Quickstart

Create a query_sk_ API key in your dashboard, then log in. The query executable should be on your PATH.

query login

The CLI prompts for the key without displaying it, validates it without submitting a query, and writes it to your user configuration directory under query/config.json — atomically, with permissions 0600. You do not configure an endpoint: the release binary already knows where the service lives.

For non-interactive setup, pipe the key over standard input, or skip the config file entirely and pass it in the environment:

printf '%s\n' "$QUERY_TOKEN" | query login

QUERY_API_TOKEN takes precedence over a saved login.

Save this as trades.sql. Values you want to vary go in as {{name}} placeholders rather than being pasted into the SQL text.

SELECT  date_trunc('hour', ts) AS hour,  count_if(is_buy) AS buys,  count_if(NOT is_buy) AS sells,  sum(sol_amount) AS sol_volumeFROM pump_fun_solana.pump_fun_evt_trade_eventWHERE ts >= {{start_time}}GROUP BY hourORDER BY hour;

Submit the file and supply each parameter with its type. OnchainQueries validates the statement, resolves the table against its catalog, selects an engine, executes, and downloads the results.

query sql \  --file trades.sql \  --param 'start_time:timestamp=2026-08-01 00:00:00' \  --output ./results

The CLI prints the query ID, the selected engine, the state and the attempt number while it waits. On completion it downloads every result part and writes the manifest last:

results/├── part-00000.parquet└── manifest.json

Each part is downloaded to a .partial file, verified against its size and checksum, then atomically renamed. Rerunning against the same output directory reuses any part that already matches, and local files stay available after the server-side result expires.

• CLI reference — every command and flag, the interactive shell, and troubleshooting.
• OnchainQL contract — what OnchainQueries accepts and what it refuses.
• Tables & naming — how to address a table, and what the convenience views cover.
• Dataset catalog — every program and every table, browsable.