跪拜 Guibai
← Back to the summary

Skip the Proxy: Volcengine's IGA Pages Handles LLM Keys and Deployment in One Step

Title: Stop Configuring Your Own Key and Setting Up Proxies for Large Models: Volcengine IGA Pages Handles It in One Step

To connect a large model to a web application, you usually need to prepare a server-side calling chain.

The model API Key cannot be placed in the frontend code, otherwise it will be exposed with client resources. To make requests from the server side, developers also have to apply for and store the API Key, configure the model service address, write forwarding logic, and then handle cross-origin issues between the frontend and the proxy.

This proxy also needs to be deployed and maintained separately. Apart from writing the page and business logic, developers also have to deal with the model service, the server runtime environment, and the hosting platform.

IGA Pages is a web application hosting and deployment capability under Volcengine's full-site acceleration product system. IGA Pages now integrates the AI acceleration gateway, connecting model access, server-side calls, and application deployment.

When you enable the AI acceleration gateway while creating a Pages project, the system automatically creates a dedicated gateway instance for your account. The model service address and API Key are written into the project configuration, and the application code can directly read them to call the model. No need to apply for your own Key, set up a proxy, or handle cross-origin issues.

To let you experience this capability directly, IGA Pages together with the AI acceleration gateway provides a free quota of 500,000 tokens per account, and waives gateway traffic fees and request count fees for the Pages dedicated gateway instance.

Model, Key, and Deployment Configured in One Go

After IGA Pages integrates the AI acceleration gateway, developers can configure models, make server-side calls, and deploy applications in a Pages project:

See the Effect First

We have made a demo template "AI Reasoning Task · Silent Route" to show what kind of applications IGA Pages + AI acceleration gateway can produce.

This is a complete interactive reasoning story: you play the executor, and the AI acts as the story host, responsible for opening narration, action feedback, character dialogue, and ending recap. The story has multiple branches, resource constraints, and multiple endings. Rule determination is executed by a deterministic engine, and the AI only handles narrative expression.

This template is already connected to a model and can be run and deployed directly.

Task main interface displays location, characters, resources, state conditions, and executable actions; the rule engine constrains gameplay.

AI host generates the prologue and scene narration, and the player reads clues before proceeding.

When the player obtains new evidence, the AI host adds narrative feedback; task determination and optional actions remain controlled by the rule engine.

The inference desk aggregates multiple types of evidence, and the player selects an interpretation and execution path before submitting for validation.

The ending page generates a recap based on the player's actual actions, listing completion conditions and the full action trail.

Hands-on Experience

Before you start, make sure the following environment is ready:

Build from a Template to Experience a Complete AI Application

Open TRAE, go to Plugin Market > Skills > Development Tools, find IGA One-Click Deployment, and click Use.

In TRAE's plugin market, select "IGA One-Click Deployment" under the "Skills" page.

Tell TRAE:

Use the IGA one-click deployment skill to create a project named my-mystery-story from the "AI Reasoning Task · Silent Route" template. Complete the IGA Pages cloud project creation and association, connect the AI acceleration gateway, and sync local and remote model configurations.

TRAE summarizes the automatic setup results: template code, Pages cloud project, AI acceleration gateway, and local/remote model config all complete.

TRAE will use the IGA one-click deployment skill to complete the following operations in sequence:

  1. Pull the complete project code from the template to the local machine.

  2. Create and associate a remote project in IGA Pages.

IGA Pages project details page shows the project created and displays the account-shared free model and 500,000 token quota.

  1. Create or reuse the account's Pages-specific AI acceleration gateway instance.

Pages-specific AI acceleration gateway instance created, providing an OpenAI-compatible Base URL and server-side API Key.

  1. Write the model service address and API Key into local and remote environment configurations.

After that, let TRAE start local preview:

Use the IGA one-click deployment skill to start Pages local preview and check the model connection.

TRAE starts Pages local preview and completes a real model call check; the right-side application runs in the browser.

Open the local address returned in the terminal, and you can try the complete AI reasoning story.

Adapt the Story into Your Own Work

You can directly adapt this template using natural language:

Change the story to "Evacuating Paintings in a Rainy Night Art Museum". I am the night shift restorer, and I need to find out why the paintings are being transferred and safely transport them. Only change the story and theme, keep the gameplay core unchanged.

Through natural language, the original story is adapted to "Evacuating Paintings in a Rainy Night Art Museum" while retaining the original gameplay, model integration, and rule engine.

TRAE will first read the AGENTS.md file in the project, then use the built-in story creation skill to modify story data and page copy while reusing the existing rule engine. After changes, refresh the local page to try it out and check whether the story content and flow meet expectations.

Deploy and Share with Others

After confirming the new story experience is satisfactory, let TRAE complete the check and deployment:

Run story validation and production build, deploy to IGA Pages after checks pass, and open the returned preview address to confirm the application is accessible.

TRAE completes story validation, production build, and IGA Pages deployment, then opens the online preview address to confirm accessibility.

TRAE first runs story structure validation and production build, then publishes the application to IGA Pages. After deployment, it returns a preview address. Share the link, and others can directly play your story.

For long-term sharing, bind a custom domain in the IGA Pages console. If you later want to try a different model, switch in the AI acceleration gateway console without changing the application code.

Custom domains in the mainland China acceleration region require ICP filing.

Start from this template and build an AI application that can go live.

Learn More: