OpenAI

Protocol: openai://
Default model: gpt-6-astra

Key envOPENAI_API_KEY

LicensePro

Runs OpenAI models on the Chat Completions API at https://api.openai.com/v1, or on the Decisions API with protocol=decisions.

Options

Option	On OpenAI
`protocol`	`openai` to chat, the default, or `decisions`
`tool`	`web_search` only, on a search model such as `gpt-5-search-api`
`max_tokens`	Sent as `max_completion_tokens`
`vision`	On by default for `gpt-4`, `gpt-5`, `gpt-6`, `gpt-image`, `o1`, `o3` and `o4` models
`body.<path>`	Fields OpenAI’s chat request names, such as `body.reasoning_effort=high`
`header`, `temperature`, `top_p`, `frames`, `timeout`, `insecure`	As listed

Decisions

protocol=decisions runs the Decisions API, which chooses among listed alternatives as typesafe does and reads screenshots. It navigates only, its default model is gpt-6-luna, and it takes no chat options. A host serving the same API at {base}/decisions takes it too: openai://KEY@host/v1?protocol=decisions.

Example

A formatter:

CLI Run locally

zshot -t llm_text \
  --llm-format-model "openai://?model=gpt-6-astra&body.reasoning_effort=high" \
  --llm-format-prompt "Summarize the lead story in one sentence." \
  -f summary.txt https://zshot-cli.com/example_assets/gazette/

A Decisions navigator with a chat model to escalate to:

CLI Run locally

zshot --navigate "Open the formats page" \
  --navigator-type "openai://?protocol=decisions&name=scout" \
  --navigator-type "openai://?name=closer" \
  -f out.png https://zshot-cli.com/