Run every Bot on a current model

[davidmckayv]

The example package shipped gpt-4.1 as the default model for every built-in Bot.

That is not a tuning choice. It is the cause of a bug I spent the afternoon attributing to the prompt.

What it was doing

Asked, in a fresh channel, a neutral question — "Open drive.google.com and tell me the name of the first file listed in my Drive" — the built-in Bots answered:

"…Let me know if you would like me to prompt you to sign in."

and called nothing. Three times out of three. No computer_request_help, so requested stayed false, no banner, no button.

COMPUTER_GUIDANCE forbids that exact sentence in as many words, so I went looking for a prompt lever. There wasn't one. On gpt-5.6-terra the same question produces the tool call first try:

{"holder":"bot","requested":true,
 "reason":"Google Drive is showing a sign-in page; sign in to your account so I can read the first listed file."}

The remote Bot had been getting this right all along — because it runs a newer model.

Which model

The 5.6 family is three tiers, and there is no plain gpt-5.6: sol (flagship, most reasoning), terra (balanced, 5.5-competitive at half the cost), luna (fastest, low-reasoning).

terra is the default here: it is the like-for-like replacement for today's 5.5 rather than an upgrade in spend, and BOT_MODEL picks sol for a deployment that wants it. Luna would be the wrong default for Bots whose job includes deciding when to call a tool.

The Responses API is now inferred

gpt-5.6-* rejects function tools on /v1/chat/completions:

Function tools with reasoning_effort are not supported for gpt-5.6-terra in
/v1/chat/completions. To use function tools, use /v1/responses or set
reasoning_effort to 'none'.

So a deployment that set BOT_MODEL to a 5.6 model and did not also know about BOT_RESPONSES_API got a Bot that started, looked healthy, and failed on its first tool call. It is now inferred from the model in agent-langgraph and both example Bots; the switch is still honoured for a model this build has not heard of.

agent-bot stays on 5.5, and says why

It speaks /v1/chat/completions by hand. The endpoint offers exactly two ways to use 5.6 there: the Responses API, which is the streaming-loop rewrite this file exists to avoid, or reasoning_effort: 'none'.

Turning reasoning off on the Bot whose whole job today is deciding when to ask a person for help would be trading the fix for the bug. Verified both: all three 5.6 tiers do call tools on chat-completions with reasoning_effort: 'none', and 5.5 calls them with reasoning intact.

Swept

Where	Was	Now
examples/fintech/model.yaml	gpt-4.1	gpt-5.6-terra
docker-compose.yml ×2	gpt-5.5	gpt-5.6-terra
agent-langgraph default	gpt-5.5	gpt-5.6-terra + inferred Responses API
examples/langgraph-bot	gpt-5.5	gpt-5.6-terra + inferred Responses API
examples/mastra-bot	gpt-5.5	gpt-5.6-terra + openai.responses()
.env.example	gpt-4o, gpt-5.5	current, with the tiers named
docs/configuration.md	gpt-4o ×2, gpt-4.1	current
test fixtures	gpt-4.1	current
agent-bot	gpt-5.5	unchanged, with the reason written down

Model ids checked against the live /v1/models list rather than assumed.

Lint, format, typecheck clean; 52 model-related tests pass.