AI Services

AI Services Skill

Configure DigitalOcean Gradient AI Platform for App Platform applications.

Tip: This is one specialized skill in the App Platform library. For complex multi-step projects, consider using the planner skill to generate a staged approach. For an overview of all available skills, see the root SKILL.md.

Quick Decision

What do you need?
├── Simple LLM API calls → Serverless Inference
│   OpenAI-compatible API, no agent management
│
└── Full AI agents → Agent Development Kit (ADK)
    Knowledge bases, RAG, guardrails, multi-agent routing

Credential Handling

Model access keys follow the standard credential hierarchy:

1. GitHub Secrets (recommended): User creates key → adds to GitHub Secrets → app spec references
2. App Platform Secrets: Set via doctl apps update with type: SECRET

# App Spec pattern
envs:
  - key: MODEL_ACCESS_KEY
    scope: RUN_TIME
    type: SECRET
    value: ${MODEL_ACCESS_KEY}   # From GitHub Secrets

Key creation: Control Panel → Serverless Inference → Model Access Keys

Keys shown only once after creation—store securely.

Quick Start: Serverless Inference

# .do/app.yaml
services:
  - name: api
    envs:
      - key: MODEL_ACCESS_KEY
        scope: RUN_TIME
        type: SECRET
        value: ${MODEL_ACCESS_KEY}
      - key: INFERENCE_ENDPOINT
        value: https://inference.do-ai.run

# Python SDK (OpenAI-compatible)
from openai import OpenAI
import os

client = OpenAI(
    base_url=os.environ["INFERENCE_ENDPOINT"] + "/v1",
    api_key=os.environ["MODEL_ACCESS_KEY"],
)

response = client.chat.completions.create(
    model="llama3.3-70b-instruct",
    messages=[{"role": "user", "content": "Hello!"}],
)

Full guide: See serverless-inference.md

Quick Start: Agent Development Kit

# Install and configure
pip install gradient-adk
gradient agent configure

# Run locally
gradient agent run
# → http://localhost:8080/run

# Deploy to DigitalOcean
gradient agent deploy

# Agent entrypoint
from gradient_adk import entrypoint

@entrypoint
def entry(payload, context):
    query = payload["prompt"]
    return {"response": "Hello from agent!"}

Full guide: See agent-development-kit.md

Available Models

# List all available models
doctl genai list-models

Check Gradient AI Models for current availability.

Reference Files

• serverless-inference.md — SDK setup, API parameters, examples
• agent-development-kit.md — ADK workflow, knowledge bases, guardrails

Quick Troubleshooting

Integration with Other Skills

• → designer: Add AI service environment variables to app spec
• → deployment: Model access key stored in GitHub Secrets
• → devcontainers: Test AI integrations locally before deployment
• → planner: Plan AI-enabled app deployments

Documentation Links

• Gradient AI Platform
• Available Models
• Serverless Inference
• Agent Development Kit