OceanPulse

Inspiration

Every DigitalOcean user has the same experience: you spin up a Droplet for a weekend project, forget about it, and three months later you're paying $48/mo for a server running at 2% CPU with no firewall and no backups. Multiply that across 600,000+ DO customers and you're looking at millions in wasted spend and exposed infrastructure.

Existing tools show you dashboards. OceanPulse actually fixes things.

The idea was simple: what if an AI agent could connect to your DO account, understand your entire infrastructure, tell you exactly what's wrong, and — with your permission — fix it on the spot?

What it does

OceanPulse is an AI infrastructure health agent. You paste your DigitalOcean API token, and it:

1. Scans every Droplet, database, volume, firewall, and snapshot on your account
2. Scores your infrastructure health across four dimensions: Cost, Performance, Security, and Architecture
3. Surfaces issues — missing firewalls, no backups, unused volumes, oversized servers
4. Fixes them — one-click actions that call the real DO API to create firewalls, enable backups, resize Droplets, and delete waste
5. Advises — an AI chat advisor that sees your live infrastructure data and gives specific recommendations referencing your actual server names and dollar amounts

Every write operation requires explicit user approval. No surprises.

How I built it

Frontend: Next.js 15 with React 19, TypeScript, and Tailwind CSS v4. Three API routes power everything:

• /api/scan — reads the DO API (Droplets, databases, volumes, firewalls, snapshots, billing)
• /api/chat — sends user messages to DO Serverless Inference with live infrastructure context injected into the system prompt
• /api/fix — executes approved write actions (POST /firewalls, POST /droplets/actions, DELETE /volumes)

AI: The chat advisor calls DigitalOcean's Serverless Inference API using openai-gpt-oss-20b. Before each request, the backend fetches the user's current infrastructure and injects it into the system prompt — so the AI sees real resource names, real costs, and real issues. A Gradient Agent is also configured as a primary route with serverless inference as fallback.

Scoring: Security deducts 25 points per Droplet without a firewall and 10 per Droplet without backups. Cost flags unattached volumes and waste. Architecture checks VPC isolation. Overall = average of all four dimensions.

Hosting: Entirely on DigitalOcean — App Platform for the frontend, Gradient Agent Platform for the agent, Serverless Inference for AI, DO API v2 for all read/write operations.

Challenges I faced

Auth layering. DigitalOcean has three different auth mechanisms: API tokens for the DO API, Model Access Keys for Serverless Inference, and Agent Access Keys for the Gradient endpoint. Getting the right key to the right service took multiple debugging rounds.

Model availability. Not all models listed in Serverless Inference are available on every subscription tier. I started with gpt-4o-mini, got 401 errors, and had to switch to openai-gpt-oss-20b — which actually works great and costs $0.05/1M tokens.

Real write operations. Making an agent that actually modifies infrastructure (not just reads it) required careful guardrails. Every fix button shows exactly what will change, what it saves, and what the risk is before the user can approve.

TypeScript strict mode vs API responses. DO API responses are loosely typed, but Next.js enforces strict TypeScript. Every forEach and map callback needed explicit any types to pass the production build.

What I learned

• DigitalOcean's Gradient AI stack is genuinely powerful — the combination of agents + serverless inference + function calling covers the full agent lifecycle
• Injecting live data into LLM system prompts is the simplest way to make AI responses contextual and specific
• Users trust AI agents more when they can see exactly what will happen before approving an action
• The DO API is remarkably consistent — every write operation follows the same pattern, which made building the fix engine straightforward

What's next

• Metrics integration — pull real CPU/memory/disk data from DO Monitoring API for accurate performance scoring
• Scheduled scans — run health checks on a cron and alert via email/Slack when scores drop
• Multi-account — manage multiple DO accounts from one dashboard
• Cost forecasting — predict next month's bill based on current usage trends

Built With

• digitalocean-api-v2
• digitalocean-app-platform
• digitalocean-gradient-agent-platform
• digitalocean-serverless-inference
• ibm-plex-mono
• instrument
• lucide-react
• marked
• next.js-15
• node.js
• react-19
• tailwind-css-v4
• typescript