Build and Deploy Semantic Search with FastAPI and pgvector

Build a personal knowledge-base search engine where users paste paragraphs and search by meaning, not just exact words.

The Stack

• Python with FastAPI
• Postgres with the pgvector extension
• Gemini API for embeddings
• Adios runtime with secrets and managed database wiring

Why This Architecture Fits Semantic Search

AI backends often import heavier Python libraries and initialize API clients at startup. On Adios, your FastAPI app runs as a normal long-lived service, so your backend can stay warm instead of rebuilding state on every request.

1. Build the API

Ask the Adios AI agent:

Build a FastAPI semantic search API. Add endpoints to insert documents and
search them using embeddings. Use Gemini for embeddings, Postgres pgvector for
similarity search, and /healthz for health checks.

2. Add adios.yaml

name: semantic-search
region: de
replicas: 1

build_cmd: python -m pip install -r requirements.txt
start_cmd: python -m uvicorn app.main:app --host 0.0.0.0 --port $PORT

runtime:
  name: python@3.13
  port: 8000
  health_path: /healthz
  memory_mb: 1024

secrets:
  DATABASE_URL: secret://DATABASE_URL
  GEMINI_API_KEY: secret://GEMINI_API_KEY

requires:
  - vector-db

Create a Postgres pgvector resource named vector-db, or set DATABASE_URL manually if you already have one.

3. Deploy

adios secrets set GEMINI_API_KEY
adios up

Your app gets a generated Adios route with TLS, runtime logs, and managed database connection details without a Dockerfile or cloud console setup.

Review the FastAPI deployment path, the pgvector deployment checks, and the exact pgvector template before adapting this example to a production document corpus.