Opper AI

Update! We have added a new API for detailed tracking of model usage and inference costs. You can now use the Opper CLI to get statistics on tokens and cost per model, project, timeframe and more. We also added a tags parameter to calls to be able to annotate calls with custom information (such as end user or customer). This is a start of our plans of providing a comprehensive understanding of AI spend and utilisation and will soon include utilisation of indexes, datasets and more.

New feature! Opper now supports testing and evaluating the performance and quality of different models and configurations for a given task. Populate a dataset with a set of expected input and outputs, and with a click (or a few keystrokes with our API/CLI) find out which models perform.

New blog post on building an agentic customer service chatbot with tool calling. We roll our own function / tool calling abstractions with task specific functions that makes it possible to monitor, control and optimize each part of the agent. https://lnkd.in/dypV-vp5

New blog post on building an agentic customer service chatbot with tool calling. We roll our own function / tool calling abstractions with task specific functions that makes it possible to monitor, control and optimize each part of the agent. https://lnkd.in/dypV-vp5

Update! We've expanded our model support beyond our default 30+ frontier and open weights models to now include custom models and BYO API keys! Great for exposing fine tuned or yet unsupported models to developers. Plus, you can now specify fallback models per call - perfect for managing rate limits and reliability.

Update! We've expanded our model support beyond our default 30+ frontier and open weights models to now include custom models and BYO API keys! Great for exposing fine tuned or yet unsupported models to developers. Plus, you can now specify fallback models per call - perfect for managing rate limits and reliability.

We are working towards open sourcing another sample AI project! This time it is an autonomous web agent built using Opper AI to autonomously perform arbitrary tasks on websites. "Sign up to Dulingo.com and take a class in French" 🚀 Will be posting more examples in the comments (autonomous sending of emails, signing up and building a recipy list, autonomously posting messages on X etc). Yes, the agent should have posted this message! But if we are lucky he/she/it (?!) might appear in the comments!

Opper goes multi modal with the addition of image inputs and outputs, and audio input to calls. Programmatically generate avatars, perform OCR and build features with voice input with the same syntax and control as any other Opper call. Mix calls using the best multi modal models from OpenAI, Google, Mistral and more. See our new patterns library for more inspiration: https://lnkd.in/d9M2sSww

How well does o1-preview and o1-mini work in existing RAG pipelines that makes heavy use of structured output? We explored this in our latest blog post "Using o1-preview and o1-mini with RAG and structured output" 

We have made the new o1 model from OpenAI available on Opper to all users. It stands out by having very impressive reasoning capabilities, with a tradeoff of higher round trip time and token usage.