Deploy on Cloud Run

Two of the three competition tasks — Tripletex and Astar Island — require you to host a public HTTPS endpoint that our validators call. Cloud Run is the easiest way to deploy one.

What is Cloud Run?

Cloud Run takes a Docker container and gives you a public HTTPS URL. You push your code, it handles scaling, TLS, and everything else. You only pay for actual requests (and with your GCP account, it's free).

Step 1: Write Your Endpoint

Here's a minimal FastAPI endpoint that matches the competition format:

from fastapi import FastAPI
 
app = FastAPI()
 
@app.get("/health")
def health():
    return {"status": "ok"}
 
@app.post("/solve")
async def solve(request: dict):
    prompt = request.get("prompt", "")
    credentials = request.get("tripletex_credentials", {})
 
    # Your AI agent logic here:
    # 1. Parse the prompt
    # 2. Call the Tripletex API using the provided credentials
    # 3. Complete the accounting task
 
    return {"status": "completed"}

Step 2: Create a Dockerfile

FROM python:3.11-slim
 
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
 
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8080"]

And a requirements.txt:

fastapi
uvicorn[standard]
requests

Step 3: Deploy

Open Cloud Shell and run:

# Clone your repo (or upload files via Cloud Shell Editor)
cd ~
git clone <your-repo-url>
cd your-project
 
# Deploy to Cloud Run (builds and deploys in one command)
gcloud run deploy my-agent \
  --source . \
  --region europe-north1 \
  --allow-unauthenticated \
  --memory 1Gi \
  --timeout 300

That's it. Cloud Run builds the Docker image, deploys it, and gives you a URL like:

https://my-agent-xxxxx-lz.a.run.app

Step 4: Submit Your URL

1. Copy the Cloud Run URL
2. Go to the submission page for your task at app.ainm.no
3. Paste the URL and submit
4. Our validators will start calling your endpoint

Tips

Use europe-north1 Region

Deploy in europe-north1 (Finland) — same region as our validators. Lower latency = faster scoring.

gcloud run deploy my-agent --region europe-north1 ...

Handle Cold Starts

Cloud Run scales to zero when idle. The first request after idle may take a few seconds. To keep it warm:

gcloud run deploy my-agent --min-instances 1 ...

This keeps one instance always running — useful during active competition.

Increase Memory for LLMs

If you're calling external LLM APIs (like Vertex AI), the default 512 MB is fine. If you're running a local model, increase memory:

gcloud run deploy my-agent --memory 2Gi --cpu 2 ...

Update Your Deployment

After making changes, just run the same deploy command again:

gcloud run deploy my-agent --source . --region europe-north1 --allow-unauthenticated

Cloud Run builds and deploys a new revision automatically.

View Logs

Check what your endpoint is doing:

gcloud run services logs read my-agent --region europe-north1 --limit 50

Or view logs in the Cloud Console under your service → Logs tab.

Which Tasks Need Cloud Run?