mirror of
https://github.com/hanzoai/chat.git
synced 2026-09-30 20:51:17 +00:00
zeekayandHanzo Dev
5c089743fe
A conversation can ask for fast mode: the gateway offers the completion to several providers at once and keeps whichever answers first. Declared beside web_search, so it appears in the OpenAI settings the same way and needs no new control of its own. It rides in modelKwargs rather than on llmConfig, which is the whole reason this is more than a schema line. The client's constructor is typed, so an unrecognised key set on llmConfig is dropped without a word — the switch would read as on, the request would go out unchanged, and nothing on the screen or in a log would say so. verbosity already had to solve this; fast follows it. Only true is sent. false is the absence of the ask, not a request to be slow, and putting it on every ordinary request would be a field with no reader. The description says what it costs, because it does cost: the question is asked more than once and the account is charged for every attempt. Co-authored-by: Hanzo Dev <[email protected]>