Guide
Workspace context
engineering
In real developer agents, context is not just prompt text. The workspace includes files, terminals, prior edits, tool permissions, and project rules. That context determines whether the agent should read code, run a command, inspect an image, or call a multimodal capability through AnyCap.
Early Access
AnyCap is currently in early access. Capabilities shown on this page are available to early access users. Request access on GitHub to get started.
What counts as workspace context
Project files
Source files, docs, configs, and generated artifacts tell the agent what kind of task it is in.
Execution state
Open terminals, running processes, and prior commands change what is safe and useful to do next.
Tool affordances
Available tools and permissions shape whether the agent can read, edit, browse, or invoke capabilities.
Task history
Previous decisions and feedback help the agent avoid repetition and pick a better next action.
Why this matters for capability routing
If the workspace shows a screenshot, a design review note, and an existing image asset, the agent should probably call an image read or editing capability. If the task history shows the team is preparing a product launch, an image generation or video generation capability may be the right next step.
AnyCap helps reduce routing complexity because multiple multimodal actions are exposed through one consistent runtime instead of many unrelated providers.