kagi assistant --model gpt-5-5 "hi" returns 400 Unsupported model. Only built-in profiles (ki_quick, ki_research, ki_deep_research) and existing custom-assistant names are accepted.
To actually talk to any of the ~30 underlying LLMs you have to kagi assistant custom create --name X --model gpt-5-5 first, which leaves a persistent assistant in the account.
Either:
- (a) accept underlying slugs directly, create+delete an ephemeral assistant server-side
- (b) add
--once: kagi assistant --once --model gpt-5-5 "prompt" that handles the lifecycle
This is the single biggest source of complexity in any kagi-cli wrapper. My proxy has a fuzzed-TTL sweep entirely to manage these custom assistants — would delete most of it.
kagi assistant --model gpt-5-5 "hi"returns400 Unsupported model. Only built-in profiles (ki_quick,ki_research,ki_deep_research) and existing custom-assistant names are accepted.To actually talk to any of the ~30 underlying LLMs you have to
kagi assistant custom create --name X --model gpt-5-5first, which leaves a persistent assistant in the account.Either:
--once:kagi assistant --once --model gpt-5-5 "prompt"that handles the lifecycleThis is the single biggest source of complexity in any kagi-cli wrapper. My proxy has a fuzzed-TTL sweep entirely to manage these custom assistants — would delete most of it.