nlp2sql 0.2.0rc14__tar.gz → 0.2.0rc15__tar.gz

nlp2sql-0.2.0rc15/CHANGELOG.md

@@ -0,0 +1,89 @@
+# Changelog
+
+All notable changes to this project are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## Versioning policy
+
+- **MAJOR** — incompatible public API changes (removing/renaming a public symbol
+  or changing a port/ABC contract in a non-backward-compatible way).
+- **MINOR** — backward-compatible functionality (new adapters, new optional
+  parameters, new optional extras).
+- **PATCH** — backward-compatible bug fixes.
+
+Pre-1.0 the project ships release candidates (e.g. `0.2.0rcN`). Public surface
+changes — moving/renaming exported symbols, or adding methods to a published
+port such as `ExampleRepositoryPort` — MUST be recorded under `Unreleased`
+before release and called out as **Breaking** if they affect external
+implementers. Backward-compatible shims are preferred over hard breaks while
+pre-1.0.
+
+## [Unreleased]
+
+## [0.2.0rc15] - 2026-09-18
+
+### Changed
+- Default chat models refreshed to current, non-retired IDs. **User-visible: cost,
+  latency and SQL style change for anyone who does not pin `model`.**
+  - OpenAI: `gpt-4o-mini` -> `gpt-5.6-luna` (1.05M context, 922K input, 128K output).
+    `gpt-4o-mini` remains callable; pin it with `ProviderConfig(model="gpt-4o-mini")`.
+  - Anthropic: `claude-sonnet-4-20250514` (retired by Anthropic on 2026-06-15) ->
+    `claude-sonnet-5` (1M context, 128K output).
+  - Gemini: `gemini-2.0-flash` (shut down by Google on 2026-06-01) ->
+    `gemini-3.6-flash` (1,048,576 context, 65,536 output).
+- `ProviderConfig.DEFAULT_MODELS` is now the single source of truth for provider
+  defaults; the adapters, `Settings.get_provider_config()` and `nlp2sql providers list`
+  read from it.
+- `AnthropicAdapter` no longer sends `temperature` to the Messages API: Claude 4.7+
+  models reject non-default sampling parameters (HTTP 400) and the `anthropic` SDK
+  removes the keyword in 1.0. `ProviderConfig(temperature=...)` is still accepted for
+  Anthropic but ignored (logged at INFO); explicitly pinned older Claude models now run
+  at the API default temperature.
+- `OpenAIAdapter` sends `max_completion_tokens` instead of the deprecated `max_tokens`.
+  For reasoning models (every model outside the `gpt-3.5`, `gpt-4` and `chatgpt-4o`
+  families, including `gpt-5.6-*`) it sets `reasoning_effort` (default `low`, new
+  `reasoning_effort` constructor argument) and omits `temperature`; `gpt-4o`/`gpt-4.1`
+  keep receiving `temperature`. Response metadata gains `reasoning_effort` and
+  `reasoning_tokens`.
+- Unknown model IDs no longer fall back to crippling context limits (OpenAI 8,192,
+  Gemini 30,720, Anthropic 100,000). They resolve by model family where possible and
+  otherwise fall back to 128,000 / 1,048,576 (Gemini family) / 200,000, logging a
+  warning when the adapter is constructed.
+- `OpenAIAdapter` token counting maps the `gpt-5`, `gpt-4.1`, `gpt-4o` and `o1`-`o4`
+  families to `o200k_base` (tiktoken has no entry for `gpt-5.6-*`) and falls back to
+  `o200k_base`, never `cl100k_base`, for unknown IDs.
+- `GeminiAdapter` reads token usage from `usage_metadata` instead of issuing extra
+  `count_tokens` calls after each response.
+- Minimum dependency versions raised: `openai>=1.58.0` (`reasoning_effort`),
+  `tiktoken>=0.7.0` (`o200k_base`), `google-generativeai>=0.8.5` (`usage_metadata`).
+
+### Fixed
+- `connect()` / `ask()` failed out of the box for Anthropic and Gemini because their
+  default model IDs had been retired by the providers.
+- Reasoning/thinking models that exhaust `max_tokens` before emitting content now raise
+  an actionable `ProviderException` instead of "Invalid JSON response" or
+  "Failed to parse response".
+- `AnthropicAdapter` and `GeminiAdapter` re-raise `ProviderException` after exhausting
+  retries instead of surfacing `tenacity.RetryError`.
+
+## [0.2.0rc14] - 2026-09-16
+
+### Added
+- `PgvectorExampleRepository`: a Postgres + `pgvector` implementation of
+  `ExampleRepositoryPort`, an alternative to the on-disk FAISS `ExampleStore`
+  for storing and semantically searching few-shot examples. Stores examples in
+  a single table with a `namespace` column (per-database isolation via
+  `md5(database_url:schema_name)`, byte-identical to `ExampleStore`), cosine
+  distance search over an HNSW index, and idempotent `auto_migrate` DDL.
+- Optional `pgvector` extra: `pip install 'nlp2sql[pgvector]'`.
+- CI: a `pgvector-integration` job that runs the adapter's integration tests
+  against a `pgvector/pgvector:pg16` service. The job fails (rather than skips)
+  when the database or `vector` extension is unavailable
+  (`NLP2SQL_REQUIRE_INTEGRATION`).
+
+### Changed
+- `docker/docker-compose.yml`: the `postgres` service image is now
+  `pgvector/pgvector:pg16` (drop-in replacement for `postgres:16`, adds the
+  `vector` extension) so integration tests run out of the box.

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: nlp2sql
-Version: 0.2.0rc14
+Version: 0.2.0rc15
 Summary: Enterprise-ready Natural Language to SQL converter with multi-provider support. Built for production scale (1000+ tables) with Clean Architecture.
 Project-URL: Homepage, https://github.com/luiscarbonel1991/nlp2sql
 Project-URL: Bug Tracker, https://github.com/luiscarbonel1991/nlp2sql/issues
@@ -26,11 +26,11 @@ Requires-Dist: azure-identity>=1.15.0
 Requires-Dist: boto3>=1.34.0
 Requires-Dist: click>=8.0.0
 Requires-Dist: faiss-cpu>=1.7.4
-Requires-Dist: google-generativeai>=0.3.0
+Requires-Dist: google-generativeai>=0.8.5
 Requires-Dist: greenlet>=3.2.3
 Requires-Dist: httpx>=0.25.0
 Requires-Dist: numpy>=1.24.0
-Requires-Dist: openai>=1.0.0
+Requires-Dist: openai>=1.58.0
 Requires-Dist: pandas>=2.0.0
 Requires-Dist: psycopg2-binary>=2.9.0
 Requires-Dist: psycopg[binary]>=3.2.13
@@ -42,12 +42,12 @@ Requires-Dist: sqlalchemy-redshift>=0.7.0
 Requires-Dist: sqlalchemy>=2.0.0
 Requires-Dist: structlog>=24.1.0
 Requires-Dist: tenacity>=8.2.0
-Requires-Dist: tiktoken>=0.5.0
+Requires-Dist: tiktoken>=0.7.0
 Provides-Extra: all-embeddings
 Requires-Dist: sentence-transformers>=2.5.0; extra == 'all-embeddings'
 Provides-Extra: all-providers
 Requires-Dist: anthropic>=0.18.0; extra == 'all-providers'
-Requires-Dist: google-generativeai>=0.3.0; extra == 'all-providers'
+Requires-Dist: google-generativeai>=0.8.5; extra == 'all-providers'
 Provides-Extra: anthropic
 Requires-Dist: anthropic>=0.18.0; extra == 'anthropic'
 Provides-Extra: dev
@@ -66,7 +66,7 @@ Provides-Extra: embeddings-local
 Requires-Dist: sentence-transformers>=2.5.0; extra == 'embeddings-local'
 Provides-Extra: embeddings-openai
 Provides-Extra: gemini
-Requires-Dist: google-generativeai>=0.3.0; extra == 'gemini'
+Requires-Dist: google-generativeai>=0.8.5; extra == 'gemini'
 Provides-Extra: mssql
 Requires-Dist: pymssql>=2.2.0; extra == 'mssql'
 Provides-Extra: mysql
@@ -344,13 +344,13 @@ postgresql://testuser:testpass@localhost:5432/testdb
 
 ## Provider Comparison
 
-| Provider | Default Model | Context Size | Best For |
-|----------|---------------|--------------|----------|
-| OpenAI | `gpt-4o-mini` | 128K | Fast general purpose usage |
-| Anthropic | `claude-sonnet-4-20250514` | 200K | Larger schemas and long prompts |
-| Gemini | `gemini-2.0-flash` | 1M | High-volume and very large contexts |
+| Provider | Default Model | Context Size | Max Output | Best For |
+|----------|---------------|--------------|------------|----------|
+| OpenAI | `gpt-5.6-luna` | 1.05M (922K input) | 128K | Fast, low-cost general purpose usage |
+| Anthropic | `claude-sonnet-5` | 1M | 128K | Larger schemas and long prompts |
+| Gemini | `gemini-3.6-flash` | 1M | 64K | High-volume and very large contexts |
 
-All models are configurable through `ProviderConfig`.
+Defaults are pinned snapshot IDs, not floating aliases. Override any of them through `ProviderConfig(model=...)`; for example `gpt-5.6-terra` is the higher-quality OpenAI option.
 
 ## Lower-Level API
 

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/README.md

@@ -264,13 +264,13 @@ postgresql://testuser:testpass@localhost:5432/testdb
 
 ## Provider Comparison
 
-| Provider | Default Model | Context Size | Best For |
-|----------|---------------|--------------|----------|
-| OpenAI | `gpt-4o-mini` | 128K | Fast general purpose usage |
-| Anthropic | `claude-sonnet-4-20250514` | 200K | Larger schemas and long prompts |
-| Gemini | `gemini-2.0-flash` | 1M | High-volume and very large contexts |
+| Provider | Default Model | Context Size | Max Output | Best For |
+|----------|---------------|--------------|------------|----------|
+| OpenAI | `gpt-5.6-luna` | 1.05M (922K input) | 128K | Fast, low-cost general purpose usage |
+| Anthropic | `claude-sonnet-5` | 1M | 128K | Larger schemas and long prompts |
+| Gemini | `gemini-3.6-flash` | 1M | 64K | High-volume and very large contexts |
 
-All models are configurable through `ProviderConfig`.
+Defaults are pinned snapshot IDs, not floating aliases. Override any of them through `ProviderConfig(model=...)`; for example `gpt-5.6-terra` is the higher-quality OpenAI option.
 
 ## Lower-Level API
 

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/docs/API.md

@@ -91,7 +91,7 @@ from nlp2sql import ProviderConfig
 config = ProviderConfig(
     provider="openai",
     api_key="sk-...",
-    model="gpt-4o",
+    model="gpt-5.6-terra",
     temperature=0.0,
     max_tokens=4000,
 )

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/examples/advanced/test_multiple_providers.py

@@ -22,7 +22,7 @@ async def test_multiple_ai_providers():
     # Test configuration for each provider
     providers = [
         {
-            "name": "OpenAI GPT-4",
+            "name": "OpenAI",
             "provider": "openai",
             "api_key": os.getenv("OPENAI_API_KEY"),
             "env_var": "OPENAI_API_KEY",
@@ -126,7 +126,7 @@ async def demo_provider_selection():
     use_cases = [
         {
             "use_case": "High accuracy complex queries",
-            "recommended": "OpenAI GPT-4",
+            "recommended": "OpenAI",
             "reason": "Most sophisticated reasoning",
         },
         {
@@ -137,7 +137,7 @@ async def demo_provider_selection():
         {
             "use_case": "Long context/large schemas",
             "recommended": "Anthropic Claude",
-            "reason": "200K token context window",
+            "reason": "1M token context window",
         },
         {
             "use_case": "Privacy-sensitive data",
@@ -152,15 +152,15 @@ async def demo_provider_selection():
         print(f"   Reason: {case['reason']}")
         print()
 
-    print("Cost Comparison (approx):")
-    print("   OpenAI GPT-4: $0.03/1K tokens")
-    print("   Anthropic Claude: $0.015/1K tokens")
-    print("   Google Gemini: $0.001/1K tokens")
+    print("Cost Comparison (approx, per 1M tokens in/out, default models):")
+    print("   OpenAI gpt-5.6-luna: $0.20 / $1.20")
+    print("   Anthropic claude-sonnet-5: $2 / $10")
+    print("   Google gemini-3.6-flash: $0.75 / $3.75")
     print()
-    print("Context Limits:")
-    print("   OpenAI GPT-4 Turbo: 128K tokens")
-    print("   Anthropic Claude: 200K tokens")
-    print("   Google Gemini Pro: 30K tokens")
+    print("Context Limits (default models):")
+    print("   OpenAI gpt-5.6-luna: 1.05M tokens (922K input)")
+    print("   Anthropic claude-sonnet-5: 1M tokens")
+    print("   Google gemini-3.6-flash: 1M tokens")
 
 
 if __name__ == "__main__":

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/examples/getting_started/simple_demo.py

@@ -58,7 +58,7 @@ def demo_basic_functionality():
     try:
         adapter = OpenAIAdapter(
             api_key="demo-key",  # Demo key
-            model="gpt-4-turbo-preview",
+            model="gpt-5.6-luna",
             temperature=0.1,
         )
 

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/mcp_server/__init__.py

@@ -1,3 +1,3 @@
 """nlp2sql MCP Server - Expose nlp2sql as MCP tools."""
 
-__version__ = "0.2.0rc14"
+__version__ = "0.2.0rc15"

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/mcp_server/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nlp2sql-mcp"
-version = "0.2.0rc14"
+version = "0.2.0rc15"
 description = "MCP server for nlp2sql - Natural Language to SQL conversion"
 readme = "README.md"
 requires-python = ">=3.9"
@@ -21,4 +21,4 @@ build-backend = "hatchling.build"
 [tool.mcp]
 name = "nlp2sql"
 description = "Convert natural language queries to SQL using multiple AI providers"
-version = "0.2.0rc14"
+version = "0.2.0rc15"

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nlp2sql"
-version = "0.2.0rc14"
+version = "0.2.0rc15"
 description = "Enterprise-ready Natural Language to SQL converter with multi-provider support. Built for production scale (1000+ tables) with Clean Architecture."
 readme = "README.md"
 license = {text = "MIT"}
@@ -27,9 +27,9 @@ dependencies = [
     "pydantic-settings>=2.0.0",
     "sqlalchemy>=2.0.0",
     "psycopg2-binary>=2.9.0",
-    "openai>=1.0.0",
+    "openai>=1.58.0",
     "anthropic>=0.18.0",
-    "google-generativeai>=0.3.0",
+    "google-generativeai>=0.8.5",
     "boto3>=1.34.0",
     "azure-identity>=1.15.0",
     "numpy>=1.24.0",
@@ -38,7 +38,7 @@ dependencies = [
     "structlog>=24.1.0",
     "httpx>=0.25.0",
     "redis>=5.0.0",
-    "tiktoken>=0.5.0",
+    "tiktoken>=0.7.0",
     "faiss-cpu>=1.7.4",
     "pandas>=2.0.0",
     "greenlet>=3.2.3",
@@ -52,8 +52,8 @@ mysql = ["pymysql>=1.1.0", "mysql-connector-python>=8.0.0"]
 oracle = ["oracledb>=2.0.0"]
 mssql = ["pymssql>=2.2.0"]
 anthropic = ["anthropic>=0.18.0"]
-gemini = ["google-generativeai>=0.3.0"]
-all-providers = ["anthropic>=0.18.0", "google-generativeai>=0.3.0"]
+gemini = ["google-generativeai>=0.8.5"]
+all-providers = ["anthropic>=0.18.0", "google-generativeai>=0.8.5"]
 embeddings-local = ["sentence-transformers>=2.5.0"]
 embeddings-openai = []
 all-embeddings = ["sentence-transformers>=2.5.0"]

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/src/nlp2sql/__init__.py

@@ -46,7 +46,7 @@ from .ports.semantic_validator import SemanticValidatorPort
 from .schema.example_store import ExampleStore
 from .services.query_service import QueryGenerationService
 
-__version__ = "0.2.0rc14"
+__version__ = "0.2.0rc15"
 __author__ = "Luis Carbonel"
 __email__ = "devhighlevel@gmail.com"
 
@@ -227,8 +227,8 @@ def create_query_service(
         database_url: Database connection URL
         ai_provider: AI provider to use ('openai', 'anthropic', 'gemini', etc.)
         api_key: API key for the AI provider
-        model: Optional model name for the AI provider (e.g., 'gpt-4o', 'claude-sonnet-4-20250514',
-            'gemini-2.0-flash'). If None, uses the provider's default model.
+        model: Optional model name for the AI provider (e.g., 'gpt-5.6-luna', 'claude-sonnet-5',
+            'gemini-3.6-flash'). If None, uses the provider's default model.
         database_type: Type of database
         schema_filters: Optional filters to limit schema scope
         embedding_provider: Optional embedding provider instance
@@ -383,7 +383,7 @@ async def create_and_initialize_service(
         database_url: Database connection URL
         ai_provider: AI provider to use ('openai', 'anthropic', 'gemini', etc.)
         api_key: API key for the AI provider
-        model: Optional model name for the AI provider (e.g., 'gpt-4o', 'claude-sonnet-4-20250514').
+        model: Optional model name for the AI provider (e.g., 'gpt-5.6-luna', 'claude-sonnet-5').
             If None, uses the provider's default model.
         database_type: Type of database
         schema_filters: Optional filters to limit schema scope
@@ -401,7 +401,7 @@ async def create_and_initialize_service(
             "postgresql://user:pass@localhost/db",
             ai_provider="anthropic",
             api_key="your-api-key",
-            model="claude-sonnet-4-20250514"
+            model="claude-sonnet-5"
         )
         result = await service.generate_sql("Show all users")
     """
@@ -459,7 +459,7 @@ async def generate_sql_from_db(
         question: Natural language question
         ai_provider: AI provider to use (default: 'openai')
         api_key: API key for the AI provider
-        model: Optional model name for the AI provider (e.g., 'gpt-4o-mini').
+        model: Optional model name for the AI provider (e.g., 'gpt-5.6-luna').
             If None, uses the provider's default model.
         database_type: Type of database (default: POSTGRES)
         schema_filters: Optional filters to limit schema scope
@@ -480,7 +480,7 @@ async def generate_sql_from_db(
             "Show me all active users",
             ai_provider="openai",
             api_key="your-api-key",
-            model="gpt-4o-mini"
+            model="gpt-5.6-luna"
         )
     """
     service = await create_and_initialize_service(

{nlp2sql-0.2.0rc14 → nlp2sql-0.2.0rc15}/src/nlp2sql/adapters/anthropic_adapter.py

@@ -8,6 +8,7 @@ import structlog
 from tenacity import retry, stop_after_attempt, wait_exponential
 
 from ..config.settings import settings
+from ..core.provider_config import ProviderConfig
 from ..exceptions import ProviderException, TokenLimitException
 from ..ports.ai_provider import AIProviderPort, AIProviderType, QueryContext, QueryResponse
 from ..utils.helpers import first_not_none
@@ -15,11 +16,35 @@ from ..utils.semantic_prompt import format_semantic_context_lines, format_sql_in
 
 logger = structlog.get_logger()
 
+# Maximum input tokens per model. Legacy rows are kept so users who pin an
+# older (or retired) model keep an accurate budget.
+CONTEXT_LIMITS: Dict[str, int] = {
+    "claude-sonnet-5": 1_000_000,
+    "claude-sonnet-4-20250514": 200_000,
+    "claude-opus-4-20250514": 200_000,
+    "claude-3-7-sonnet-20250219": 200_000,
+    "claude-3-5-sonnet-20241022": 200_000,
+    "claude-3-5-haiku-20241022": 200_000,
+    "claude-3-opus-20240229": 200_000,
+    "claude-3-sonnet-20240229": 200_000,
+    "claude-3-haiku-20240307": 200_000,
+    "claude-2.1": 200_000,
+    "claude-2.0": 100_000,
+}
+# Every Claude model since 2.1 has at least a 200K window.
+DEFAULT_CONTEXT_LIMIT = 200_000
+
 
 class AnthropicAdapter(AIProviderPort):
-    """Anthropic Claude adapter for natural language to SQL generation."""
+    """Anthropic Claude adapter for natural language to SQL generation.
+
+    Sampling parameters (``temperature``, ``top_p``, ``top_k``) are never sent:
+    Claude 4.7+ models reject non-default values with HTTP 400 and the
+    ``anthropic`` SDK removes the keywords in 1.0. A ``temperature`` passed to
+    the constructor is kept for backward compatibility but not forwarded.
+    """
 
-    DEFAULT_MODEL = "claude-sonnet-4-20250514"
+    DEFAULT_MODEL = ProviderConfig.DEFAULT_MODELS["anthropic"]
     DEFAULT_MAX_TOKENS = 2000
     DEFAULT_TEMPERATURE = 0.1
 
@@ -37,6 +62,15 @@ class AnthropicAdapter(AIProviderPort):
         self.model = model or self.DEFAULT_MODEL
         self.max_tokens = first_not_none(max_tokens, self.DEFAULT_MAX_TOKENS)
         self.temperature = first_not_none(temperature, self.DEFAULT_TEMPERATURE)
+        self._max_context_size = self._resolve_context_limit(self.model)
+
+        if temperature is not None:
+            # INFO, not WARNING: the CLI always forwards its --temperature default.
+            logger.info(
+                "Anthropic adapter does not forward temperature (rejected by Claude 4.7+ models)",
+                model=self.model,
+                temperature=temperature,
+            )
 
         self.client = anthropic.AsyncAnthropic(api_key=self.api_key)
 
@@ -45,13 +79,36 @@ class AnthropicAdapter(AIProviderPort):
             provider="anthropic",
             model=self.model,
             max_tokens=self.max_tokens,
-            temperature=self.temperature,
+            context_limit=self._max_context_size,
         )
 
     @property
     def provider_type(self) -> AIProviderType:
         return AIProviderType.ANTHROPIC
 
+    @staticmethod
+    def _resolve_context_limit(model: str) -> int:
+        """Exact match, then longest model-family prefix, then a logged fallback."""
+        if model in CONTEXT_LIMITS:
+            return CONTEXT_LIMITS[model]
+        matches = [key for key in CONTEXT_LIMITS if model.startswith(key)]
+        if matches:
+            family = max(matches, key=len)
+            logger.debug(
+                "Resolved context limit by model family",
+                model=model,
+                family=family,
+                context_limit=CONTEXT_LIMITS[family],
+            )
+            return CONTEXT_LIMITS[family]
+        logger.warning(
+            "Unknown Anthropic model; using fallback context limit",
+            model=model,
+            context_limit=DEFAULT_CONTEXT_LIMIT,
+            hint="pin a model listed in anthropic_adapter.CONTEXT_LIMITS or add it",
+        )
+        return DEFAULT_CONTEXT_LIMIT
+
     def get_token_count(self, text: str) -> int:
         """Estimate token count for Claude models."""
         # Claude uses similar tokenization to GPT models
@@ -59,22 +116,10 @@ class AnthropicAdapter(AIProviderPort):
         return len(text) // 4
 
     def get_max_context_size(self) -> int:
-        """Get maximum context size for the model."""
-        context_limits = {
-            "claude-sonnet-4-20250514": 200000,
-            "claude-opus-4-20250514": 200000,
-            "claude-3-7-sonnet-20250219": 200000,
-            "claude-3-5-sonnet-20241022": 200000,
-            "claude-3-5-haiku-20241022": 200000,
-            "claude-3-opus-20240229": 200000,
-            "claude-3-sonnet-20240229": 200000,
-            "claude-3-haiku-20240307": 200000,
-            "claude-2.1": 200000,
-            "claude-2.0": 100000,
-        }
-        return context_limits.get(self.model, 100000)
-
-    @retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=4, max=10))
+        """Get maximum input context size for the model."""
+        return self._max_context_size
+
+    @retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=4, max=10), reraise=True)
     async def generate_query(self, context: QueryContext) -> QueryResponse:
         """Generate SQL query using Claude."""
         try:
@@ -88,11 +133,10 @@ class AnthropicAdapter(AIProviderPort):
             # Create messages
             messages = [{"role": "user", "content": prompt}]
 
-            # Call Claude API
+            # Call Claude API (no sampling parameters: see class docstring)
             response = await self.client.messages.create(
                 model=self.model,
                 max_tokens=context.max_tokens,
-                temperature=context.temperature,
                 system=system_prompt,
                 messages=messages,
             )
@@ -310,7 +354,6 @@ Check for:
             message = await self.client.messages.create(
                 model=self.model,
                 max_tokens=1000,
-                temperature=0,
                 system="You are a SQL validation expert. Analyze the given SQL query and provide validation results in JSON format only.",
                 messages=[{"role": "user", "content": validation_prompt}],
             )