Skip to main content

​
Deployment

Deployment is the process of making a BindAI project available for production use. A deployed project includes its applications, agents, workflows, tools, knowledge, configuration, and runtime services, allowing users or external systems to interact with it reliably. BindAI is designed so the same project can run locally during development and scale to production environments with minimal changes.

​
Deployment Architecture

A typical deployment consists of several layers.
Each layer has a clearly defined responsibility.

​
What Gets Deployed?

A BindAI deployment typically includes:
  • project configuration
  • applications
  • agents
  • workflows
  • shared tools
  • knowledge resources
  • memory providers
  • model provider configuration
Together these components form a complete AI solution.

​
Development vs Production

Development focuses on rapid iteration. Production focuses on reliability. The project structure remains the same in both environments.

​
Configuration

Deployment-specific settings should be stored outside application logic. Examples include:
  • API keys
  • model selection
  • provider endpoints
  • database connections
  • environment variables
  • feature flags
Separating configuration from code simplifies deployment across multiple environments.

​
Environment Variables

Sensitive information should be supplied through environment variables. Examples:
Avoid hardcoding secrets inside your project.

​
Model Providers

Different environments may use different model providers. Example:
Changing providers should not require changes to workflow or agent logic.

​
Knowledge Resources

Knowledge bases should be indexed before deployment. Typical deployment process:
This avoids rebuilding indexes every time the application starts.

​
Scheduled Workflows

Production deployments often execute scheduled workflows automatically. Examples include:
  • nightly reports
  • document indexing
  • data synchronization
  • monitoring tasks
The workflow scheduler manages recurring execution.

​
Human Tasks

Applications using Human Tasks require a mechanism for users to review and complete pending tasks. Conceptually:
The workflow engine remains independent of the interface used to manage tasks.

​
Logging

Production deployments should log important events. Examples include:
  • workflow execution
  • agent requests
  • tool execution
  • retries
  • timeout events
  • human task creation
  • provider errors
Logging simplifies monitoring and troubleshooting.

​
Monitoring

Monitoring provides visibility into production behavior. Useful metrics include:
  • request count
  • workflow duration
  • model latency
  • retry frequency
  • timeout frequency
  • tool failures
  • provider usage
  • token consumption
Monitoring helps identify performance bottlenecks before they affect users.

​
Scaling

As demand grows, applications can scale independently.
Because applications are modular, additional instances can be added without changing workflow definitions.

​
Reliability

Production systems should combine multiple reliability features. Examples include:
  • retry policies
  • timeout policies
  • scheduled execution
  • health monitoring
  • human approval
  • structured logging
These mechanisms improve stability under real-world conditions.

​
Deployment Checklist

Before deploying a project:
  • Validate all workflows.
  • Configure model providers.
  • Register required tools.
  • Index knowledge resources.
  • Configure memory providers.
  • Store secrets securely.
  • Enable logging and monitoring.
  • Test scheduled workflows.
  • Verify Human Task handling.
  • Confirm environment configuration.

​
Best Practices

  • Keep application logic independent of deployment details.
  • Store secrets outside the project source code.
  • Deploy knowledge indexes separately from application startup.
  • Monitor production metrics continuously.
  • Use retries and timeouts appropriately.
  • Test workflows before production releases.
  • Maintain separate configurations for development, staging, and production.

​
Summary

Deployment transforms a BindAI project into a production-ready AI system. By separating configuration from application logic and combining workflows, agents, shared tools, scheduling, monitoring, and external providers, BindAI projects can scale from local development to enterprise deployments while maintaining the same architecture.