Data, Optimize, Skills

This release introduces a range of new features and updates including fine-tuning API improvements, new data management and export capabilities, ability to fine-tune OpenAI models, SDK enhancements for better data navigation and streaming, and new AI Skills creation with Python and Zapier NLA integration.

Klu Observe

Data Filter, sort, and search

We've introduced the ability to filter, sort, and search your data, allowing you to gain insights and easily locate specific datasets.

Data Export JSONL, CSV

We now support exporting your data in both JSONL and CSV formats, offering more options for you to interact with your data outside the platform.

Klu Optimize

Optimize Fine-tune OpenAI models

You can now fine-tune OpenAI models within Klu.ai using your data, bringing the power of OpenAI's LLM directly to your applications.

New Skills

Skills Python, Zapier NLA

You can now create AI Skills using Python, allowing you to take advantage of the language's extensive ecosystem of libraries and frameworks. We have introduced Natural Language Actions (NLA) integration with Zapier, enabling you to connect and automate tasks in over 5,000+ apps and 30,000+ actions using natural language commands.

API Enhancements

API Finetune

We have expanded capabilities allowing you to create, read, and delete fine-tuning jobs. This provides you with a more flexible way to control and manage your fine-tuning workflows.

API Skills

We introduced the ability to create skills, making it easier to manage and deploy different AI models and algorithms in your applications.

SDK

SDK Pagination, Prompt Streaming

We've improved our SDK to support pagination, allowing you to efficiently navigate through large data sets or results. We have introduced the ability for prompt streaming in our SDK. This allows you to feed in data to your apps in a stream, allowing for more flexible and efficient end user experience.

Templates

Actions Configure model config

Action Templates Quickly create new Actions

App Management Move, copy, delete Actions

In-App Feedback Send feedback to Klu team

Integrations Delete Context, Skills

Workspace Remove Members, delete Workspace