Analysis: Factory object collection timing - bootstrap vs per-request

[Copilot]

Analyzed factory.py to determine optimal timing for plugin object collection (LLMs, agents, auth handlers, MCP servers).

Current Implementation (Correct ✅)

Objects are collected at app bootstrap via on_mad_hatter_refresh() callback:

# cheshire_cat.py
async def on_mad_hatter_refresh(self):
    await self.factory.load_objects(self)  # Bootstrap + plugin refresh only
    self.llms = self.factory.get_objects("llm")
    self.agents = self.factory.get_objects("agent")
    # ...

Per-request access is lightweight:

# stray_cat.py (per-request instance)
@property
def _llm(self):
    return self._ccat.llms[slug]  # Direct dict lookup, no hook execution

Recommendation: Keep at Bootstrap

Performance: Avoids executing mad_hatter.execute_hook() (iterates all plugins) on every request

Architecture:

• CheshireCat = app singleton, Factory objects = stateless/shared
• StrayCat = per-request, references shared objects via _ccat

Refresh triggers: App start, plugin toggle, hot reload (infrequent vs requests)

No Changes Required

Current implementation follows FastAPI best practices for plugin-based apps.

Original prompt

check the contents of factory.py and tell me:

• should I run the collection of objects per request or at app bootstrap?

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.