05 / 07 · App deployment
Deploy the AI application from laptop to server
AI Application Deployment
Key concepts
- During development, the laptop can call the server inference API. For team use, deploy the application or agent to a host that can keep running.
- Deploy a versioned build with dependencies or a container image. Configure the inference URL, credentials, data paths, network access, service startup, logs, health checks, and rollback on the server.
- Evaluate Agent2Agent (A2A) when independent agents need a shared interoperability protocol.
Checkpoint
The laptop prototype works. How will you deploy it for daily team use?
- Single-host deployment
Fits smaller services with limited operations capacity and maintenance windows.
Tradeoffs · Keeps deployment simple but shares resources and failure impact.
Verify · Test resource limits, restart behavior, startup, and host recovery time.
- Separate application and GPU hosts
Fits shared inference or separate application and model operations.
Tradeoffs · Allows independent updates but adds networking, authentication, and tracing.
Verify · Simulate inference timeouts, updates, and outages and verify graceful handling.
- Prepare replacement capacity
Fits services with explicit downtime and recovery targets.
Tradeoffs · Requires spare capacity, state backups, and tested failover procedures.
Verify · Rehearse outage and rollback and measure recovery time and data loss.
Hands-on check
Turn off the development laptop and complete a task from another device. Repeat after restarting the service.
References
Illustration credits
- SINGLE HOST · AI-generated concept illustration. Verify actual specifications in product documentation.
- SPLIT SERVICES · AI-generated concept illustration. Verify actual specifications in product documentation.
- RECOVERY CAPACITY · AI-generated concept illustration. Verify actual specifications in product documentation.