ai-services_skill

Overview

This skill configures DigitalOcean Gradient AI serverless inference and the Agent Development Kit (ADK) for App Platform apps. It provides patterns for wiring model access keys, setting environment secrets, choosing between simple inference and full agent workflows, and deploying ADK agents to App Platform. Use it to add LLM inference endpoints or build managed agents with knowledge bases and guardrails.

How this skill works

The skill explains two primary flows: Serverless Inference for direct OpenAI-compatible API calls, and ADK for building deployable agents. It shows how to create and store model access keys, reference them in app specs or GitHub Secrets, and configure runtime environment variables. It also includes quick start examples for the Python SDK and ADK CLI for local runs and App Platform deployment.

When to use it

• You need simple LLM API calls without agent orchestration (Serverless Inference).
• You want full-featured conversational agents with RAG, guardrails, or multi-agent routing (ADK).
• You must securely provide model access keys to App Platform services.
• You are deploying an AI service on DigitalOcean App Platform.
• You want to test AI integrations locally before pushing to production.

Best practices

• Store model access keys in GitHub Secrets and reference them in the app spec; keys are shown only once at creation.
• Use App Platform runtime secrets (type: SECRET) for production environment variables rather than hard-coding keys.
• Choose smaller or cheaper models for dev/testing (e.g., llama3-8b) and larger models for production QA.
• Implement exponential backoff and retries to handle rate limits.
• Verify CLI token scopes for ADK deploys (genai CRUD + project read) to avoid permission errors.

Example use cases

• Add a hosted LLM endpoint to a web app for chat completion using the OpenAI-compatible endpoint.
• Deploy an ADK agent that answers product documentation using a vectorized knowledge base and guardrails.
• Run local developer testing against the ADK before deploying to App Platform via gradient agent deploy.
• Rotate model access keys by updating GitHub Secrets and app spec without changing application code.
• Create multi-agent pipelines that route user queries to specialized agents via ADK.