LLM Connections
Selects the LLM provider and model behind --llm-format-model and --navigator-type, as a connection string:
protocol://[key@][host][/base-path]?model=NAME[&options]| Part | Value |
|---|---|
protocol | The provider, from its page below |
key | The provider’s API key. Omitted, ZShot reads the provider’s key variable, and the first one set wins |
host | An endpoint in place of the provider’s own, for a self-hosted or proxied API |
base-path | The endpoint’s path prefix, for a gateway whose route carries one. Omitted, ZShot uses the provider’s own |
model | The model. Omitted, ZShot runs the provider’s default model |
options | Connection string options |
A bare protocol name such as anthropic runs the provider’s default model with the key from its variable. A key or model written in the string overrides both. There is no default provider; setting only one of the two flags supplies the other.
| Provider | Description |
|---|---|
| Anthropic | Anthropic's Claude models |
| OpenAI | OpenAI's GPT models |
| Google's Gemini models | |
| Ollama | Models on your own Ollama server |
| Mistral | Mistral AI's models |
| DeepSeek | DeepSeek's models |
| Mercury | Inception Labs' Mercury models |
| Groq | Open models hosted by Groq |
| HuggingFace | Models hosted through HuggingFace |
| TypeSafe | TypeSafe's decision models, for navigation only |
| Cloudflare | Models on Cloudflare Workers AI |
| OpenRouter | Many makers' models through one API |
| stdio | An external agent, or a person at a terminal |
For a local model, run Ollama and use ollama.
Example
Through a Cloudflare AI Gateway, whose route is the base path, with the gateway’s token in a header:
CLI
Run locally
zshot -t llm_json \
--llm-format-model "openai://$OPENAI_API_KEY@gateway.ai.cloudflare.com/v1/$ACCOUNT/$GATEWAY/openai?model=gpt-5.6&header=cf-aig-authorization:Bearer%20$AIG_TOKEN" \
--llm-format-prompt "Summarize this page." \
https://zshot-cli.com/example_assets/gazette/ -f out.json