4 ms·
Show HN: Opper AI – Task-Completion API for LLMs
Hey HN! We’re building Opper AI (https://opper.ai https://opper.ai), a task-completion API for large language models. You describe tasks once in JSON - inputs, expected output, and a success test — and call a single endpoint. Opper handles prompting, retries, fallbacks, and evaluation. You can check out our docs here: https://docs.opper.ai https://docs.opper.ai.
LLM features often fail in production due to fragile, model-specific prompt chains. That’s why we built a task-completion API to make model calls as reliable as any other API: you declare what you want, and it manages the interaction with 80+ proprietary and open-source models. If the output fails, Opper retries or falls back automatically. Successful completions can be saved as task specific dataset entries and serve as examples for future generations. When the completion passes (or retries are exhausted), you get structured output plus a pass/fail flag.
For example, gettested.io previously spent weeks rewriting prompts to generate consistent blood test summaries. With Opper, a single task definition now delivers reports in 62 countries without any edits.
Key features:
- Define tasks in JSON: inputs, outputs, success test
- Automatic prompt construction, retries and fallbacks
- Quality through task specific datasets and in-context learning
- Full observability (llm-as-a-judge, prompts, responses, tokens, costs)
- Free tier up to $5/mon. Utility plan from $5/month gets you high rate limits
We built Opper after our last startup, Unomaly (ML observability, acquired 2020). Our vision with Opper is to let developers work with LLMs the same way they write code and get reliable results.
The task-completion API is live today. We’d love for you to give it a try and would appreciate any feedback you have.
Thanks for checking it out!
- jpnagel 1y agoBeen using Opper for a while! Great tool for speeding up work with LLMs
- jhgustafsson 1y agoThx!
- gsandahl 1y agoCo-Founder here thanks for taking a look at Opper! I’m hanging around the thread all day, so feel free to ask anything, share feedback, or tell us where you’d like the product to go next
- lucasalska 1y agoBeen using Opper for a while, and my absolute favorite way of building AI apps. Talk about DX and implementation speed!
- gsandahl 1y agoThanks for the shout out!
- sebastian1998 1y agoHey, I am the founder of TenderRender. We're looking into question specific in-context learning examples. So retrieving different examples based on the prompt. Is this possible?
- gsandahl 1y agoYes that's possible! You can populate examples of great outputs to task specific datasets and have those be automatically populated to the prompt. More info here: https://docs.opper.ai/capabilities/learning https://docs.opper.ai/capabilities/learning
- ValtteriK 1y agoDo you have a demo video available for the latest features/use cases? We are building a digital end-of-life service platform with customized "Bereavement AI" where we have tried to switch between GPT-4o mini and Gemini 2.5 Flash for different types user questions. Would be interesting to give this a spin in our use case!
- gsandahl 1y agoNo up to date demo video unfortunately :( Sounds like a great use case though!
- jlundberg 1y agoWhat are your thoughts on longetivity, i.e. depending on your service over time? Do you have any public commitments in that regard? Being able to run your application forever is one of the strong benefits I see with local models, besides being able to feed it with sensitive data.
- gsandahl 1y agoWe have been thinking a bit about this, and one option would be to have some form of locally hosted runner. You can optimize the task in the cloud and deploy it locally. Something like that. It is possible to plug in custom models so technically feasible.
- jhgustafsson2 1y agoHey, other co-founder here. Happy to answer questions.
- nielsbosma 1y agoLong-time user of Opper! Can recommend! Makes building LLM apps so much faster.
- felix089 1y agoThanks happy to hear Niels!
- WilhelmHomer 1y agoWay to go Opper!
- felix089 1y agothx!
- while1 1y agoReally cool tool! Great job!
- trematerra 1y agoAmazing Opper! Definitely worth giving a try
- gsandahl 1y agoPlease do and give us some feedback!
- viborcip 1y agoCongrats, team! I've been following you for quite some time, happy to see you outta stealth and crushing it! My only question is: What has been the most exciting learning experience you've had while building Opper so far?
- gsandahl 1y agoI think just how far you can go with examples has been an interesting learning! As these models have become smarter, they are also getting better at reasoning from examples and understanding intent. We will be publishing some research in the next few days!
- not_alex 1y agoAwesome!