selectools 0.17.3__tar.gz → 0.17.4__tar.gz

{selectools-0.17.3/src/selectools.egg-info → selectools-0.17.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: selectools
-Version: 0.17.3
+Version: 0.17.4
 Summary: Production-ready AI agents with tool calling, structured output, execution traces, and RAG. Provider-agnostic (OpenAI, Anthropic, Gemini, Ollama) with fallback chains, batch processing, tool policies, streaming, caching, and cost tracking.
 Author-email: John Nichev <johnnichev@gmail.com>
 Maintainer-email: NichevLabs <support@nichevlabs.com>

{selectools-0.17.3 → selectools-0.17.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "selectools"
-version = "0.17.3"
+version = "0.17.4"
 description = "Production-ready AI agents with tool calling, structured output, execution traces, and RAG. Provider-agnostic (OpenAI, Anthropic, Gemini, Ollama) with fallback chains, batch processing, tool policies, streaming, caching, and cost tracking."
 readme = "README.md"
 requires-python = ">=3.9"

{selectools-0.17.3 → selectools-0.17.4}/src/selectools/__init__.py

@@ -1,6 +1,6 @@
 """Public exports for the selectools package."""
 
-__version__ = "0.17.3"
+__version__ = "0.17.4"
 
 # Import submodules (lazy loading for optional dependencies)
 from . import embeddings, evals, guardrails, models, rag, toolbox
@@ -34,7 +34,13 @@ from .guardrails import (
     TopicGuardrail,
     ToxicityGuardrail,
 )
-from .knowledge import KnowledgeMemory
+from .knowledge import (
+    FileKnowledgeStore,
+    KnowledgeEntry,
+    KnowledgeMemory,
+    KnowledgeStore,
+    SQLiteKnowledgeStore,
+)
 from .knowledge_graph import (
     InMemoryTripleStore,
     KnowledgeGraphMemory,
@@ -73,6 +79,7 @@ from .sessions import (
     SQLiteSessionStore,
 )
 from .structured import ResponseFormat
+from .token_estimation import TokenEstimate, estimate_run_tokens, estimate_tokens
 from .tools import Tool, ToolParameter, ToolRegistry, tool
 from .trace import AgentTrace, StepType, TraceStep
 from .types import AgentResult, Message, Role, ToolCall
@@ -177,6 +184,14 @@ __all__ = [
     "EntityMemory",
     # Knowledge Memory
     "KnowledgeMemory",
+    "KnowledgeEntry",
+    "KnowledgeStore",
+    "FileKnowledgeStore",
+    "SQLiteKnowledgeStore",
+    # Token estimation
+    "TokenEstimate",
+    "estimate_tokens",
+    "estimate_run_tokens",
     # Knowledge Graph
     "Triple",
     "TripleStore",

{selectools-0.17.3 → selectools-0.17.4}/src/selectools/agent/_provider_caller.py

@@ -37,7 +37,7 @@ class _ProviderCallerMixin:
             self.config.stream and getattr(self.provider, "supports_streaming", False)
         ):
             cache_key = CacheKeyBuilder.build(
-                model=self.config.model,
+                model=self._effective_model,
                 system_prompt=self._system_prompt,
                 messages=self._history,
                 tools=self.tools,
@@ -53,7 +53,7 @@ class _ProviderCallerMixin:
                         "on_llm_start",
                         run_id,
                         self._history,
-                        self.config.model,
+                        self._effective_model,
                         self._system_prompt,
                     )
                     self._notify_observers(
@@ -65,7 +65,7 @@ class _ProviderCallerMixin:
                     self._notify_observers(
                         "on_cache_hit",
                         run_id,
-                        self.config.model,
+                        self._effective_model,
                         cached_msg.content or "",
                     )
                     self._notify_observers("on_usage", run_id, cached_usage)
@@ -76,8 +76,8 @@ class _ProviderCallerMixin:
                         TraceStep(
                             type=StepType.CACHE_HIT,
                             duration_ms=(time.time() - call_start) * 1000,
-                            model=self.config.model,
-                            summary=f"Cache hit: {self.config.model}",
+                            model=self._effective_model,
+                            summary=f"Cache hit: {self._effective_model}",
                         )
                     )
                 return cached_msg
@@ -93,7 +93,7 @@ class _ProviderCallerMixin:
                         "on_llm_start",
                         run_id,
                         self._history,
-                        self.config.model,
+                        self._effective_model,
                         self._system_prompt,
                     )
 
@@ -104,7 +104,7 @@ class _ProviderCallerMixin:
                     return Message(role=Role.ASSISTANT, content=response_text)
 
                 response_msg, usage_stats = self.provider.complete(
-                    model=self.config.model,
+                    model=self._effective_model,
                     system_prompt=self._system_prompt,
                     messages=self._history,
                     tools=self.tools,
@@ -144,11 +144,11 @@ class _ProviderCallerMixin:
                         TraceStep(
                             type=StepType.LLM_CALL,
                             duration_ms=(time.time() - call_start) * 1000,
-                            model=self.config.model,
+                            model=self._effective_model,
                             prompt_tokens=usage_stats.prompt_tokens,
                             completion_tokens=usage_stats.completion_tokens,
                             cost_usd=usage_stats.cost_usd,
-                            summary=f"{self.config.model} → {len(response_text)} chars",
+                            summary=f"{self._effective_model} → {len(response_text)} chars",
                         )
                     )
                 return response_msg
@@ -185,7 +185,7 @@ class _ProviderCallerMixin:
                 TraceStep(
                     type=StepType.LLM_CALL,
                     duration_ms=(time.time() - call_start) * 1000,
-                    model=self.config.model,
+                    model=self._effective_model,
                     error=last_error,
                     summary=f"Provider error: {last_error}",
                 )
@@ -200,7 +200,7 @@ class _ProviderCallerMixin:
 
         aggregated: List[str] = []
         for chunk in self.provider.stream(
-            model=self.config.model,
+            model=self._effective_model,
             system_prompt=self._system_prompt,
             messages=self._history,
             tools=self.tools,
@@ -233,7 +233,7 @@ class _ProviderCallerMixin:
             self.config.stream and getattr(self.provider, "supports_streaming", False)
         ):
             cache_key = CacheKeyBuilder.build(
-                model=self.config.model,
+                model=self._effective_model,
                 system_prompt=self._system_prompt,
                 messages=self._history,
                 tools=self.tools,
@@ -249,7 +249,7 @@ class _ProviderCallerMixin:
                         "on_llm_start",
                         run_id,
                         self._history,
-                        self.config.model,
+                        self._effective_model,
                         self._system_prompt,
                     )
                     self._notify_observers(
@@ -261,7 +261,7 @@ class _ProviderCallerMixin:
                     self._notify_observers(
                         "on_cache_hit",
                         run_id,
-                        self.config.model,
+                        self._effective_model,
                         cached_msg.content or "",
                     )
                     self._notify_observers("on_usage", run_id, cached_usage)
@@ -272,8 +272,8 @@ class _ProviderCallerMixin:
                         TraceStep(
                             type=StepType.CACHE_HIT,
                             duration_ms=(time.time() - call_start) * 1000,
-                            model=self.config.model,
-                            summary=f"Cache hit: {self.config.model}",
+                            model=self._effective_model,
+                            summary=f"Cache hit: {self._effective_model}",
                         )
                     )
                 return cached_msg
@@ -289,14 +289,14 @@ class _ProviderCallerMixin:
                         "on_llm_start",
                         run_id,
                         self._history,
-                        self.config.model,
+                        self._effective_model,
                         self._system_prompt,
                     )
                     await self._anotify_observers(
                         "on_llm_start",
                         run_id,
                         self._history,
-                        self.config.model,
+                        self._effective_model,
                         self._system_prompt,
                     )
 
@@ -312,7 +312,7 @@ class _ProviderCallerMixin:
                     self.provider, "supports_async", False
                 ):
                     response_msg, usage_stats = await self.provider.acomplete(
-                        model=self.config.model,
+                        model=self._effective_model,
                         system_prompt=self._system_prompt,
                         messages=self._history,
                         tools=self.tools,
@@ -328,7 +328,7 @@ class _ProviderCallerMixin:
                         response_msg, usage_stats = await loop.run_in_executor(
                             executor,
                             lambda: self.provider.complete(
-                                model=self.config.model,
+                                model=self._effective_model,
                                 system_prompt=self._system_prompt,
                                 messages=self._history,
                                 tools=self.tools,
@@ -371,11 +371,11 @@ class _ProviderCallerMixin:
                         TraceStep(
                             type=StepType.LLM_CALL,
                             duration_ms=(time.time() - call_start) * 1000,
-                            model=self.config.model,
+                            model=self._effective_model,
                             prompt_tokens=usage_stats.prompt_tokens,
                             completion_tokens=usage_stats.completion_tokens,
                             cost_usd=usage_stats.cost_usd,
-                            summary=f"{self.config.model} → {len(response_text)} chars",
+                            summary=f"{self._effective_model} → {len(response_text)} chars",
                         )
                     )
                 return response_msg
@@ -412,7 +412,7 @@ class _ProviderCallerMixin:
                 TraceStep(
                     type=StepType.LLM_CALL,
                     duration_ms=(time.time() - call_start) * 1000,
-                    model=self.config.model,
+                    model=self._effective_model,
                     error=last_error,
                     summary=f"Provider error: {last_error}",
                 )
@@ -430,7 +430,7 @@ class _ProviderCallerMixin:
 
         if hasattr(self.provider, "astream") and getattr(self.provider, "supports_async", False):
             stream = self.provider.astream(  # type: ignore[attr-defined]
-                model=self.config.model,
+                model=self._effective_model,
                 system_prompt=self._system_prompt,
                 messages=self._history,
                 tools=self.tools,
@@ -445,7 +445,7 @@ class _ProviderCallerMixin:
                         stream_handler(chunk)
         else:
             for chunk in self.provider.stream(
-                model=self.config.model,
+                model=self._effective_model,
                 system_prompt=self._system_prompt,
                 messages=self._history,
                 tools=self.tools,

{selectools-0.17.3 → selectools-0.17.4}/src/selectools/agent/config.py

@@ -18,6 +18,7 @@ if TYPE_CHECKING:
     from ..policy import ToolPolicy
     from ..providers.base import Provider
     from ..sessions import SessionStore
+    from ..usage import AgentUsage
 
 # Hook type definitions
 HookCallable = Callable[..., None]
@@ -157,6 +158,7 @@ class AgentConfig:
     knowledge_graph: Optional[KnowledgeGraphMemory] = None
     knowledge_memory: Optional[KnowledgeMemory] = None
     stop_condition: Optional[Callable[[str, str], bool]] = None
+    model_selector: Optional[Callable[[int, List, "AgentUsage"], str]] = None
     max_total_tokens: Optional[int] = None
     max_cost_usd: Optional[float] = None
     cancellation_token: Optional[CancellationToken] = None

{selectools-0.17.3 → selectools-0.17.4}/src/selectools/agent/core.py

@@ -189,6 +189,15 @@ class Agent(_ToolExecutorMixin, _ProviderCallerMixin, _LifecycleMixin, _MemoryMa
         """Return the agent's name from config."""
         return self.config.name
 
+    @property
+    def _effective_model(self) -> str:
+        """The model to use for the current iteration.
+
+        Returns the model set by ``model_selector`` if active, otherwise
+        falls back to ``self.config.model``.
+        """
+        return getattr(self, "_current_model", None) or self.config.model
+
     def __call__(
         self,
         messages: Union[str, List[Message]],
@@ -313,6 +322,7 @@ class Agent(_ToolExecutorMixin, _ProviderCallerMixin, _LifecycleMixin, _MemoryMa
         wires observers, runs input guardrails, loads memory/session, injects
         knowledge context, and returns a _RunContext carrying all state.
         """
+        self._current_model: Optional[str] = None
         original_system_prompt = self._system_prompt
         if response_format is not None:
             schema = schema_from_response_format(response_format)
@@ -867,6 +877,18 @@ class Agent(_ToolExecutorMixin, _ProviderCallerMixin, _LifecycleMixin, _MemoryMa
                 if budget_msg:
                     return self._build_budget_exceeded_result(ctx, budget_msg)
 
+                # Model selection (R10)
+                if self.config.model_selector:
+                    selected = self.config.model_selector(
+                        ctx.iteration, ctx.all_tool_calls, self.usage
+                    )
+                    if selected != self._effective_model:
+                        old_model = self._effective_model
+                        self._current_model = selected
+                        self._notify_observers(
+                            "on_model_switch", ctx.run_id, ctx.iteration, old_model, selected
+                        )
+
                 self._notify_observers(
                     "on_iteration_start", ctx.run_id, ctx.iteration, self._history
                 )
@@ -1053,6 +1075,21 @@ class Agent(_ToolExecutorMixin, _ProviderCallerMixin, _LifecycleMixin, _MemoryMa
                     yield StreamChunk(content=budget_msg)
                     return
 
+                # Model selection (R10)
+                if self.config.model_selector:
+                    selected = self.config.model_selector(
+                        ctx.iteration, ctx.all_tool_calls, self.usage
+                    )
+                    if selected != self._effective_model:
+                        old_model = self._effective_model
+                        self._current_model = selected
+                        self._notify_observers(
+                            "on_model_switch", ctx.run_id, ctx.iteration, old_model, selected
+                        )
+                        await self._anotify_observers(
+                            "on_model_switch", ctx.run_id, ctx.iteration, old_model, selected
+                        )
+
                 self._notify_observers(
                     "on_iteration_start", ctx.run_id, ctx.iteration, self._history
                 )
@@ -1369,6 +1406,21 @@ class Agent(_ToolExecutorMixin, _ProviderCallerMixin, _LifecycleMixin, _MemoryMa
                 if budget_msg:
                     return self._build_budget_exceeded_result(ctx, budget_msg)
 
+                # Model selection (R10)
+                if self.config.model_selector:
+                    selected = self.config.model_selector(
+                        ctx.iteration, ctx.all_tool_calls, self.usage
+                    )
+                    if selected != self._effective_model:
+                        old_model = self._effective_model
+                        self._current_model = selected
+                        self._notify_observers(
+                            "on_model_switch", ctx.run_id, ctx.iteration, old_model, selected
+                        )
+                        await self._anotify_observers(
+                            "on_model_switch", ctx.run_id, ctx.iteration, old_model, selected
+                        )
+
                 self._notify_observers(
                     "on_iteration_start", ctx.run_id, ctx.iteration, self._history
                 )