Questions become tool calls
The agent does not guess table names. It uses tools to list tables, inspect columns, sample values, then writes SQL.
Talk to your database
Talk to your data.
Let an agent
handle the SQL.
Connect PostgreSQL, ask in plain English, and get the SQL plus results. Writes pause until you approve them.
Works with your stack
PostgreSQLLangGraphLangChainNext.jsExpressPrismaTypeScriptOllamaGemini
PostgreSQLLangGraphLangChainNext.jsExpressPrismaTypeScriptOllamaGemini
Features
Everything you need to talk to your data.
Connect your database, choose a model, and start asking questions. Source Agent inspects the schema, writes the SQL, and shows its work.
Postgres is the source of truth
Point it at a real database. Schema, keys, and relationships come from the connection - not from a prompt.
Writes wait for you
INSERT, UPDATE, and DELETE pause the graph. You approve or reject before anything hits the database.
A conversation, not a console
Results come back in chat with the SQL the model used, so you can see how it got there - and catch mistakes.
One connection.
Every question answered.
Schema tools, safe SQL execution, and human approval for writes - in one conversation.
Try a conversationHow it works
Connect, inspect, then ask.
Three steps from a Postgres connection to an answer - with the SQL visible, and writes waiting on you.
01Connect
New connection
Postgres URL
AI provider
Model + key
Plug in Postgres and a model.
Credentials stay encrypted. The agent only sees schema and the rows it is allowed to query.
02Inspect
Tool loop
● get_table_schema
Then
execute_sql
or interrupt →
Ask in plain English.
The graph picks tools, retries bad SQL, and shows the query it ran so you can follow the reasoning.
03Approve
Write querypaused
UPDATE users SET email = '...' WHERE name = 'Aakash';
Waiting for your approval
Reads run. Writes wait.
SELECT comes back immediately. Anything that changes rows interrupts the graph until you say yes.
From a question to a query you can audit.
One chat. Clear SQL.
FAQ
Frequently asked questions.
How Source Agent connects, writes SQL, and keeps writes behind your approval.
How does Source Agent connect to my database?
+
Add a PostgreSQL connection string. Source Agent inspects tables, columns, primary keys, and foreign keys - then uses that schema to answer questions.
Which databases are supported?
+
PostgreSQL. The tools read Postgres catalogs for schema, keys, and relationships.
Does the model see my whole database?
+
It sees what the tools return: table lists, schemas, samples, and query results. Credentials are encrypted at rest.
Why does it stop on UPDATE or DELETE?
+
Write queries pause the agent. You approve or reject the SQL before it runs against your database.
Which models can I use?
+
Bring your own provider and key. Configure it in the studio and the agent binds tools to that model.
Where do I start?
+
Create an account, add a database, pick a model, and ask something your schema can actually answer.
Ready when you are
Bring a database. Ask it something real.
Sign in, connect Postgres, pick a model, and start asking questions your schema can actually answer.
Your Postgres
Your model key
Human approval
DB
AI
SQL
HITL