integrating-ai-models-into-your-development-workflow.md

title	intro	versions	shortTitle
Integrating AI models into your development workflow	Call AI models in the tools you use every day.	feature github-models	Integrate AI models

With {% data variables.product.prodname_github_models %} extensions, you can call specific AI models from both {% data variables.product.prodname_copilot_chat_short %} and {% data variables.product.prodname_cli %}. These extensions integrate directly into your development workflow, allowing you to prompt models without context switching.

Using AI models in {% data variables.product.prodname_copilot_chat_short %}

If you have a {% data variables.product.prodname_copilot_short %} subscription, you can work with AI models in {% data variables.product.prodname_copilot_chat_short %} in two different ways:

• Using the {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %}. With this extension, you can ask for model recommendations based on certain criteria and chat with specific models. See Using the {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %}.
• Using multiple model support in {% data variables.product.prodname_copilot_chat_short %}. With multi-model {% data variables.product.prodname_copilot_chat_short %}, you can choose a specific model to use for a conversation, then prompt {% data variables.product.prodname_copilot_chat_short %} as usual. See AUTOTITLE.

Using the {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %}

[!NOTE] The {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %} is in {% data variables.release-phases.public_preview %} and is subject to change.

1. Install the {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %}.

   • If you have a {% data variables.product.prodname_copilot_pro_short %} subscription, you can install the extension on your personal account.
   • If you have access to {% data variables.product.prodname_copilot_short %} through a {% data variables.product.prodname_copilot_business_short %} or {% data variables.product.prodname_copilot_enterprise_short %} subscription:
     • An organization owner or enterprise owner needs to enable the {% data variables.product.prodname_copilot_extensions_short %} policy for your organization or enterprise.
     • An organization owner needs to install the extension for your organization.

2. Open any implementation of {% data variables.product.prodname_copilot_chat_short %} that supports {% data variables.product.prodname_copilot_extensions %}. For a list of supported {% data variables.product.prodname_copilot_chat_short %} implementations, see AUTOTITLE.

3. In the chat window, type @models YOUR-PROMPT, then send your prompt. There are several use cases for the {% data variables.product.prodname_github_models %} {% data variables.product.prodname_copilot_extension_short %}, including:

   • Recommending a particular model based on context and criteria you provide. For example, you can ask for a low-cost OpenAI model that supports function calling.
   • Executing prompts using a particular model. This is especially useful when you want to use a model that is not currently available in multi-model {% data variables.product.prodname_copilot_chat_short %}.
   • Listing models currently available through {% data variables.product.prodname_github_models %}

Using AI models with {% data variables.product.prodname_actions %}

You can use the {% data variables.product.prodname_actions %} token (GITHUB_TOKEN) to call AI models directly inside your workflows.

Setting permissions

To use AI models in your workflows, you need to ensure that the models permission is enabled in your workflow configuration. This permission allows workflows to access the {% data variables.product.prodname_github_models %} inference API. You can either set this permission itself or use the general read-all or write-all permissions. See AUTOTITLE.

Writing your workflow file

You can call the inference API directly from your workflow. For instance:

name: Use GitHub Models

on:
  workflow_dispatch:

permissions:
  models: read

jobs:
  call-model:
    runs-on: ubuntu-latest
    steps:
      - name: Call AI model
        env:
          GITHUB_TOKEN: {% raw %}${{ secrets.GITHUB_TOKEN }}{% endraw %}
        run: |
          curl "https://models.github.ai/inference/chat/completions" \
             -H "Content-Type: application/json" \
             -H "Authorization: Bearer $GITHUB_TOKEN" \
             -d '{
              "messages": [
                  {
                     "role": "user",
                     "content": "Explain the concept of recursion."
                  }
               ],
               "model": "openai/gpt-4o"
            }'

Using AI models from the command line

[!NOTE] The {% data variables.product.prodname_github_models %} extension for {% data variables.product.prodname_cli %} is in {% data variables.release-phases.public_preview %} and is subject to change.

You can use the {% data variables.product.prodname_github_models %} extension for {% data variables.product.prodname_cli %} to prompt AI models from the command line, and even pipe in the output of a command as context.

Prerequisites

To use the {% data variables.product.prodname_github_models %} CLI extension, you need to have {% data variables.product.prodname_cli %} installed. {% data reusables.cli.cli-installation %}

Installing the extension

1. If you have not already authenticated to the {% data variables.product.prodname_cli %}, run the following command in your terminal.

   gh auth login

2. To install the {% data variables.product.prodname_github_models %} extension, run the following command.

   gh extension install https://github.com/github/gh-models

Using the extension

To see a list of all available commands, run gh models.

There are a few key ways you can use the extension:

• To ask a model multiple questions using a chat experience, run gh models run. Select your model from the listed models, then send your prompts.
• To ask a model a single question, run gh models run MODEL-NAME "QUESTION" in your terminal. For example, to ask the GPT 4o model why the sky is blue, you can run gh models run gpt-4o "why is the sky blue?".
• To provide the output of a command as context when you call a model, you can join a separate command and the call to the model with the pipe character (|). For example, to summarize the README file in the current directory using the GPT 4o model, you can run cat README.md | gh models run gpt-4o "summarize this text".