text2ql 0.2.4

Natural language to any query language, starting with GraphQL

text2ql

Natural-language to query-language toolkit for GraphQL and SQL.

text2ql is built for practical usage: deterministic generation, LLM-assisted generation, schema/mapping normalization, evaluation utilities, and benchmark adapters (Spider/BIRD). Current first-class targets are GraphQL and SQL; planned future targets include JSONata, jq-style queries, Cypher, and dialect-focused SQL paths (PostgreSQL, MySQL, and more).

Why teams use it

• Fast path from text -> query for GraphQL and SQL.
• Three generation modes: deterministic, llm, function_calling.
• Schema + mapping aware generation with validation and confidence scoring.
• Built-in dataset ingestion, synthetic rewrites, and evaluation hooks.
• Built-in Spider/BIRD benchmark loaders and runner.

Install

Python: >=3.10

pip install text2ql

Optional extras:

# Streamlit playground
pip install "text2ql[app]"

# SQL execution backend support (SQLAlchemy)
pip install "text2ql[sql]"

# Local development tools
pip install -e ".[dev]"

Quick Start (CLI)

1. Deterministic GraphQL:

text2ql "show top 5 client records with mail state enabled" \
  --target graphql \
  --schema '{"entities":["customers"],"fields":{"customers":["id","email","status"]}}' \
  --mapping '{"entities":{"client":"customers"},"fields":{"mail":"email"},"filters":{"state":"status"},"filter_values":{"status":{"enabled":"active"}}}'

2. Deterministic SQL:

text2ql "show customers highest total first 5 offset 10" \
  --target sql \
  --schema '{"entities":["customers"],"fields":{"customers":["id","total","status"]}}'

3. LLM mode:

export OPENAI_API_KEY=...
text2ql "show latest 5 orders with status active" \
  --target sql \
  --mode llm \
  --llm-model gpt-4o-mini \
  --schema '{"entities":["orders"],"fields":{"orders":["id","status","createdAt"]}}'

4. Function-calling / structured output mode:

export OPENAI_API_KEY=...
text2ql "show latest 5 orders with status active" \
  --target graphql \
  --mode function_calling \
  --llm-model gpt-4o-mini \
  --schema '{"entities":["orders"],"fields":{"orders":["id","status","createdAt"]},"args":{"orders":["status","limit","orderBy","orderDirection"]}}'

Quick Start (Python API)

from text2ql import Text2QL

service = Text2QL()
result = service.generate(
    text="list active customers",
    target="graphql",
    schema={"entities": ["customers"], "fields": {"customers": ["id", "status", "email"]}},
    mapping={"filters": {"state": "status"}, "filter_values": {"status": {"active": "active"}}},
)

print(result.query)
print(result.confidence)

LLM provider wiring:

from text2ql import Text2QL
from text2ql.providers.openai_compatible import OpenAICompatibleProvider

provider = OpenAICompatibleProvider(
    api_key="...",  # or use OPENAI_API_KEY / TEXT2QL_API_KEY
    model="gpt-4o-mini",
    use_structured_output=True,
)
service = Text2QL(provider=provider)

Streamlit Playground

Use hosted app: https://text2ql.streamlit.app/ For local/private data: pip install -e ".[app]". Run locally: python -m streamlit run examples/streamlit_app.py.

Modes at a glance

Mode	What it does	Best for
deterministic	Rule/schema-driven generation	Fast, predictable production defaults
llm	Direct LLM-assisted generation	Broad language coverage
function_calling	Schema-constrained structured intent path	Better output shape control

Production Setup (brief)

Recommended project files:

project/
  schema.json
  mapping.json
  data.json                # optional; used for payload execution checks
  expected_query.sql       # optional
  expected_rows.json       # optional

Production checklist:

1. Keep schema.json and mapping.json in source control.
2. Start with --mode deterministic for baseline reliability.
3. Enable --mode llm or --mode function_calling after baseline tests pass.
4. Use retry settings for provider calls (--llm-max-retries, --llm-retry-backoff).
5. Run evaluation/benchmark checks in CI before releases.
6. Use -v during rollout/debugging for detailed logs.

Useful CLI operations:

# Generate hybrid mapping (auto baseline + optional overrides)
text2ql --generate-hybrid-mapping \
  --schema-file ./schema.json \
  --data-file ./data.json \
  --mapping-output-file ./mapping.generated.json

# Execute generated query against payload JSON and compare expected output
text2ql "how many qqq do i own" \
  --target graphql \
  --schema-file ./schema.json \
  --mapping-file ./mapping.json \
  --data-file ./data.json \
  --expected-execution-file ./expected_rows.json

Benchmarking (Spider & BIRD)

text2ql ships with benchmark adapters and runner APIs:

• load_spider(...)
• load_bird(...)
• run_benchmark(...), arun_benchmark(...)
• format_report(...)

CLI examples:

# Spider
text2ql --benchmark spider --benchmark-path /path/to/spider --benchmark-mode structural

# BIRD
text2ql --benchmark bird --benchmark-path /path/to/bird-minidev --benchmark-mode execution

Latest LLM benchmark snapshot (synthetic harness)

Using run_llm_benchmark.py (50 Spider-style + 50 BIRD-style synthetic examples, mode=llm, gpt-4o-mini):

Benchmark	Exact	Structural	Execution	Errors
Spider (50)	62.0%	64.0%	84.0%	0/50
BIRD (50)	70.0%	78.0%	90.0%	0/50

Raw summary:

  SUMMARY
====================================================================
  Spider    Exact=62.0%  Structural=64.0%  Execution=84.0%  Errors=0/50
  BIRD      Exact=70.0%  Structural=78.0%  Execution=90.0%  Errors=0/50

Reproduce:

OPENAI_API_KEY=... ./venv/bin/python run_llm_benchmark.py
# or
OPENAI_API_KEY=... python3 run_llm_benchmark.py

Evaluation + Synthetic Data Utilities

Public helpers:

• ingest_dataset(...)
• generate_synthetic_examples(...)
• evaluate_examples(...)
• aevaluate_examples(...)
• rewrite_user_utterance(...)

Built-in rewrite plugins include:

• generic, portfolio, banking, crm, healthcare, ecommerce

Repo layout (high level)

src/text2ql/
  core.py                 # Text2QL facade
  cli.py                  # CLI entrypoint
  engines/                # GraphQL/SQL engines + stage modules
  renderers.py            # GraphQLIRRenderer / SQLIRRenderer
  evaluate.py             # exact/structural/backend evaluation
  benchmarks/             # Spider/BIRD loaders + runner
  mapping.py              # hybrid mapping generation
  dataset.py              # ingestion + synthetic variants
  sql_executor.py         # SQLAlchemy-backed execution backend

Testing

python3 -m pytest -m unit
python3 -m pytest -m e2e
python3 -m pytest

License

Apache-2.0