Skip to main content
Workflows are in beta. Please share any feedback with us in GitHub discussions.
Workflows are a way to run and manage background agents in Continue. You can use them for:
  • Addressing small nitpicks and bugs
  • Building boilerplate-heavy features
  • Investigating an issue to kickstart your work
  • Automated security scanning
  • Running repeatable tasks with your own rules, prompts, and MCP servers
  • Much more!

​
Getting Started

To run your first workflow:
  1. Go to hub.continue.dev/agents Workflows Github and Model setup
  2. Connect with your GitHub account - This is required to enable repository access and branch selection
  3. Set up a model - You can either provide an Anthropic API key, or follow the instructions to add a different model Workflows Interface
  4. Select a repository - Choose which codebase the workflow should operate on
  5. Choose a branch - Pick the specific branch for the workflow to work with
  6. Enter your prompt - Describe the task you want the AI agent to accomplish
  7. Click the send button or press enter to kick off your workflow

​
Example Workflow Tasks

Here are some example tasks you can try with your workflows:
  • “Fix the TypeError in api/users.ts where the user object might be undefined”
  • “Add null checks to all database query results in the services/ directory”
  • “Fix all ESLint warnings in the components folder”
  • “Update deprecated React lifecycle methods to hooks in legacy components”
  • “Create a new REST endpoint for user profile updates with validation and error handling”
  • “Add pagination to the products list page with previous/next buttons”
  • “Implement dark mode toggle using Tailwind CSS classes across all pages”
  • “Add unit tests for the authentication service using Jest”
  • “Scan the codebase for hardcoded API keys and move them to environment variables”
  • “Add input sanitization to all user-facing form fields”
  • “Update all npm packages with known security vulnerabilities”
  • “Implement rate limiting on the /api/login endpoint”
  • “Add JSDoc comments to all exported functions in the utils/ directory”
  • “Create a README.md for the new payment-processing module with setup instructions”
  • “Generate TypeScript interfaces for all API response schemas”
  • “Add error handling boilerplate to all async functions missing try-catch blocks”
  • “Investigate why the login API is returning 500 errors intermittently and suggest fixes”
  • “Analyze the performance bottleneck in the data processing pipeline”
  • “Review the database schema for the orders table and suggest optimizations”
  • “Find all TODO comments related to authentication and create a summary”
  • “Extract the repeated validation logic in controllers into a shared utility function”
  • “Convert all class components in src/legacy to functional components with hooks”
  • “Rename all instances of ‘userId’ to ‘accountId’ across the codebase”
  • “Split the 500-line UserService.ts into smaller, single-responsibility services”

​
Workflow Documentation

Creating Workflow Agents

Build custom workflow configurations with tools, rules, and prompts

Editing Workflow Agents

Modify existing workflows and manage versions

Continue CLI

Test workflow agents locally before deploying

Continuous AI Blog

Latest insights and best practices for using AI workflows

​
Best Practices for Continuous AI

The practice of using background agents, which we call Continuous AI, requires practice and forethought to set up the right guiderails and habits to fit your development workflow, much like learning to work with a larger engineering team. We are constantly sharing our learnings on the Continuous AI Blog.

Getting Started Successfully

Start Local: Practice running workflows locally using the Continue CLI --workflow flag in “TUI mode”. The Continue CLI is used to run workflows in the hub, so you can also test your workflow agents and prompts locally.Begin Small: Identify and begin with tasks that you are confident can be accomplished by Continue, e.g., a small bug with a simple and isolated fix.Iterate and Improve: Once you have merged a PR created by Continue, try larger tasks. By being willing to start tasks that might not succeed on the first try; you will learn about prompting best practices and limitations of current language models.

Effective Prompting

Be Thorough: Workflows can run for a long time to complete their task, so it is worthwhile to invest in sharing all of the important details.Provide Context: Include information about:
  • Project structure and conventions
  • Expected coding standards
  • Integration requirements
  • Success criteria
Use Examples: When possible, reference existing code patterns or provide examples of desired output.

Team Integration

Communicate with Your Team: Truly embracing Continuous AI likely means acknowledging that a higher volume of PRs will be created and adjusting your code review habits.Establish Guidelines: Work with your team to establish:
  • When to use workflows vs manual development
  • Code review standards for AI-generated code
  • Testing requirements for workflow outputs
  • Rollback procedures if issues arise
Monitor and Adjust: Regularly assess workflow effectiveness and adjust prompts, tools, and processes based on team feedback.
⌘I