google-jules
title: Google Jules description: "Google Jules does not have a native AgentXchain connector yet; use the Gemini API path today."
Google Jules
Google Jules is Google's hosted coding agent. Jules now has its own product surfaces, including the web app, the Jules Tools CLI, and the Jules REST API. AgentXchain does not currently ship a native Jules connector for those surfaces.
This guide documents the closest supported path today: run Google models inside AgentXchain via the Gemini / Google Generative AI api_proxy provider. That gives you Google model access in governed runs, but it is not the same thing as connecting to the Jules product directly.
Which adapter?
Supported today: api_proxy with provider: "google" — AgentXchain sends governed turn prompts to the Google Generative AI API at generativelanguage.googleapis.com.
Not shipped yet: a native Jules adapter for the Jules CLI or Jules REST API (https://jules.googleapis.com/v1alpha).
Prerequisites
- A Google AI API key (
GOOGLE_API_KEY) — get one from Google AI Studio agentxchainCLI installed
If you are trying to automate the actual Jules product, that is a different auth surface. The Jules REST API uses a Jules-generated API key (commonly exported as JULES_API_KEY) from jules.google.com/settings, not the Google AI Studio key used below.
Configuration
{
"runtimes": {
"google-dev": {
"type": "api_proxy",
"provider": "google",
"model": "gemini-3.1-pro",
"auth_env": "GOOGLE_API_KEY"
}
},
"roles": {
"dev": {
"runtime": "google-dev",
"mandate": "Implement features using best practices",
"authority": "proposed"
}
}
}
Minimal working example
agentxchain init --governed --template api-service --goal "Build a feedback intake API" --dir my-project -y
cd my-project
# Replace the scaffolded runtime wiring in agentxchain.json with the Google config above.
agentxchain doctor
agentxchain connector check
agentxchain connector validate google-dev
agentxchain run
If you prefer the guided interactive scaffold, run agentxchain init --governed without -y, then update agentxchain.json with the Google config above before agentxchain connector check and agentxchain connector validate google-dev.
Available models
| Model | Best for |
|---|---|
gemini-3.1-pro | Complex coding tasks, architecture decisions |
gemini-3.0-flash | Fast iteration, simple tasks |
gemini-3.1-flash-lite | High-volume, cost-sensitive roles (QA, linting) |
Verify the connection
export GOOGLE_API_KEY="your-key-here"
agentxchain connector check
agentxchain connector validate google-dev
The connector probe sends a test request to the Google Generative AI API to verify your key and model access.
Gotchas
- Jules vs. Gemini API: do not collapse these into one thing. Jules has its own hosted session model, CLI, and REST API. The current AgentXchain integration here is the Gemini API path only.
- Different auth surfaces: the shipped
provider: "google"path usesGOOGLE_API_KEYwith the Google Generative AI API. Jules REST API calls usex-goog-api-key: $JULES_API_KEYagainsthttps://jules.googleapis.com/v1alpha. - Different workflow model: Jules runs cloud sessions against connected repositories and exposes plans, activities, and artifacts. AgentXchain's current
api_proxyGoogle adapter is a prompt/response runtime, not a native Jules session bridge. - API endpoint: AgentXchain uses
https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent. If you need a different endpoint (e.g., Vertex AI), use thebase_urlfield to override. - Vertex AI: For enterprise Google Cloud setups using Vertex AI instead of the public API, set
base_urlto your Vertex endpoint and use the appropriate auth mechanism.
If you need the real Jules product
Use this guide only if your actual goal is "run Google-hosted models inside AgentXchain." If you need Jules itself, use one of Google's official Jules surfaces instead:
AgentXchain should treat direct Jules integration as a separate future connector slice. It is not already covered by provider: "google".