The Text-2-SQL (agent.query_custom_table) seems had a BUG

[Apri1221]

I try to run example RAG with sqlite / postgres - biz_bot.py.
Ask: what is the annual rate of the base salary?

And then everything is error, the query seems no space:
SELECTSUM(annual_spend)FROMcustomer_table_1WHEREcustomer_name='basesalary'

Sorry I could not find an answer to your question.<br/>Here is the SQL query that was generated by your question: <br/>SELECTSUM(annual_spend)FROMcustomer_table_1WHEREcustomer_name='basesalary'.<br/> If this missed the mark, please try asking the question again with a little more specificity.

I debug:

model_response = agent.query_custom_table(prompt, db=db, table=table_name)

for x in range(0, len(agent.research_list)):
      print("research: ", x, agent.research_list[x])

research: 1 {'step': 11, 'tool': 'sql', 'db_response': [], 'sql_query': "SELECTSUM(annual_spend)FROMcustomer_table_1WHEREcustomer_name='basesalary'", 'query': 'what is the annual rate of the base salary?', 'db': 'postgres', 'work_item': 'CREATE TABLE customer_table_1 (customer_name text, account_number bigint, customer_level text, vip_customer text, annual_spend bigint, user_name text )'}

everything same code with: https://github.com/llmware-ai/llmware/blob/main/examples/Use_Cases/biz_bot.py

[onestardao]

It looks like what you’re hitting is very close to Problem #11 (SQL agent output drift / formatting mismatch).
Basically the model generates a query with subtle spacing/format issues, so the executor fails even though the logic looks fine.

There’s a checklist / workaround for this (how to normalize and guard against the drift).
If you’d like, I can share the full problem map I use (16 known failure cases, including this one) — just let me know and I’ll send the link.