Gemini AI

Category: AI & Agents

Get this pack →

This page is generated from the Air Pipe marketplace. Browse it live to install into your organization.

Query Google's Gemini models over a simple HTTP API. Three variants: return the full Gemini payload, extract just the answer text, or constrain the answer with an instruction prefix. The API key and model are managed variables — swap models without touching the config or redeploying.

---

What's included

File	Purpose
config.yml	Air Pipe config with docs: true

---

Endpoints

Method	Path	Description
POST	/gemini/query	Send a prompt, receive Gemini's full response payload
POST	/gemini/query-text	Send a prompt, receive only the answer text
POST	/gemini/query-prefixed	Constrain the answer with a fixed instruction prefix

---

Setup

1. Get a Google AI Studio API key

Create a key at https://aistudio.google.com/app/apikey.

2. Set managed variables

Variable	Example	Notes
GEMINI_API_KEY	AIza…	Your AI Studio API key (store as a secret)
GEMINI_MODEL	gemini-2.5-flash	Any Gemini model that supports generateContent

Free-tier note: available models vary by key. A 429 with limit: 0 means that model isn't on your key's free tier (e.g. gemini-2.0-flash often isn't) — switch GEMINI_MODEL to gemini-2.5-flash / gemini-flash-latest, or enable billing.

3. Deploy the pack and call it

curl -X POST https://<your-airpipe-host>/gemini/query-text \
  -H 'content-type: application/json' \
  -d '{"text": "Tell me about dogs"}'

Response (/gemini/query-text):

{ "data": { "QueryGemini": { "data": "Dogs are loyal, intelligent companions..." } } }

Use /gemini/query instead if you need the full Gemini payload (candidates, safety ratings, token usage), or /gemini/query-prefixed to force short answers.

---

How it works

Each endpoint validates the incoming text, then calls the Gemini generateContent API with the model and key from managed variables. The query-text and query-prefixed variants add a post_transforms step that pulls candidates[0].content.parts[0].text out of the response with jq.

Configuration

config.yml

name: GeminiAI
description: >
  Query Google's Gemini models over a simple HTTP API. Three variants: return
  the full Gemini payload, extract just the answer text, or constrain the answer
  with a system-style prefix. The API key and model are managed variables — swap
  models without redeploying.

docs: true

# Required managed variables:
#   GEMINI_API_KEY  — Google AI Studio API key (https://aistudio.google.com/app/apikey)
#   GEMINI_MODEL    — Model name, e.g. gemini-2.5-flash

interfaces:

  # POST /gemini/query
  # Validate the prompt, call Gemini, return the full response payload.
  gemini/query:
    output: http
    method: POST
    summary: Query Gemini (full payload)
    description: Send a prompt and receive Gemini's complete response.
    tags: [gemini, ai]
    request_example:
      text: "Tell me about dogs"

    actions:
      - name: CheckBody
        input: a|body|
        hide_data_on_success: true
        assert:
          http_code_on_error: 400
          tests:
            - value: text
              is_not_null: true
              is_not_empty: true
              description: "The prompt to send to the model"

      - name: QueryGemini
        run_when_succeeded: [CheckBody]
        http:
          url: https://generativelanguage.googleapis.com/v1beta/models/a|ap_var::GEMINI_MODEL|:generateContent?key=a|ap_var::GEMINI_API_KEY|
          method: POST
          headers:
            content-type: application/json
          body: |
            {"contents":[{"parts":[{"text":"a|CheckBody::text->json_escape|"}]}]}

  # POST /gemini/query-text
  # Same as above but extracts only the answer text from the payload.
  gemini/query-text:
    output: http
    method: POST
    summary: Query Gemini (answer text only)
    description: Send a prompt and receive only the model's answer text.
    tags: [gemini, ai]
    request_example:
      text: "Tell me about dogs"

    actions:
      - name: CheckBody
        input: a|body|
        hide_data_on_success: true
        assert:
          http_code_on_error: 400
          tests:
            - value: text
              is_not_null: true
              is_not_empty: true

      - name: QueryGemini
        run_when_succeeded: [CheckBody]
        http:
          url: https://generativelanguage.googleapis.com/v1beta/models/a|ap_var::GEMINI_MODEL|:generateContent?key=a|ap_var::GEMINI_API_KEY|
          method: POST
          headers:
            content-type: application/json
          body: |
            {"contents":[{"parts":[{"text":"a|CheckBody::text->json_escape|"}]}]}
        post_transforms:
          - extract_with_jq: ".body.candidates[0].content.parts[0].text"

  # POST /gemini/query-prefixed
  # Constrain the answer by prepending an instruction to the prompt.
  gemini/query-prefixed:
    output: http
    method: POST
    summary: Query Gemini with an instruction prefix
    description: >
      Prepend a fixed instruction to every prompt to constrain the response
      (e.g. "answer in one short sentence").
    tags: [gemini, ai]
    request_example:
      text: "Tell me about dogs"

    actions:
      - name: CheckBody
        input: a|body|
        hide_data_on_success: true
        assert:
          http_code_on_error: 400
          tests:
            - value: text
              is_not_null: true
              is_not_empty: true

      - name: QueryGemini
        run_when_succeeded: [CheckBody]
        http:
          url: https://generativelanguage.googleapis.com/v1beta/models/a|ap_var::GEMINI_MODEL|:generateContent?key=a|ap_var::GEMINI_API_KEY|
          method: POST
          headers:
            content-type: application/json
          body: |
            {"contents":[{"parts":[{"text":"Answer in one short sentence: a|CheckBody::text->json_escape|"}]}]}
        post_transforms:
          - extract_with_jq: ".body.candidates[0].content.parts[0].text"