Every tool on this page is one we use in production — for our own Friday One product, for our reference architectures, and for client engagements. No vendor kickbacks. Honest engineering choices.
✦ Every tool on this page is in active use on real projects — not aspirational.
Looking for our Partner Program? It moved.Our preferred LLM provider for production systems. Claude's long context, tool-use reliability, and strong safety properties make it our default for agentic workflows.
GPT-4o for multimodal tasks and function-calling pipelines. We use the Assistants API for stateful agent contexts and the Batch API for high-volume evaluation runs.
LangGraph is our primary framework for building stateful, multi-step agentic workflows — graph-based orchestration with full control over agent state and decision branching.
Managed vector database for production RAG pipelines. We use Pinecone for high-throughput semantic search where pgvector's operational overhead is a constraint.
LLMOps and agent observability. LangSmith traces every agent run, records prompt/response pairs, and powers our prompt regression test suites.
Experiment tracking for fine-tuning runs and ML pipeline monitoring. Every model training job we run is tracked in W&B for reproducibility.
Primary cloud platform. We use EKS for Kubernetes, SageMaker for ML workloads, Bedrock for managed LLM inference, and CDK for infrastructure as code.
Preferred for BigQuery analytics, Vertex AI Agent Builder, and Gemini API integrations. GCP is our default for data-intensive AI workloads.
Enterprise integrations, Azure OpenAI Service, Active Directory, and hybrid cloud deployments. Common in Microsoft-stack enterprise environments.
Source control, CI/CD via GitHub Actions, and code review on every project. All agent evaluation runs are gated in CI.
Next.js deployments, edge functions, and preview environments for every PR. Our default for frontend-heavy products.
Postgres-backed backend with pgvector enabled — our default for RAG pipelines where a managed vector DB is unnecessary. Also used for auth and realtime.
Type-safe ORM for production-grade database access. Our standard for any TypeScript backend with a relational data model.
Error tracking and performance monitoring for production deployments. We instrument every service before go-live.
Infrastructure as code for all cloud environments. Every piece of infrastructure we manage is version-controlled and reproducible.
Every engagement we take on runs on this exact toolset — no switching costs, no vendor lock-in surprises. Tell us what you are building and we will put together a proposal within 48 hours.
