Practice
What is Context Engineering?
Also known as: context management
Context engineering is the practice of deciding what goes into a model's context window on each call — instructions, retrieved data, tool results, history — so the model has what it needs and little else.
Prompt engineering asks how to phrase the request. Context engineering asks a bigger question: of everything that could go in the window — the system prompt, the last twenty turns, six retrieved documents, forty tool schemas, the output of the previous step — what actually earns its place? Context is finite and not free, and models degrade as it fills: relevant detail gets lost among the filler, cost and latency climb with every token. The craft is curation — retrieve what's needed now, summarise what's old, drop the rest.
The capability mesh is context engineering applied to tools. A thousand connectors cannot live in a prompt, so Velaris keeps them as searchable data the agent queries at runtime and loads only the handful a step needs. The same logic runs through the rest of the system: multi-step agents keep state in the graph rather than re-sending the whole history, and retrieval pulls documents in when a task calls for them. The aim isn't a bigger window — it's a window with the right things in it.
Related terms
See it in action
Velaris puts context engineering to work inside a real AI Operating System.
Get early access