The stack

What we build the agents with

Six groups of tools, one gesture each. Open any of them to see the names and what we use them for.

The stack

Frameworks & multi-agent orchestration

The agent’s backbone: stateful workflows, hand-offs between agents, and the human-in-the-loop gate where it waits for your word.

Frameworks & multi-agent orchestration

The agent’s backbone: stateful workflows, hand-offs between agents, and the human-in-the-loop gate where it waits for your word.

  • LangGraph / LangChainstateful, cyclic graphs, complex workflows and human-in-the-loop controls
  • CrewAIrole-based multi-agent systems and task delegation
  • Microsoft Agent FrameworkAutoGen + Semantic Kernel — unified enterprise orchestration
  • OpenAI Agents SDK · Anthropic Claude Agent SDK · Google ADKthe providers’ own agent SDKs
  • Mastra · PydanticAITypeScript and type-safe Python agent frameworks

The stack

Protocol, tooling & execution standards

How an agent reaches your tools and the web, and where it is allowed to run code — in a box, never on your desk.

Protocol, tooling & execution standards

How an agent reaches your tools and the web, and where it is allowed to run code — in a box, never on your desk.

  • Model Context Protocol (MCP)the open standard that connects agents to APIs, databases and enterprise tools, with native OAuth
  • Playwright · Stagehand · Browserbase · Firecrawlautonomous web interaction, scraping and browser agents
  • Docker · E2B · Modalsandboxed code generation and execution loops

The stack

Foundation models & reasoning engines

The minds we put behind an agent — a frontier reasoning model where judgement matters, a small model on your own hardware where privacy or cost does.

Foundation models & reasoning engines

The minds we put behind an agent — a frontier reasoning model where judgement matters, a small model on your own hardware where privacy or cost does.

  • Anthropic Claude — Opus · Sonnet · Haikufrontier reasoning and long, careful work
  • OpenAI GPT and o-seriesgeneral and reasoning models
  • Google Geminimultimodal reasoning at scale
  • DeepSeek R-seriesopen reasoning models
  • Llama · Phi · Gemma · Qwensmall and on-device language models for cost-efficient, private or edge deployments

The stack

Agent memory, state & knowledge

What the agent remembers about you and your business, and how it finds the right page of it in a second.

Agent memory, state & knowledge

What the agent remembers about you and your business, and how it finds the right page of it in a second.

  • Qdrant · Pinecone · Weaviate · Milvusvector databases
  • Neo4j · pgvector (PostgreSQL)graph and hybrid search for graph RAG and long-term agent state
  • LlamaIndex Workflows · UnstructuredRAG, document parsing and context engineering

The stack

Evaluation, observability & guardrails

How we prove an agent did the right thing, step by step, and the rails that keep it from doing the wrong one.

Evaluation, observability & guardrails

How we prove an agent did the right thing, step by step, and the rails that keep it from doing the wrong one.

  • LangSmith · Phoenix (Arize) · Helicone · AgentOpsstep-by-step latency, token cost and execution tracing
  • NeMo Guardrails · Guardrails AI · Pydanticagent guardrails and strict output schemas
  • Ragas · DeepEvaldeterministic evaluation of multi-step agent actions

The stack

Infrastructure & deployment

Where it all runs — a cloud if you want one, your own server room if you don’t — and the queues that keep a long-running agent alive.

Infrastructure & deployment

Where it all runs — a cloud if you want one, your own server room if you don’t — and the queues that keep a long-running agent alive.

  • FastAPI · Python · Node.js / TypeScriptbackend services
  • AWS Bedrock / Lambda · GCP Vertex AI · Azure AI Studio · Vercel AI SDKcloud and serverless runtimes
  • Celery · Redis · Temporalresilient, long-running agent workflows

Say hello

That is the stack.

Tell us what your business does at night, and we will tell you what its AI could be doing.

HomeWhat we have builtWebsites