Agent.ini is not limited to one endpoint. Declare as many [Agent.{Name}] sections as you like and combine them across three independent roles:

• DialogAgent — what NPCs say when you talk to them (and in NPC-to-NPC replies). Your quality tier.
• BarkAgent — ambient idle comments. A cheap, fast model is plenty here.
• MemoryAgent — background memory summarization.

Each role takes a single agent or a pipe-separated chain. On any failure Numen advances to the next agent in the chain automatically, so you can put a cloud model first with a local one as a safety net (or the other way around).

Free-tier load balancing
Several providers (Google's Gemini API among them) offer a free tier capped at so many requests per minute and per day. A single key runs dry quickly once your companions get chatty — so fetch API keys from different providers, give each key its own [Agent.] section, chain them, and switch on random ordering. Numen then starts each request at a random agent in the chain and spreads the load evenly across them.

[General]
DialogAgent=Free1 | Free2
BarkAgent=Free1 | Free2
MemoryAgent=Free1 | Free2
bRandomAgentOrder=1

[Agent.Free1]
sURL=https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
sModel=gemma-4-12b-it
sFormat=openai
sReasoningEffort=none
sAPIKey=key-from-account-1

[Agent.Free2]
sURL=https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
sModel=gemma-4-12b-it
sFormat=openai
sReasoningEffort=none
sAPIKey=key-from-account-2

Add as many [Agent.FreeN] sections as you have keys.
The default Agent.ini demonstrates how to use this feature with their free-tier Gemini and Gemma models.

Keep in mind - some providers' terms restrict using multiple free accounts to get around limits.