Skip to content

README.md

Latest commit

 

History

History
111 lines (78 loc) · 3.03 KB

README.md

File metadata and controls

111 lines (78 loc) · 3.03 KB

Redis SQL Search Agent

This sample shows an ADK agent answering catalog questions by emitting SQL SELECT statements through the new RedisSQLSearchTool. The tool wraps redisvl.query.SQLQuery (and the optional sql-redis extra), which translates SQL into FT.SEARCH / FT.AGGREGATE against a Redis search index.

What this sample shows

  • Configuring a RedisSQLSearchTool against a bound RedisVL index.
  • Letting the LLM build SQL queries (with :name parameter placeholders) to answer structured questions about a product catalog.
  • Wiring the tool into ADK's Agent and running it under adk web.

Prerequisites

  1. Redis 8.4 running locally (or Redis Cloud with the RediSearch module enabled). The repo root has ./scripts/start-redis.sh if you want an automated start.
  2. A Gemini API key. Get one at aistudio.google.com.

No embeddings are required for this example, so you don't need a vectorizer or NLTK stopwords.

Setup

1. Install dependencies

From the repository root:

uv pip install 'adk-redis[sql,examples]'

The sql extra pulls in redisvl[sql-redis]>=0.18.2, which provides the SQL-to-Redis query translation.

2. Start Redis 8.4

docker run -d --name redis -p 6379:6379 redis:8.4
docker exec redis redis-cli ping   # -> PONG

3. Set your Gemini API key

Copy .env.example to .env and fill in GOOGLE_API_KEY.

4. Load the sample catalog

cd examples/redis_sql_search
python load_data.py

This builds the adk_sql_catalog index and loads ten products spanning electronics, fitness, kitchen, and office categories.

5. Run the agent

adk web redis_sql_search_agent

ADK web opens at http://127.0.0.1:8000. Pick the redis_sql_search_agent app from the dropdown.

Try these prompts

  • "What electronics cost less than 100 dollars?"
  • "Show me kitchen items with a rating of 4.5 or higher."
  • "Find office furniture from DeskWorks under 500 dollars."
  • "Which products have fewer than 20 in stock?"
  • "List BrewLab products sorted by rating."

You will see the agent emit SQL like:

SELECT title, brand, price
FROM adk_sql_catalog
WHERE category = 'electronics' AND price < :max_price

with params={"max_price": 100} flowing through the same tool call.

How it works

RedisSQLSearchTool exposes a two-parameter FunctionDeclaration to the LLM:

  • sql: a SQL SELECT statement (required).
  • params: an optional object whose keys substitute the :name placeholders inside the SQL.

The tool builds a redisvl.query.SQLQuery and runs it against the bound index, returning matching rows as {"status": "success", "count": N, "results": [...]}. Errors become {"status": "error", "error": "..."}, so the agent can fall back rather than crash.

See RedisSQLSearchTool in src/adk_redis/tools/search/sql.py for the implementation, and redisvl.query.SQLQuery for the underlying SQL dialect.

Cleanup

docker stop redis && docker rm redis