Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
Lorenze/fix/file input not working reliably (#6020)
Lorenze Jay770d1b284f31
diff --git a/2605.26112v1.pdf b/2605.26112v1.pdfnew file mode 100644index 000000000..c46fed704Binary files /dev/null and b/2605.26112v1.pdf differdiff --git a/lib/crewai-files/src/crewai_files/formatting/api.py b/lib/crewai-files/src/crewai_files/formatting/api.pyindex b056f014e..c192cabba 100644--- a/lib/crewai-files/src/crewai_files/formatting/api.py+++ b/lib/crewai-files/src/crewai_files/formatting/api.py@@ -11,7 +11,10 @@ from crewai_files.formatting.anthropic import AnthropicFormatterfrom crewai_files.formatting.bedrock import BedrockFormatterfrom crewai_files.formatting.gemini import GeminiFormatterfrom crewai_files.formatting.openai import OpenAIFormatter, OpenAIResponsesFormatter-from crewai_files.processing.constraints import get_constraints_for_provider+from crewai_files.processing.constraints import (+ get_constraints_for_provider,+ uses_openai_responses_api,+)from crewai_files.processing.processor import FileProcessorfrom crewai_files.resolution.resolver import FileResolver, FileResolverConfigfrom crewai_files.uploaders.factory import ProviderType@@ -120,9 +123,11 @@ def format_multimodal_content(if not files:return content_blocks- constraints_key: str = provider_type- if api == "responses" and "openai" in provider_type.lower():- constraints_key = "openai_responses"+ constraints_key = (+ "openai_responses"+ if uses_openai_responses_api(provider_type, api)+ else provider_type+ )processor = FileProcessor(constraints=constraints_key)processed_files = processor.process_files(files)@@ -184,9 +189,11 @@ async def aformat_multimodal_content(if not files:return content_blocks- constraints_key: str = provider_type- if api == "responses" and "openai" in provider_type.lower():- constraints_key = "openai_responses"+ constraints_key = (+ "openai_responses"+ if uses_openai_responses_api(provider_type, api)+ else provider_type+ )processor = FileProcessor(constraints=constraints_key)processed_files = await processor.aprocess_files(files)diff --git a/lib/crewai-files/src/crewai_files/processing/constraints.py b/lib/crewai-files/src/crewai_files/processing/constraints.pyindex ba05827d5..798f313b1 100644--- a/lib/crewai-files/src/crewai_files/processing/constraints.py+++ b/lib/crewai-files/src/crewai_files/processing/constraints.py@@ -346,6 +346,20 @@ def get_constraints_for_provider(return None+def uses_openai_responses_api(provider: str, api: str | None = None) -> bool:+ """Return whether provider/API should use OpenAI Responses file support."""+ if api != "responses":+ return False++ provider_lower = provider.lower()+ return (+ "openai" in provider_lower+ or provider_lower == "gpt"+ or provider_lower.startswith("gpt-")+ or "/gpt-" in provider_lower+ )++def get_supported_content_types(provider: str, api: str | None = None) -> list[str]:"""Get supported MIME type prefixes for a provider.@@ -356,9 +370,9 @@ def get_supported_content_types(provider: str, api: str | None = None) -> list[sReturns:List of supported MIME type prefixes (e.g., ["image/", "application/pdf"])."""- lookup_key = provider- if api == "responses" and "openai" in provider.lower():- lookup_key = "openai_responses"+ lookup_key = (+ "openai_responses" if uses_openai_responses_api(provider, api) else provider+ )constraints = get_constraints_for_provider(lookup_key)if not constraints:diff --git a/lib/crewai-files/tests/processing/test_constraints.py b/lib/crewai-files/tests/processing/test_constraints.pyindex c90dc760e..cd3f64f12 100644--- a/lib/crewai-files/tests/processing/test_constraints.py+++ b/lib/crewai-files/tests/processing/test_constraints.py@@ -11,6 +11,7 @@ from crewai_files.processing.constraints import (ProviderConstraints,VideoConstraints,get_constraints_for_provider,+ get_supported_content_types,)import pytest@@ -70,6 +71,13 @@ class TestPDFConstraints:assert constraints.max_size_bytes == 1000assert constraints.max_pages is None+ @pytest.mark.parametrize("provider", ["openai", "gpt", "gpt-4o-mini"])+ def test_openai_responses_supports_pdf_for_gpt_aliases(self, provider):+ """OpenAI Responses PDF support applies to concrete GPT model names."""+ supported_types = get_supported_content_types(provider, api="responses")++ assert "application/pdf" in supported_types+class TestAudioConstraints:"""Tests for AudioConstraints dataclass."""diff --git a/lib/crewai/src/crewai/experimental/agent_executor.py b/lib/crewai/src/crewai/experimental/agent_executor.pyindex 81b431a24..3cc9cdd7b 100644--- a/lib/crewai/src/crewai/experimental/agent_executor.py+++ b/lib/crewai/src/crewai/experimental/agent_executor.py@@ -93,6 +93,7 @@ from crewai.utilities.agent_utils import (track_delegation_if_needed,)from crewai.utilities.constants import TRAINING_DATA_FILE+from crewai.utilities.file_store import aget_all_files, get_all_filesfrom crewai.utilities.i18n import I18N_DEFAULTfrom crewai.utilities.planning_types import (PlanStep,@@ -2771,7 +2772,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):mark_cache_breakpoint(format_message_for_llm(user_prompt)))- self._inject_files_from_inputs(inputs)+ await self._ainject_files_from_inputs(inputs)self.state.ask_for_human_input = bool(inputs.get("ask_for_human_input", False)@@ -2982,12 +2983,42 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):training_handler.save(training_data)def _inject_files_from_inputs(self, inputs: dict[str, Any]) -> None:- """Inject files from inputs into the last user message.+ """Inject files into the last user message.Args:inputs: Input dictionary that may contain a 'files' key."""- files = inputs.get("files")+ files: dict[str, Any] = {}++ if self.crew and self.task:+ stored_files = get_all_files(self.crew.id, self.task.id)+ if stored_files:+ files.update(stored_files)++ if inputs.get("files"):+ files.update(inputs["files"])++ if not files:+ return++ for i in range(len(self.state.messages) - 1, -1, -1):+ msg = self.state.messages[i]+ if msg.get("role") == "user":+ msg["files"] = files+ break++ async def _ainject_files_from_inputs(self, inputs: dict[str, Any]) -> None:+ """Async inject files into the last user message."""+ files: dict[str, Any] = {}++ if self.crew and self.task:+ stored_files = await aget_all_files(self.crew.id, self.task.id)+ if stored_files:+ files.update(stored_files)++ if inputs.get("files"):+ files.update(inputs["files"])+if not files:returndiff --git a/lib/crewai/src/crewai/tools/agent_tools/read_file_tool.py b/lib/crewai/src/crewai/tools/agent_tools/read_file_tool.pyindex e41d5390d..048552748 100644--- a/lib/crewai/src/crewai/tools/agent_tools/read_file_tool.py+++ b/lib/crewai/src/crewai/tools/agent_tools/read_file_tool.py@@ -3,6 +3,7 @@from __future__ import annotationsimport base64+from io import BytesIOfrom typing import TYPE_CHECKINGfrom pydantic import BaseModel, Field, PrivateAttr@@ -64,6 +65,9 @@ class ReadFileTool(BaseTool):content_type = file_input.content_typefilename = file_input.filename or file_name+ if content_type == "application/pdf":+ return self._read_pdf_text(content, filename)+text_types = ("text/","application/json",@@ -76,3 +80,22 @@ class ReadFileTool(BaseTool):encoded = base64.b64encode(content).decode("ascii")return f"[Binary file: {filename} ({content_type})]\nBase64: {encoded}"++ def _read_pdf_text(self, content: bytes, filename: str) -> str:+ """Extract text from a PDF instead of returning base64."""+ try:+ from pypdf import PdfReader+ except ImportError:+ encoded = base64.b64encode(content).decode("ascii")+ return f"[Binary file: {filename} (application/pdf)]\nBase64: {encoded}"++ try:+ reader = PdfReader(BytesIO(content))+ page_text = [text for page in reader.pages if (text := page.extract_text())]+ except Exception as exc:+ return f"Unable to extract text from PDF '{filename}': {exc}"++ if not page_text:+ return f"[PDF file with no extractable text: {filename}]"++ return "\n\n".join(page_text)diff --git a/lib/crewai/tests/agents/test_agent_executor.py b/lib/crewai/tests/agents/test_agent_executor.pyindex c7f2d0107..5868a7ce2 100644--- a/lib/crewai/tests/agents/test_agent_executor.py+++ b/lib/crewai/tests/agents/test_agent_executor.py@@ -7,9 +7,11 @@ flow methods, routing logic, and error handling.from __future__ import annotationsimport asyncio+from types import SimpleNamespaceimport timefrom typing import Anyfrom unittest.mock import AsyncMock, Mock, patch+from uuid import uuid4import pytestfrom pydantic import BaseModel@@ -64,6 +66,8 @@ from crewai.events.types.tool_usage_events import (from crewai.tools.tool_types import ToolResultfrom crewai.utilities.step_execution_context import StepExecutionContextfrom crewai.utilities.planning_types import TodoItem+from crewai.utilities.file_store import clear_files, clear_task_files, store_files+from crewai_files import TextFileclass TestAgentExecutorState:"""Test AgentExecutorState Pydantic model."""@@ -112,6 +116,58 @@ class TestAgentExecutor:class StructuredResult(BaseModel):value: str+ def test_inject_files_from_crew_task_store(self):+ """Crew-level input_files should attach to the LLM user message."""+ crew_id = uuid4()+ task_id = uuid4()+ stored_file = TextFile(source=b"stored content")+ executor = _build_executor(+ crew=SimpleNamespace(id=crew_id),+ task=SimpleNamespace(id=task_id),+ )+ executor.state.messages = [{"role": "user", "content": "Analyze this file"}]++ try:+ store_files(crew_id, {"document": stored_file})+ executor._inject_files_from_inputs({})+ finally:+ clear_files(crew_id)+ clear_task_files(task_id)++ assert executor.state.messages[0]["files"] == {"document": stored_file}++ @pytest.mark.asyncio+ async def test_ainject_files_from_crew_task_store_uses_async_store(self):+ """Async file injection should not call the sync file store helper."""+ crew_id = uuid4()+ task_id = uuid4()+ stored_file = TextFile(source=b"stored content")+ local_file = TextFile(source=b"local content")+ inputs = {"files": {"local": local_file}}+ executor = _build_executor(+ crew=SimpleNamespace(id=crew_id),+ task=SimpleNamespace(id=task_id),+ )+ executor.state.messages = [{"role": "user", "content": "Analyze this file"}]++ with (+ patch(+ "crewai.experimental.agent_executor.aget_all_files",+ new=AsyncMock(return_value={"document": stored_file}),+ ) as async_get_files,+ patch(+ "crewai.experimental.agent_executor.get_all_files",+ side_effect=AssertionError("sync file store should not be called"),+ ),+ ):+ await executor._ainject_files_from_inputs(inputs)++ async_get_files.assert_awaited_once_with(crew_id, task_id)+ assert executor.state.messages[0]["files"] == {+ "document": stored_file,+ "local": local_file,+ }+@pytest.fixturedef mock_dependencies(self):"""Create mock dependencies for executor."""diff --git a/lib/crewai/tests/llms/test_multimodal.py b/lib/crewai/tests/llms/test_multimodal.pyindex ab851efca..18f37f50d 100644--- a/lib/crewai/tests/llms/test_multimodal.py+++ b/lib/crewai/tests/llms/test_multimodal.py@@ -108,6 +108,16 @@ class TestLiteLLMMultimodal:assert result == []+ def test_format_responses_pdf_with_concrete_gpt_model(self) -> None:+ """Test OpenAI Responses PDF support with an inferred GPT provider."""+ files = {"doc": PDFFile(source=MINIMAL_PDF)}++ result = format_multimodal_content(files, "gpt-4o-mini", api="responses")++ assert len(result) == 1+ assert result[0]["type"] == "input_file"+ assert result[0]["file_data"].startswith("data:application/pdf;base64,")+@pytest.mark.skipif(not HAS_ANTHROPIC, reason="Anthropic SDK not installed")class TestAnthropicMultimodal:@@ -370,4 +380,4 @@ class TestMultipleFilesFormatting:result = format_multimodal_content({}, llm.model)- assert result == []\ No newline at end of file+ assert result == []diff --git a/lib/crewai/tests/tools/agent_tools/test_read_file_tool.py b/lib/crewai/tests/tools/agent_tools/test_read_file_tool.pyindex de2be4cc9..455372399 100644--- a/lib/crewai/tests/tools/agent_tools/test_read_file_tool.py+++ b/lib/crewai/tests/tools/agent_tools/test_read_file_tool.py@@ -1,11 +1,20 @@"""Unit tests for ReadFileTool."""import base64+from pathlib import Pathfrom crewai.tools.agent_tools.read_file_tool import ReadFileToolfrom crewai_files import ImageFile, PDFFile, TextFile+TEST_FIXTURES_DIR = (+ Path(__file__).parent.parent.parent.parent.parent+ / "crewai-files"+ / "tests"+ / "fixtures"+)++class TestReadFileTool:"""Tests for ReadFileTool."""@@ -72,15 +81,15 @@ class TestReadFileTool:decoded = base64.b64decode(b64_part)assert decoded == png_bytes- def test_run_pdf_file_returns_base64(self) -> None:- """Test reading a PDF file returns base64 encoded content."""- pdf_bytes = b"%PDF-1.4 some content here"+ def test_run_pdf_file_returns_extracted_text(self) -> None:+ """Test reading a PDF file returns extracted text instead of base64."""+ pdf_bytes = (TEST_FIXTURES_DIR / "agents.pdf").read_bytes()self.tool.set_files({"doc.pdf": PDFFile(source=pdf_bytes)})result = self.tool._run(file_name="doc.pdf")- assert "[Binary file:" in result- assert "application/pdf" in result+ assert "Base64:" not in result+ assert "agents" in result.lower()def test_set_files_none(self) -> None:"""Test setting files to None."""diff --git a/scripts/age90_file_input_runner.py b/scripts/age90_file_input_runner.pynew file mode 100644index 000000000..3f97267da--- /dev/null+++ b/scripts/age90_file_input_runner.py@@ -0,0 +1,226 @@+# ruff: noqa: T201+"""Manual runner for AGE-90 PDF input handling.++Usage examples:+ uv run python scripts/age90_file_input_runner.py+ uv run python scripts/age90_file_input_runner.py --mode fallback+ uv run python scripts/age90_file_input_runner.py --mode payload --pdf ./sample_story.pdf+ uv run python scripts/age90_file_input_runner.py --mode kickoff --pdf ./sample_story.pdf+"""++from __future__ import annotations++import argparse+from collections.abc import Mapping, Sequence+from contextlib import nullcontext+import os+from pathlib import Path+from typing import Any+from unittest.mock import patch++from crewai_files import PDFFile, format_multimodal_content, get_supported_content_types+++ROOT = Path(__file__).resolve().parents[1]+DEFAULT_PDF = ROOT / "lib" / "crewai-files" / "tests" / "fixtures" / "agents.pdf"+++def _content_summary(block: dict[str, Any]) -> dict[str, str]:+ """Return a compact, non-base64 summary of a content block."""+ summary: dict[str, str] = {"type": str(block.get("type"))}+ for key in ("file_id", "file_url", "filename", "image_url"):+ if key in block:+ value = str(block[key])+ summary[key] = value[:100] + ("..." if len(value) > 100 else "")+ if "file_data" in block:+ value = str(block["file_data"])+ summary["file_data"] = value[:80] + f"... ({len(value)} chars)"+ return summary+++def _sanitize_payload(value: Any) -> Any:+ """Shorten large fields before printing API payloads."""+ if isinstance(value, Mapping):+ sanitized: dict[str, Any] = {}+ for key, item in value.items():+ if key == "file_data" and isinstance(item, str):+ sanitized[key] = item[:100] + f"... ({len(item)} chars)"+ else:+ sanitized[str(key)] = _sanitize_payload(item)+ return sanitized++ if isinstance(value, Sequence) and not isinstance(value, str | bytes):+ return [_sanitize_payload(item) for item in value]++ return value+++def inspect_native_path(pdf_path: Path, provider: str, api: str | None) -> None:+ """Show whether the PDF is treated as a native multimodal input."""+ pdf = PDFFile(source=str(pdf_path))+ supported_types = get_supported_content_types(provider, api=api)+ blocks = format_multimodal_content(+ {"document": pdf},+ provider=provider,+ api=api,+ text="Summarize this PDF.",+ )++ print("\n== Native File Formatting ==")+ print(f"PDF: {pdf_path}")+ print(f"Provider/API: {provider} / {api or 'default'}")+ print(f"Supported content types: {supported_types}")+ print(f"Content block count: {len(blocks)}")+ for index, block in enumerate(blocks, start=1):+ print(f" {index}. {_content_summary(block)}")++ has_pdf_block = any(block.get("type") == "input_file" for block in blocks)+ print(f"PDF native input_file block: {'YES' if has_pdf_block else 'NO'}")+++def inspect_fallback_tool(pdf_path: Path) -> None:+ """Show what read_file returns if a PDF falls back to the tool path."""+ from crewai.tools.agent_tools.read_file_tool import ReadFileTool++ tool = ReadFileTool()+ tool.set_files({"document": PDFFile(source=str(pdf_path))})+ result = tool._run("document")++ print("\n== read_file Fallback ==")+ print(f"Returned {len(result)} chars")+ print(f"Contains Base64 marker: {'YES' if 'Base64:' in result else 'NO'}")+ print("\nPreview:")+ print(result[:1200])+ if len(result) > 1200:+ print("...")+++def run_crew_kickoff(+ pdf_path: Path,+ model: str,+ api: str | None,+ prompt: str,+ *,+ payload_only: bool = False,+) -> None:+ """Run a real Crew kickoff against the supplied model."""+ from crewai import LLM, Agent, Crew, Task++ if model.startswith("openai/") and not os.getenv("OPENAI_API_KEY") and not payload_only:+ raise SystemExit(+ "OPENAI_API_KEY is not set. Export it before running --mode kickoff."+ )++ kwargs: dict[str, Any] = {"model": model, "temperature": 0}+ if api:+ kwargs["api"] = api++ llm = LLM(**kwargs)+ agent = Agent(+ role="PDF Analyst",+ goal="Read the provided PDF and answer accurately from its contents",+ backstory="You inspect uploaded files carefully and avoid guessing.",+ llm=llm,+ verbose=True,+ )+ task = Task(+ description=prompt,+ expected_output="A concise answer grounded in the uploaded PDF.",+ agent=agent,+ )+ crew = Crew(agents=[agent], tasks=[task], verbose=True)++ print("\n== Crew Kickoff ==")+ print(f"Model/API: {model} / {api or 'default'}")+ print(f"PDF: {pdf_path}")++ context = nullcontext()+ if payload_only:+ from crewai.llms.providers.openai.completion import OpenAICompletion++ def print_payload_and_stop(+ self: OpenAICompletion,+ params: dict[str, Any],+ *_args: Any,+ **_kwargs: Any,+ ) -> str:+ print("\n== Sanitized Responses Payload ==")+ print(_sanitize_payload(params))+ return "Payload debug complete."++ context = patch.object(+ OpenAICompletion,+ "_handle_responses",+ print_payload_and_stop,+ )++ with context:+ result = crew.kickoff(input_files={"document": PDFFile(source=str(pdf_path))})++ print("\n== Final Output ==")+ print(result.raw)+++def parse_args() -> argparse.Namespace:+ parser = argparse.ArgumentParser(description=__doc__)+ parser.add_argument(+ "--mode",+ choices=("inspect", "fallback", "payload", "kickoff", "all"),+ default="inspect",+ help="What to run. 'inspect', 'fallback', and 'payload' do not call an LLM.",+ )+ parser.add_argument(+ "--pdf",+ type=Path,+ default=DEFAULT_PDF,+ help="PDF file to test.",+ )+ parser.add_argument(+ "--provider",+ default="gpt-4o-mini",+ help="Provider/model string for file formatting inspection.",+ )+ parser.add_argument(+ "--model",+ default="openai/gpt-4o-mini",+ help="CrewAI model for real kickoff mode.",+ )+ parser.add_argument(+ "--api",+ default="responses",+ help="API variant. Use '' to omit.",+ )+ parser.add_argument(+ "--prompt",+ default="Summarize the uploaded PDF in 3 bullet points. Do not guess.",+ help="Task prompt for kickoff mode.",+ )+ return parser.parse_args()+++def main() -> None:+ args = parse_args()+ pdf_path = args.pdf.expanduser().resolve()+ api = args.api or None++ if not pdf_path.exists():+ raise SystemExit(f"PDF not found: {pdf_path}")++ if args.mode in ("inspect", "all"):+ inspect_native_path(pdf_path, args.provider, api)+ if args.mode in ("fallback", "all"):+ inspect_fallback_tool(pdf_path)+ if args.mode == "payload":+ run_crew_kickoff(pdf_path, args.model, api, args.prompt, payload_only=True)+ if args.mode in ("kickoff", "all"):+ run_crew_kickoff(+ pdf_path,+ args.model,+ api,+ args.prompt,+ payload_only=args.mode == "all",+ )+++if __name__ == "__main__":+ main()