agent-dispatch

Just-in-time agent dispatch reduces context window bloat by keeping a lightweight registry of specialized subagents and only downloading the specific agent needed to handle a requested task at execution time, minimizing startup overhead.

You route a request to a specialized AI agent by mapping keywords from the user's input to agent names and categories via a TOML index, which facilitates efficient task delegation and automatically downloads the matching agent.

Yes, the JIT agent router is platform agnostic and works with various AI coding platforms that support the SKILL.md format, enabling automated agent discovery, caching, and dispatch across different environments.

The best way to manage a registry of specialized subagents is using a keyword-based router that supports automated agent discovery, caching, and on-demand downloading to efficiently delegate tasks without pre-installing agents.

You should use on-demand agent downloading instead of pre-installing agents when you want to minimize startup overhead and prevent an AI from handling tasks outside its expertise without bloating the main context window.