data-analyst

Upload CSVs or JSON datasets, then ask questions in plain English. Workers AI generates SQL, runs it against your D1 tables, and returns structured results.

How it works

1. Upload a CSV or JSON file via POST /upload — the agent parses it, infers column types, and creates a D1 table
2. Ask a question via POST /query — Workers AI (llama-3.3-70b) generates a SELECT statement from your question and the table schema
3. The SQL runs against D1 and the results are returned as JSON alongside the generated SQL

API

Write endpoints (/upload, DELETE /tables/:name) require Authorization: Bearer <ADMIN_TOKEN>. All read endpoints are public.

Method	Path	Auth	Description
POST	/upload	yes	Upload CSV or JSON — creates a D1 table
GET	/tables	no	List all datasets
GET	/tables/:name/schema	no	Column names + types for one table
POST	/query	no	Natural language → SQL → results
GET	/history	no	Recent query history (?dataset=, ?limit=)
DELETE	/tables/:name	yes	Drop a dataset and its table
GET	/	no	Health check

Upload a CSV

# Multipart form (recommended)
curl -X POST https://data-analyst.proagentstore.online/upload \
-H "Authorization: Bearer $ADMIN_TOKEN" \
-F "file=@sales.csv" \
-F "name=Sales 2025"# Raw CSV body
curl -X POST "https://data-analyst.proagentstore.online/upload?name=sales" \
-H "Authorization: Bearer $ADMIN_TOKEN" \
-H "Content-Type: text/csv" \
--data-binary @sales.csv

Upload JSON

# Accepts an array of objects or an object with a top-level array property
curl -X POST https://data-analyst.proagentstore.online/upload \
-H "Authorization: Bearer $ADMIN_TOKEN" \
-F "file=@orders.json" \
-F "name=Orders"

Ask a question

curl -X POST https://data-analyst.proagentstore.online/query \
-H "Content-Type: application/json" \
-d '{"question": "What are the top 5 products by total revenue?"}'# When multiple datasets exist, specify which one
curl -X POST https://data-analyst.proagentstore.online/query \
-H "Content-Type: application/json" \
-d '{"question": "Show average order value by month", "table": "orders"}'

Response:

{
"id": "abc123",
"question": "What are the top 5 products by total revenue?",
"sql": "SELECT \"product\", SUM(\"revenue\") AS total_revenue FROM \"t_sales_2025_xyz789\" GROUP BY \"product\" ORDER BY total_revenue DESC LIMIT 5;",
"row_count": 5,
"results": [
{ "product": "Widget A", "total_revenue": 48200 },
...
]
}

List datasets

curl https://data-analyst.proagentstore.online/tables

Get schema

curl https://data-analyst.proagentstore.online/tables/orders/schema

Query history

curl "https://data-analyst.proagentstore.online/history?limit=20"
curl "https://data-analyst.proagentstore.online/history?dataset=orders"

Delete a dataset

curl -X DELETE https://data-analyst.proagentstore.online/tables/orders \
-H "Authorization: Bearer $ADMIN_TOKEN"

Upload formats

CSV

Standard comma-separated. Quoted fields (including embedded commas and newlines) are handled. The first row is the header.

JSON

Either a top-level array of objects:

[{"id": 1, "name": "Alice", "revenue": 1200}, ...]

Or an object whose first array-valued property is used:

{"orders": [{"id": 1, ...}, ...]}

Nested objects are stringified to TEXT.

Column type inference

The agent inspects up to 200 rows to decide the SQLite type:

Inferred type	Condition
INTEGER	All non-empty values are whole numbers
REAL	All non-empty values are valid numbers (including decimals)
TEXT	Anything else

Setup

1. Create the D1 database

wrangler d1 create data-analyst-db
# Copy the database_id into wrangler.toml

2. Run migrations

wrangler d1 migrations apply data-analyst-db --remote

3. Deploy the Worker

pnpm install
wrangler deploy

4. Set secrets

wrangler secret put ADMIN_TOKEN
# Also update Doppler:
doppler secrets set ADMIN_TOKEN=<value> --project pags --config prd

Local dev

pnpm dev
# Upload a file:
curl -X POST "http://localhost:8787/upload?name=test" \
-H "Content-Type: text/csv" \
--data-binary @sample.csv
# Query it:
curl -X POST http://localhost:8787/query \
-H "Content-Type: application/json" \
-d '{"question": "How many rows are there?"}'

Bindings

Binding	Type	Purpose
AI	Workers AI	Translates natural language questions to SQL
DB	D1	Stores dataset metadata, query history, and all uploaded data tables

D1 schema

Metadata tables (created by migration)

Table	Purpose
datasets	Registry of uploaded datasets — name, table_name, column schema, row count
query_history	Every query run: question, generated SQL, result row count, any error

Data tables (created dynamically per upload)

Each upload creates a table named t_<sanitised-name>_<6-char-uid>. Columns are named after the CSV headers / JSON keys, with types inferred from the data. These tables are dropped when the dataset is deleted.