Anthropic's Claude is our go-to model for complex reasoning, long-context analysis, and safe, controllable AI agent behavior. We use Claude 3.5 Sonnet and Claude 3 Opus for tasks requiring nuanced judgment, multi-step reasoning, and reliable output.

GPT-4o and GPT-4 Turbo power our conversational AI products, customer-facing chatbots, and vision-enabled workflows. OpenAI's function calling and structured outputs make it the most API-flexible model for complex integrations.

Gemini 1.5 Pro's 1M-token context window makes it uniquely capable for processing entire codebases, large PDFs, and hour-long video transcripts. We use it for enterprise document intelligence and multi-modal AI pipelines.

The industry standard for building LLM-powered applications. LangChain gives us composable chains, tool use, memory management, and a massive ecosystem of integrations. We use it to build RAG systems, multi-step pipelines, and LLM-backed APIs in production.

LangGraph extends LangChain with stateful, cyclical graph execution — the key to building truly autonomous AI agents with loops, branching, and self-correction. We use it for complex multi-agent workflows where an agent needs to evaluate its own outputs and iterate.

CrewAI makes multi-agent collaboration feel natural — define a crew of specialized agents (Researcher, Writer, Analyst) and have them work together on a shared objective. We use it for document research workflows, competitive intelligence pipelines, and content generation systems.

Our primary open-source automation platform. n8n lets us self-host complex AI workflows, connecting LLMs to CRMs, databases, webhooks, and 400+ integrations. We run n8n for lead qualification bots, AI-powered email routing, and multi-step AI pipelines for enterprise clients.

Make (formerly Integromat) is our cloud-native automation choice for clients who need visual, no-code AI workflows. We build AI-augmented scenarios that sync data between Salesforce, Slack, Google Sheets, and custom APIs — all triggered by AI-classified events.

The Vercel AI SDK gives us a unified TypeScript interface to stream responses from Claude, OpenAI, and Gemini directly into Next.js applications. We use it for AI-powered SaaS products, real-time chat interfaces, and server-side AI generation with edge deployment.

Pinecone is our managed vector database of choice for production RAG systems requiring millisecond-latency semantic search at scale. We use it to store embedding vectors from client documents, product catalogs, and knowledge bases — powering AI search that understands meaning, not just keywords.

Qdrant is our self-hosted vector search engine for clients with data sovereignty requirements. Written in Rust for maximum performance, Qdrant handles filtered vector search with metadata — perfect for multi-tenant AI applications where you need to isolate each user's knowledge space.

Weaviate brings a GraphQL interface and native multi-modal vector search — enabling AI systems to search across text, images, and structured data simultaneously. We use it for knowledge graph-style RAG pipelines and enterprise search applications combining semantic and keyword retrieval (hybrid search).

Supabase is the open-source Firebase alternative we use as the backend for AI SaaS products — combining Postgres, real-time subscriptions, Auth, Storage, and Edge Functions in one platform. We rely on Supabase's pgvector extension to run vector search inside Postgres when a dedicated vector DB isn't required.

Postgres is the relational backbone of every production AI application we build. It stores structured application data, audit logs, user records, and — via the pgvector extension — serves as a lightweight vector store. Its JSONB support makes it ideal for storing unstructured LLM outputs alongside structured metadata.

Redis is the memory and caching layer for our AI applications — storing conversation history for LLM context windows, caching expensive LLM responses to cut API costs, rate-limiting AI endpoints, and managing real-time queues for async AI job processing. Essential for production-grade AI at scale.