Context Development

A context is the primary unit of customization in Beaker. It defines what subkernel runs, what tools the agent has access to, what custom messages flow between the front-end and the kernel, and what previews the notebook renders. The pages in this section cover authoring each of those pieces.

If you have not yet read the conceptual overview, see Key Concepts > Contexts first.

Setting a context from a front-end

External applications (such as a custom front-end) set a context by sending a context_setup_request message to the Beaker kernel:

{
  "subkernel": "<subkernel slug>",
  "context": "<context slug>",
  "context_info": {"<json payload of any settings/info required to start the context>"}
}

The list of available subkernels and contexts depends on what has been installed. The contents of context_info are defined by the context itself. The Beaker service exposes a discovery endpoint at http://{jupyter_url}/beaker/contexts/ that returns the installed contexts and which subkernels each supports.

Tool toggling

Tools on the agent can be toggled individually using either an environment variable or a class attribute. To toggle a tool, create an attribute on your context class or set a variable in your environment named TOOL_ENABLED_{YOUR_TOOL_NAME_IN_UPPER_CASE}. True enables the tool; False disables it. If both are set, the class attribute wins. For example, the environment variable TOOL_ENABLED_ASK_USER=false is overridden by:

class FooContext(BaseContext):
    TOOL_ENABLED_ASK_USER = True
    ...

If no toggle is set for a tool, it defaults to enabled.