laminar

All-in-one platform for production-grade AI applications.

Productivity

API

SaaS

Hey Product Hunt 👋  We were building an AI startup and quickly realized that hard-coding prompts and chains of model calls is very inflexible and severely slows down development cycle. Solving these problems for ourselves, we gradually ended up with a universal platform not just limited to our use case. Thus, ➡️ laminar was born. https://www.lmnr.ai (desktop only) We realized that modern AI apps converge to graph-like flows, where you have chains of different model calls, semantic search, and so on. However, majority of AI startups right now hard-code these sequential calls on their backends, and this results in inflexible and slow experimentation. This is where ➡️ laminar comes into light. At its core laminar is an extremely fast async flow execution backend written in Rust, coupled with user-friendly graph UI to build LLM flows (e.g. RAG, HyDE search). Due to its async nature, you can build parallel workflows, such as running multiple models at the same time, out of the box. Moreover, for each pipeline you build, you can experiment with dozens of inputs in parallel! In the next release we will add extremely fast, parallel evaluations i.e. “model unit tests”! Once we’ve built the core, we realized that we needed to separate our experiments from our production endpoints, so we’ve added different types of pipeline versions — workshop and deployed — one for development, and the other one to be called via exposed API from user applications. Later, every time new deployment version was ready, we had to alter our client code to call the updated version. We understood that this experience is far from perfect, so we’ve added an ability to redeploy (update) any deployed version, thus resulting in an intuitive and powerful version-control system. Thanks to Rust, it was very easy to add async, zero-overhead logs for each step of each pipeline execution. We wanted to have logs for each core node - CoT Agent, Chat, Search … - so that in (very) near future you can fine-tune each node/component using stored logs. On top of that, our UI provides simple way of monitoring performance of your pipelines from logs. Stay tuned for AI analytics tools for ➡️ laminar, built using ➡️ laminar 😉! We already support OpenAI and Mistral models, and we are planning to expand this list to other LLM providers and soon vision/STT/TTS models. ➡️ laminar is still rough around the edges, but we were too excited to release it and get first feedback from you! We welcome all next-gen AI builders to give it try and give us feedback in our discord server or in the comments!