Cloudflare AI Gateway exposes an OpenAI-compatible unified API, so you reach multiple vendors through a single endpoint by setting the model id to a vendor/model route. The default vendor is Workers AI, which needs only a Cloudflare token and account id.
The Cloudflare class defaults to @cf/meta/llama-3.3-70b-instruct-fp8-fast. Agno rewrites any @cf/... id to workers-ai/@cf/... for the gateway.
Authentication
Cloudflare needs an API token and your account id. The AI Gateway id is optional and falls back to default.
| Variable | Required | Description |
|---|
CLOUDFLARE_API_TOKEN | Yes | API token. Create one at dash.cloudflare.com. |
CLOUDFLARE_ACCOUNT_ID | Yes | Your account id, found in the Cloudflare dashboard. |
CLOUDFLARE_AI_GATEWAY_ID | No | AI Gateway id. Defaults to default. |
export CLOUDFLARE_API_TOKEN=***
export CLOUDFLARE_ACCOUNT_ID=***
export CLOUDFLARE_AI_GATEWAY_ID=*** # optional
Example
Use Cloudflare with your Agent:
from agno.agent import Agent
from agno.models.cloudflare import Cloudflare
agent = Agent(
model=Cloudflare(id="@cf/meta/llama-3.3-70b-instruct-fp8-fast"),
markdown=True,
)
# Print the response in the terminal
agent.print_response("Share a 2 sentence horror story.")
Parameters
| Parameter | Type | Default | Description |
|---|
id | str | "@cf/meta/llama-3.3-70b-instruct-fp8-fast" | The gateway model id. See Model selection. |
name | str | "Cloudflare" | The name of the model |
provider | str | "Cloudflare" | The provider of the model |
api_key | Optional[str] | None | API token (defaults to CLOUDFLARE_API_TOKEN env var) |
account_id | Optional[str] | None | Account id (defaults to CLOUDFLARE_ACCOUNT_ID env var) |
gateway_id | Optional[str] | None | AI Gateway id (defaults to CLOUDFLARE_AI_GATEWAY_ID, or default) |
base_url | Optional[str] | None | Override the gateway URL. Skips env-var lookup when set. |
max_tokens | Optional[int] | None | Maximum tokens to generate |
Cloudflare extends the OpenAI-compatible interface and supports most parameters from the OpenAI model. Use use_json_mode=True for structured output when a model does not support native structured outputs.
Model selection
Pick the route by setting id:
| Vendor | Id format | Notes |
|---|
| Workers AI | @cf/<org>/<model> | Copy the binding id from the model catalog. Agno prepends workers-ai/. |
| OpenAI, Google, etc | openai/<model>, google/<model> | Requires the vendor’s BYOK key stored in the AI Gateway dashboard. |
| Dynamic route | dynamic/<route> | A route configured in the dashboard for fallbacks. |
from agno.agent import Agent
from agno.models.cloudflare import Cloudflare
# Workers AI (Agno rewrites to workers-ai/@cf/...)
agent = Agent(model=Cloudflare(id="@cf/google/gemma-4-26b-a4b-it"), markdown=True)
Workers AI serves Google’s open-weight Gemma models, not the standard Gemini API. Invented ids such as @cf/.../gemini-... return HTTP 400. For Gemini, use a google/... route with BYOK credentials. Vendor routes also need a model id from the Cloudflare docs.
