A working AI system, layer by layer.
Read it top to bottom, from what people touch to what it all runs on. Open a layer to see its tools.
Surface
Where people meet the system.
Capability
What turns a raw model into useful work.
- Agent frameworks Libraries for wiring models into agents that plan, call tools, and loop towards a goal.
- Workflow & automation Engines that run multi-step processes, with AI as one decision node among many.
- RAG & knowledge Retrieval pipelines that ground a model's answers in your own documents and data.
- Document extraction Turning PDFs, scans, and messy files into clean, structured text a model can use.
- Voice & transcription Speech to text and text to speech: the audio edges of an AI system.
- Browser automation Driving a real browser so an agent can use the web like a person does.
- MCP & connectors The Model Context Protocol and the connectors that plug models into live systems.
Model
The intelligence itself, and how you reach it.
- Gateways & routers Proxies that sit in front of many models: one API, with failover and spend control.
- LLM providers The commercial frontier model APIs, and the default intelligence most systems start with.
- Open-weight models Model weights you can download, inspect, and run yourself.
- Model serving Runtimes that turn model weights into a fast, callable endpoint.
Infrastructure
What everything above runs on.
