What we have shipped on AWS, OpenAI, LangChain, LiveKit and Langfuse, and how working with us affects your vendor choices.
Most of what we ship runs on AWS: Bedrock for models, AgentCore for hosting agents, Lambda and ECS for services, and pgvector on RDS for retrieval. We set it up in your AWS account, so the bill and the access stay with you.
We use OpenAI models where they win the evaluation for the task. GPT-4.1 handles reasoning in a production voice assistant, and GPT-4 writes the lesson scripts in a content pipeline that has produced more than 200 course modules. The API keys and the bill sit in your own OpenAI account, and model calls go through your own code, so the provider can change later.
LangGraph is our default for stateful agents that pause for human approval, call real backend services and need to be observable. Our founder has contributed to the LangChain open-source project.
Hire for it
LangGraph developers
We build voice agents on LiveKit Agents with response times under half a second, and we run self-hosted LiveKit for teams that need the media server in their own cloud.
Shipped with it
Hire for it
LiveKit and voice AI developersTracing, evaluation and cost per request go in with the first release, and Langfuse is how we usually do it. We have run it self-hosted and on Langfuse Cloud, and we share feedback with their team through a shared Slack channel.
Shipped with it
Hire for it
Langfuse developersNo. We pick the platform per project and write down why. Model calls go through a thin layer of your own code, so switching from one provider to another is a configuration change plus an evaluation run, not a rewrite.
We do not resell licences or cloud usage. You pay AWS, Langfuse or any other vendor directly, on your own account. If a vendor ever paid us for a referral on your project, we would tell you before you signed.
Yours, from day one. The repository, cloud account, model API keys and observability project are created in your company's name, and we work in them with access you can revoke.
Usually, yes. We start with a fixed-price review of what exists, so you know what is worth keeping before anyone commits to a build.
