grasp_agents 0.5.8__tar.gz → 0.5.9__tar.gz

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: grasp_agents
-Version: 0.5.8
+Version: 0.5.9
 Summary: Grasp Agents Library
 License-File: LICENSE.md
 Requires-Python: <4,>=3.11.4
@@ -37,31 +37,30 @@ Description-Content-Type: text/markdown
 
 ## Features
 
-- Clean formulation of agents as generic entities over:
-  - I/O schemas
-  - Memory
-  - Shared context
+- Clean formulation of agents as generic entities over I/O schemas and shared context.
 - Transparent implementation of common agentic patterns:
-  - Single-agent loops with an optional "ReAct mode" to enforce reasoning between the tool calls
+  - Single-agent loops
   - Workflows (static communication topology), including loops
   - Agents-as-tools for task delegation
   - Freeform A2A communication via the in-process actor model
-- Parallel processing with flexible retries and rate limiting
-- Simple logging and usage/cost tracking
+- Built-in parallel processing with flexible retries and rate limiting.
+- Support for all popular API providers via LiteLLM.
+- Granular event streaming with separate events for standard outputs, thinking, and tool calls.
+- Callbacks via decorators or subclassing for straightforward customisation of agentic loops and context management.
 
 ## Project Structure
 
-- `processor.py`, `comm_processor.py`, `llm_agent.py`: Core processor and agent class implementations.
-- `packet.py`, `packet_pool.py`: Communication management.
+- `processors/`, `llm_agent.py`: Core processor and agent class implementations.
+- `packet.py`, `packet_pool.py`, `runner.py`: Communication management.
 - `llm_policy_executor.py`: LLM actions and tool call loops.
 - `prompt_builder.py`: Tools for constructing prompts.
 - `workflow/`: Modules for defining and managing static agent workflows.
 - `llm.py`, `cloud_llm.py`: LLM integration and base LLM functionalities.
 - `openai/`: Modules specific to OpenAI API integration.
-- `memory.py`, `llm_agent_memory.py`: Memory management.
+- `litellm/`: Modules specific to LiteLLM integration.
+- `memory.py`, `llm_agent_memory.py`: Basic agent memory management.
 - `run_context.py`: Shared context management for agent runs.
 - `usage_tracker.py`: Tracking of API usage and costs.
-- `costs_dict.yaml`: Dictionary for cost tracking (update if needed).
 - `rate_limiting/`: Basic rate limiting tools.
 
 ## Quickstart & Installation Variants (UV Package manager)
@@ -190,7 +189,7 @@ teacher = LLMAgent[None, Problem, None](
 )
 
 async def main():
-    ctx = RunContext[None](print_messages=True)
+    ctx = RunContext[None](log_messages=True)
     out = await teacher.run("start", ctx=ctx)
     print(out.payloads[0])
     print(ctx.usage_tracker.total_usage)

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/README.md

@@ -20,31 +20,30 @@
 
 ## Features
 
-- Clean formulation of agents as generic entities over:
-  - I/O schemas
-  - Memory
-  - Shared context
+- Clean formulation of agents as generic entities over I/O schemas and shared context.
 - Transparent implementation of common agentic patterns:
-  - Single-agent loops with an optional "ReAct mode" to enforce reasoning between the tool calls
+  - Single-agent loops
   - Workflows (static communication topology), including loops
   - Agents-as-tools for task delegation
   - Freeform A2A communication via the in-process actor model
-- Parallel processing with flexible retries and rate limiting
-- Simple logging and usage/cost tracking
+- Built-in parallel processing with flexible retries and rate limiting.
+- Support for all popular API providers via LiteLLM.
+- Granular event streaming with separate events for standard outputs, thinking, and tool calls.
+- Callbacks via decorators or subclassing for straightforward customisation of agentic loops and context management.
 
 ## Project Structure
 
-- `processor.py`, `comm_processor.py`, `llm_agent.py`: Core processor and agent class implementations.
-- `packet.py`, `packet_pool.py`: Communication management.
+- `processors/`, `llm_agent.py`: Core processor and agent class implementations.
+- `packet.py`, `packet_pool.py`, `runner.py`: Communication management.
 - `llm_policy_executor.py`: LLM actions and tool call loops.
 - `prompt_builder.py`: Tools for constructing prompts.
 - `workflow/`: Modules for defining and managing static agent workflows.
 - `llm.py`, `cloud_llm.py`: LLM integration and base LLM functionalities.
 - `openai/`: Modules specific to OpenAI API integration.
-- `memory.py`, `llm_agent_memory.py`: Memory management.
+- `litellm/`: Modules specific to LiteLLM integration.
+- `memory.py`, `llm_agent_memory.py`: Basic agent memory management.
 - `run_context.py`: Shared context management for agent runs.
 - `usage_tracker.py`: Tracking of API usage and costs.
-- `costs_dict.yaml`: Dictionary for cost tracking (update if needed).
 - `rate_limiting/`: Basic rate limiting tools.
 
 ## Quickstart & Installation Variants (UV Package manager)
@@ -173,7 +172,7 @@ teacher = LLMAgent[None, Problem, None](
 )
 
 async def main():
-    ctx = RunContext[None](print_messages=True)
+    ctx = RunContext[None](log_messages=True)
     out = await teacher.run("start", ctx=ctx)
     print(out.payloads[0])
     print(ctx.usage_tracker.total_usage)

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "grasp_agents"
-version = "0.5.8"
+version = "0.5.9"
 description = "Grasp Agents Library"
 readme = "README.md"
 requires-python = ">=3.11.4,<4"

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/src/grasp_agents/litellm/lite_llm.py

@@ -1,5 +1,7 @@
 import logging
+from collections import defaultdict
 from collections.abc import AsyncIterator, Mapping
+from copy import deepcopy
 from typing import Any, cast
 
 import litellm
@@ -90,10 +92,19 @@ class LiteLLM(CloudLLM[LiteLLMSettings, LiteLLMConverters]):
                 "was specified. Please provide a valid API provider or use a different "
                 "model."
             )
+
+        if llm_settings is not None:
+            stream_options = llm_settings.get("stream_options") or {}
+            stream_options["include_usage"] = True
+            _llm_settings = deepcopy(llm_settings)
+            _llm_settings["stream_options"] = stream_options
+        else:
+            _llm_settings = LiteLLMSettings(stream_options={"include_usage": True})
+
         super().__init__(
             model_name=model_name,
             model_id=model_id,
-            llm_settings=llm_settings,
+            llm_settings=_llm_settings,
             converters=LiteLLMConverters(),
             tools=tools,
             response_schema=response_schema,
@@ -192,7 +203,17 @@ class LiteLLM(CloudLLM[LiteLLMSettings, LiteLLMConverters]):
         )
         stream = cast("CustomStreamWrapper", stream)
 
+        tc_indices: dict[int, set[int]] = defaultdict(set)
+
         async for completion_chunk in stream:
+            # Fix tool call indices to be unique within each choice
+            for n, choice in enumerate(completion_chunk.choices):
+                for tc in choice.delta.tool_calls or []:
+                    # Tool call ID is not None only when it is a new tool call
+                    if tc.id and tc.index in tc_indices[n]:
+                        tc.index = max(tc_indices[n]) + 1
+                    tc_indices[n].add(tc.index)
+
             yield completion_chunk
 
     def combine_completion_chunks(

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/src/grasp_agents/llm.py

@@ -25,7 +25,7 @@ from .typing.events import (
     AnnotationsEndEvent,
     AnnotationsStartEvent,
     CompletionChunkEvent,
-    CompletionEndEvent,
+    # CompletionEndEvent,
     CompletionEvent,
     CompletionStartEvent,
     LLMStateChangeEvent,
@@ -196,7 +196,9 @@ class LLM(ABC, Generic[SettingsT_co, ConvertT_co]):
         annotations_op_evt: AnnotationsChunkEvent | None = None
         tool_calls_op_evt: ToolCallChunkEvent | None = None
 
-        def _close_open_events() -> list[LLMStateChangeEvent[Any]]:
+        def _close_open_events(
+            _event: CompletionChunkEvent[CompletionChunk] | None = None,
+        ) -> list[LLMStateChangeEvent[Any]]:
             nonlocal \
                 chunk_op_evt, \
                 thinking_op_evt, \
@@ -206,26 +208,21 @@ class LLM(ABC, Generic[SettingsT_co, ConvertT_co]):
 
             events: list[LLMStateChangeEvent[Any]] = []
 
-            if tool_calls_op_evt:
+            if not isinstance(_event, ThinkingChunkEvent) and thinking_op_evt:
+                events.append(ThinkingEndEvent.from_chunk_event(thinking_op_evt))
+                thinking_op_evt = None
+
+            if not isinstance(_event, ToolCallChunkEvent) and tool_calls_op_evt:
                 events.append(ToolCallEndEvent.from_chunk_event(tool_calls_op_evt))
+                tool_calls_op_evt = None
 
-            if response_op_evt:
+            if not isinstance(_event, ResponseChunkEvent) and response_op_evt:
                 events.append(ResponseEndEvent.from_chunk_event(response_op_evt))
+                response_op_evt = None
 
-            if thinking_op_evt:
-                events.append(ThinkingEndEvent.from_chunk_event(thinking_op_evt))
-
-            if annotations_op_evt:
+            if not isinstance(_event, AnnotationsChunkEvent) and annotations_op_evt:
                 events.append(AnnotationsEndEvent.from_chunk_event(annotations_op_evt))
-
-            if chunk_op_evt:
-                events.append(CompletionEndEvent.from_chunk_event(chunk_op_evt))
-
-            chunk_op_evt = None
-            thinking_op_evt = None
-            tool_calls_op_evt = None
-            response_op_evt = None
-            annotations_op_evt = None
+                annotations_op_evt = None
 
             return events
 
@@ -252,14 +249,14 @@ class LLM(ABC, Generic[SettingsT_co, ConvertT_co]):
                 sub_events = event.split_into_specialized()
 
                 for sub_event in sub_events:
+                    for close_event in _close_open_events(sub_event):
+                        yield close_event
+
                     if isinstance(sub_event, ThinkingChunkEvent):
                         if not thinking_op_evt:
                             thinking_op_evt = sub_event
                             yield ThinkingStartEvent.from_chunk_event(sub_event)
                         yield sub_event
-                    elif thinking_op_evt:
-                        yield ThinkingEndEvent.from_chunk_event(thinking_op_evt)
-                        thinking_op_evt = None
 
                     if isinstance(sub_event, ToolCallChunkEvent):
                         tc = sub_event.data.tool_call
@@ -273,27 +270,18 @@ class LLM(ABC, Generic[SettingsT_co, ConvertT_co]):
                             tool_calls_op_evt = sub_event
                             yield ToolCallStartEvent.from_chunk_event(sub_event)
                         yield sub_event
-                    elif tool_calls_op_evt:
-                        yield ToolCallEndEvent.from_chunk_event(tool_calls_op_evt)
-                        tool_calls_op_evt = None
 
                     if isinstance(sub_event, ResponseChunkEvent):
                         if not response_op_evt:
                             response_op_evt = sub_event
                             yield ResponseStartEvent.from_chunk_event(sub_event)
                         yield sub_event
-                    elif response_op_evt:
-                        yield ResponseEndEvent.from_chunk_event(response_op_evt)
-                        response_op_evt = None
 
                     if isinstance(sub_event, AnnotationsChunkEvent):
                         if not annotations_op_evt:
                             annotations_op_evt = sub_event
                             yield AnnotationsStartEvent.from_chunk_event(sub_event)
                         yield sub_event
-                    elif annotations_op_evt:
-                        yield AnnotationsEndEvent.from_chunk_event(annotations_op_evt)
-                        annotations_op_evt = None
 
                 prev_completion_id = chunk.id
 

{grasp_agents-0.5.8 → grasp_agents-0.5.9}/src/grasp_agents/openai/openai_llm.py

@@ -127,8 +127,8 @@ class OpenAILLM(CloudLLM[OpenAILLMSettings, OpenAIConverters]):
             provider_name, provider_model_name = model_name_parts
             if provider_name not in compat_providers_map:
                 raise ValueError(
-                    f"OpenAI compatible API provider '{provider_name}' "
-                    "is not supported. Supported providers are: "
+                    f"API provider '{provider_name}' is not a supported OpenAI "
+                    f"compatible provider. Supported providers are: "
                     f"{', '.join(compat_providers_map.keys())}"
                 )
             api_provider = compat_providers_map[provider_name]
@@ -138,10 +138,18 @@ class OpenAILLM(CloudLLM[OpenAILLMSettings, OpenAIConverters]):
                 "you must provide an 'api_provider' argument."
             )
 
+        if llm_settings is not None:
+            stream_options = llm_settings.get("stream_options") or {}
+            stream_options["include_usage"] = True
+            _llm_settings = deepcopy(llm_settings)
+            _llm_settings["stream_options"] = stream_options
+        else:
+            _llm_settings = OpenAILLMSettings(stream_options={"include_usage": True})
+
         super().__init__(
             model_name=provider_model_name,
             model_id=model_id,
-            llm_settings=llm_settings,
+            llm_settings=_llm_settings,
             converters=OpenAIConverters(),
             tools=tools,
             response_schema=response_schema,