---
title: "通过 RunAPI 在爱马仕 (Hermes Agent) 中使用 Gemini — 大模型API 指南"
url: "https://runapi.ai/zh-CN/hermes-gemini.md"
canonical: "https://runapi.ai/zh-CN/hermes-gemini"
locale: "zh-CN"
model: "gemini"
---

# 在 Hermes Agent 中使用 Gemini。

Google Gemini 可通过 RunAPI 的 OpenAI 兼容端点调用。Hermes Agent 使用 custom:runapi provider 调用它 — Gemini 3.5 Flash 用于对速度敏感的 agent 循环，3.x Pro 用于多步推理，2.5 Pro 用于长上下文生产任务。无需 Google Cloud 项目或 Vertex AI 凭据 — 只需你已为聊天配置的同一个 RUNAPI_API_KEY 和 base_url。

## API example

```bash
curl -X POST https://runapi.ai/v1/chat/completions \
  -H "Authorization: Bearer $RUNAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.5-flash",
    "messages": [
      {"role": "system", "content": "You are a concise technical assistant."},
      {"role": "user", "content": "Explain the difference between gRPC and REST in three sentences."}
    ],
    "temperature": 0.7,
    "max_tokens": 256
  }'

```

### Response

```json
{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "model": "gemini-3.5-flash",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "gRPC uses HTTP/2 and Protocol Buffers for strongly-typed, multiplexed RPC calls with built-in code generation. REST uses HTTP/1.1 (or 2) with JSON payloads and relies on URL paths and HTTP verbs for resource semantics. gRPC is faster for service-to-service calls; REST is simpler to debug and more widely supported by browsers."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 34,
    "completion_tokens": 71,
    "total_tokens": 105
  }
}

```

## How it works

1. **Configure RunAPI** — Set the RUNAPI_API_KEY environment variable. If you already added RunAPI as a custom:runapi provider in Hermes Agent, the same key and base_url work for Gemini — change only the model ID. No Google Cloud credentials needed.
2. **Call Gemini via chat completions** — Send a POST request to /v1/chat/completions with model set to gemini-3.5-flash. Pass a messages array with system and user roles. Hermes Agent sends the same OpenAI-compatible request shape it uses for GPT — RunAPI routes to Gemini based on the model parameter.
3. **Read the response** — The response arrives synchronously in OpenAI chat completion format. The assistant reply is in choices[0].message.content, with token usage in the usage object. For streaming, set stream to true and Hermes Agent parses the SSE delta events automatically.

## Parameters

| Parameter | Type | Description |
|-----------|------|-------------|
| `model` | `string` | Required. gemini-3.5-flash, gemini-2.5-flash, gemini-2.5-pro, gemini-3-flash-preview, gemini-3-pro-preview, or gemini-3.1-pro-preview. |
| `messages` | `array` | Required. Array of message objects with role (system, user, assistant) and content fields. |
| `temperature` | `number` | Optional. Sampling temperature between 0 and 2. Lower values produce more deterministic output. Default varies by model. |
| `max_tokens` | `integer` | Optional. Maximum number of tokens to generate in the response. |
| `stream` | `boolean` | Optional. When true, the response streams as server-sent events. Each event contains a delta with partial content. |
| `top_p` | `number` | Optional. Nucleus sampling threshold between 0 and 1. Alternative to temperature for controlling output randomness. |

## FAQ

### Can I use Google Gemini in Hermes Agent without Google Cloud credentials?

Yes. RunAPI provides Gemini through its OpenAI-compatible endpoint. Configure RunAPI as a custom:runapi provider with base_url https://runapi.ai/v1 and key_env RUNAPI_API_KEY. No Google Cloud project, service account, or Vertex AI setup required.

### What is the difference between Gemini Flash vs Pro -- when should I use each?

Flash (gemini-3.5-flash) is fastest and cheapest -- best for real-time agent loops, classification, and tool-calling chains. Pro (gemini-2.5-pro) handles complex reasoning, long-context analysis, and multi-step tasks. Use Flash for speed, Pro for depth.

### How do I use context caching to reduce costs with long documents?

When sending the same large context across multiple requests (like a codebase or document set), Gemini's context caching reduces input token costs on subsequent calls. This is especially useful in agent loops where the system prompt and reference material stay the same across many turns.

### Can Hermes Agent switch between Gemini and other LLMs mid-session?

Yes. All RunAPI LLMs share the same custom:runapi provider and API key. Use the /model command or hermes model to switch between gemini-3.5-flash, gpt-5.5, claude-opus-4.6, or any other RunAPI model without changing provider config.

### Does Gemini through RunAPI support function calling and tool use?

Yes. RunAPI passes the OpenAI-compatible tools and tool_choice parameters to Gemini. Define tools in the request body and Gemini returns tool_calls in the assistant message. Hermes Agent processes these the same way it handles tool calls from GPT or Claude.

### Can Hermes Agent mix Gemini with other LLM providers in one workflow?

Yes. Hermes Agent can call Gemini Flash for cheap preprocessing, GPT-5.5 for complex reasoning, and Claude for long-context analysis, all through the same RunAPI key and custom:runapi provider.


## Links

- [Hermes Agent 配置指南 →](https://runapi.ai/zh-CN/hermes-agent)
- [Gemini 模型 →](https://runapi.ai/zh-CN/models/gemini)
- [Model catalog](https://runapi.ai/zh-CN/models)
- [API docs](https://runapi.ai/zh-CN/docs)