Always-On Tools#

Enkaidu includes a small number of tools that are always available. They are intended to provide the models with dynamic control over how agentic work is done.

  1. For complex tasks, you can ask a model to use an agent to reduce the impact on the context window, especially when using small models.
  2. A model can also query and install tools without you having to pre-load them all the team. Caveat: tool calling models really want to call tools and the idea of installing before using a tool is not well represented in many models’ latent spaces.

Each tool is described below, including its purpose, parameters, and usage notes.

Portions generated byAI

install_tools#

Install one or more tools from the list of installable tools. Tool names are unique across toolsets.

Parameters#

Parameter Type Required Description
tools List of strings Yes List of one or more tools to install so they are available to call afterwards

Notes#

  • The user must confirm the installation before the tools are loaded.
  • Installation is performed across toolsets, and the function returns a JSON payload indicating success or partial success.

list_installable_tools#

Obtain a list of available tools that you can install. Use the list to determine the tools that will help with your task. IMPORTANT: YOU MUST install a tool from this list BEFORE you can call it.

Parameters#

Parameter Type Required Description
(none)

Notes#

  • The response includes the catalog of installable tools and a reminder about installing them prior to use.

spawn_agent#

Run a prompt in a completely fresh, isolated context window and receive the result as [query, response]. Ideal for consuming large amounts of tokens, handling self‑contained tasks, or avoiding context pollution.

Parameters#

Parameter Type Required Description
prompt String Yes The full instruction for the sub‑agent; be explicit, especially when not including history.
include_caller_history Boolean No Set to true only when the task requires awareness of the current session (e.g., summarizing, continuing a thread). Defaults to false.

Notes#

  • The tool spawns a temporary session, executes the prompt, and returns the query‑response pair. It is useful for multi‑step research or large codebase analysis.