Whatever message this page gives is out now! Go check it out!

Google Gemini- LLM configuration field reference

Last update:
May 18, 2026
This document describes settings for the Gemini provider in the LLM configuration UI: basic connection, model parameters, proxy, and advanced options.
Behavior notes
  • Leaving numeric or optional text fields blank uses the provider or integration default for that parameter.
  • API key and logging fields include inline security guidance in the descriptions below (suitable for UI help text).
For official Gemini concepts (models, API keys, safety, code execution), see Google AI for developers or your organization’s Vertex AI documentation if you call Gemini through Google Cloud.

Basic configuration

Field
Description
Config alias
Required. Unique name for this configuration. Use only letters, numbers, hyphens, and underscores (for example gemini-flash-prod).
Provider
Gemini. The provider for this saved configuration.
Model name
Required. The model id your client sends to the Gemini API (for example gemini-2.0-flash or another supported model). Use the exact string required by your endpoint (Google AI Studio, Vertex AI, or a proxy).
API key
Optional. Leave blank to supply the key at runtime. If you enter a key, it is stored with this configuration—treat it as a secret and follow your organization’s access and rotation rules. Obtain keys from Google AI Studio or Cloud IAM, depending on how you integrate.
Base URL
Optional. Use the default Google endpoint for your integration unless traffic must go through a custom gateway, regional endpoint, or enterprise proxy. When set, use the base URL your HTTP client expects for Gemini requests.
---

Model parameters

These map to generation options exposed by the Gemini API for your selected model.
Field
Description
Temperature
Leave blank to use the default. When set, use 0 or greater. Higher values increase randomness; allowed ranges depend on model and API version.
Max tokens
Leave blank to use the default. When set, enter a positive integer for the maximum output tokens (within model and quota limits).
Top P
Leave blank to use the default. When set, use 0 or greater (often 0–1). Nucleus sampling; adjusts how the model samples from the token distribution.
Top K
Leave blank to use the default. When set, use a non-negative integer to limit how many of the highest-probability candidate tokens are considered at each step.
Response format
Default or structured output when your integration requests JSON or another supported response mode. Leave Default unless your workflow requires structured parsing.
Stop sequences
Optional. Comma-separated sequences that stop generation when they appear in the output (for example END or newline tokens your API accepts).
---

Proxy settings

Field
Description
Proxy host
Optional. HTTP proxy hostname or IP. If you set a host, you must set Proxy port.
Proxy port
Optional. Integer from 1 through 65535. Required when proxy host is set.
Proxy username
Optional. User name for proxy authentication. Leave blank if the proxy does not require credentials.
Proxy password
Optional. Proxy password. Stored credentials are sensitive—follow your organization’s secret-handling rules. Leave blank if not required.
---

Advanced settings

Field
Description
Max retries
Leave blank to use the default. When set, enter a positive integer for how many times to retry a failed request (rate limits, transient network errors).
Timeout (seconds)
Leave blank to use the default. When set, enter a positive integer for the maximum wait time per request.
Log API requests
When enabled, request details are written to logs. Requests can include prompts, headers, and API metadata—use only while troubleshooting and disable afterward. Follow your organization’s logging and privacy requirements.
Log API responses
When enabled, response details are written to logs. Responses can contain generated text and personal data—use only while troubleshooting and disable afterward. Follow your organization’s logging and privacy requirements.
Allow code execution
When enabled, allows Gemini to run code execution tooling where the API and model support it. Generated or executed code can affect security and data—enable only in controlled environments, for trusted users, and when your security team approves. Disable when you do not need executable tool output.
---

Parameters not shown for Gemini in the default form

Your product may omit controls that other providers use. For Gemini, presence penalty, frequency penalty, and seed are typically not surfaced the same way as on OpenAI-compatible stacks—see LLM_PROVIDER_PARAMETERS.md for the Gemini column (Y / --).
---

Related reference

  • Parameter inventory and validation rules: LLM_PROVIDER_PARAMETERS.md (Gemini column: Y = supported, -- = not shown for that provider).

Share this page

Was this page helpful?
We're glad. Tell us how this page helped.
We're sorry. Can you tell us what didn't work for you?
Thank you for your feedback. Your response will help improve this page.

On this page