feat: Add ModelsLab LLM for text-to-SQL

[adhikjoshi]

feat: Add ModelsLab LLM integration

Summary

Adds ModelsLab as a new LLM integration in
src/vanna/integrations/modelslab/, giving Vanna users access to ModelsLab's
uncensored Llama-based models for SQL generation, explanation, and chart code.

ModelsLab's chat API is fully OpenAI-compatible, so ModelsLabLlmService
extends OpenAILlmService and inherits all streaming, tool-calling, and
structured-output features — zero code duplication.

---

Files changed

File	Change
src/vanna/integrations/modelslab/__init__.py	NEW: Package init & exports
src/vanna/integrations/modelslab/llm.py	NEW: ModelsLabLlmService(OpenAILlmService)
tests/integrations/test_modelslab_llm.py	NEW: 12 unit tests

---

Usage

import vanna
from vanna.integrations.modelslab import ModelsLabLlmService
from vanna.integrations.chromadb import ChromaDBVectorStore  # or your store

vn = vanna.create_client(
    llm=ModelsLabLlmService(
        model="llama-3.1-70b-uncensored",   # optional, default: 8b
        api_key="YOUR_MODELSLAB_API_KEY",   # or set MODELSLAB_API_KEY env var
    ),
    vector_store=ChromaDBVectorStore(),
)

# Train on your schema
vn.train(ddl="CREATE TABLE orders (id INT, amount DECIMAL, created_at DATE)")

# Generate SQL
sql = await vn.generate_sql("Show me monthly revenue for 2024")
print(sql)
# SELECT DATE_TRUNC('month', created_at) AS month,
#        SUM(amount) AS revenue
# FROM orders
# WHERE created_at BETWEEN '2024-01-01' AND '2024-12-31'
# GROUP BY 1 ORDER BY 1

---

Implementation

# src/vanna/integrations/modelslab/llm.py

class ModelsLabLlmService(OpenAILlmService):
    def __init__(self, model=None, api_key=None, base_url=None, **kwargs):
        resolved_api_key = api_key or os.getenv("MODELSLAB_API_KEY")
        if not resolved_api_key:
            raise ValueError("... set MODELSLAB_API_KEY ...")
        super().__init__(
            model=model or os.getenv("MODELSLAB_MODEL") or "llama-3.1-8b-uncensored",
            api_key=resolved_api_key,
            base_url=base_url or os.getenv("MODELSLAB_BASE_URL")
                     or "https://modelslab.com/api/uncensored-chat/v1",
            **kwargs,
        )

---

Supported models

Model	Context	Notes
llama-3.1-8b-uncensored (default)	128K	Fast
llama-3.1-70b-uncensored	128K	Higher quality

---

Environment variables

Variable	Description
MODELSLAB_API_KEY	API key (required)
MODELSLAB_MODEL	Model ID override (optional)
MODELSLAB_BASE_URL	Endpoint URL override (optional)

---

Testing

pip install pytest pytest-asyncio
pytest tests/integrations/test_modelslab_llm.py -v
# 12 tests — all pass

---

Checklist

• New directory src/vanna/integrations/modelslab/ following existing structure
• ModelsLabLlmService extends OpenAILlmService (no code duplication)
• API key from arg OR MODELSLAB_API_KEY env var; clear error when missing
• Model and base URL also configurable via env vars
• __init__.py with __all__ exports
• Unit tests covering: inheritance, init, default/custom model, API key
  (arg + env + missing), base URL (arg + env + default), constants

---

About ModelsLab

ModelsLab provides API access to uncensored Llama,
Flux, SDXL, video, audio, and 3D models. Their uncensored LLMs are useful for
Vanna when SQL generation requires reasoning about sensitive data topics without
content filters.

API docs: docs.modelslab.com
API keys: modelslab.com/api-keys