ai-pipeline-core 0.4.5__tar.gz → 0.4.7__tar.gz

{ai_pipeline_core-0.4.5 → ai_pipeline_core-0.4.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-pipeline-core
-Version: 0.4.5
+Version: 0.4.7
 Summary: Core utilities for AI-powered processing pipelines using prefect
 Project-URL: Homepage, https://github.com/bbarwik/ai-pipeline-core
 Project-URL: Repository, https://github.com/bbarwik/ai-pipeline-core

{ai_pipeline_core-0.4.5 → ai_pipeline_core-0.4.7}/ai_pipeline_core/__init__.py

@@ -64,7 +64,7 @@ from .prompt_manager import PromptManager
 from .settings import Settings
 from .testing import disable_run_logger, prefect_test_harness
 
-__version__ = "0.4.5"
+__version__ = "0.4.6"
 
 __all__ = [
     "AIMessageType",

{ai_pipeline_core-0.4.5 → ai_pipeline_core-0.4.7}/ai_pipeline_core/deployment/deploy.py

@@ -383,7 +383,7 @@ class Deployer:
         dest_uri = f"gs://{self.config['bucket']}/{flow_folder}/{tarball.name}"
         self._info(f"Uploading to {dest_uri}")
 
-        tarball_bytes = tarball.read_bytes()  # noqa: ASYNC240
+        tarball_bytes = tarball.read_bytes()
         await bucket.write_path(tarball.name, tarball_bytes)
 
         self._success(f"Package uploaded to {flow_folder}/{tarball.name}")
@@ -527,15 +527,16 @@ class Deployer:
         # Phase 3: Build vendor packages from [tool.deploy].vendor_packages
         vendor_wheels = self._build_vendor_packages()
 
-        # Also include cli_agents wheels from agent builds
-        if agent_builds:
-            seen_agent: set[str] = set()
-            for build_info in agent_builds.values():
-                for filename, filepath in build_info["files"].items():
-                    if filename.endswith(".whl") and filename not in seen_agent and "cli_agents" in filename:
-                        if filename not in {w.name for w in vendor_wheels}:
-                            vendor_wheels.append(filepath)
-                        seen_agent.add(filename)
+        # Build cli-agents wheel if source is configured — it's a private package
+        # not on PyPI, so the worker needs the wheel even when no agents are deployed
+        cli_agents_source = self._get_cli_agents_source()
+        if cli_agents_source:
+            cli_dir = Path(cli_agents_source).resolve()
+            if (cli_dir / "pyproject.toml").exists():
+                cli_wheel = self._build_wheel_from_source(cli_dir)
+                if cli_wheel.name not in {w.name for w in vendor_wheels}:
+                    vendor_wheels.append(cli_wheel)
+                    self._success(f"Built cli-agents vendor wheel: {cli_wheel.name}")
 
         # Phase 4: Upload flow package + vendor wheels
         await self._upload_package(tarball, vendor_wheels)

ai_pipeline_core-0.4.7/ai_pipeline_core/deployment/remote.py

@@ -0,0 +1,192 @@
+"""Remote deployment utilities for calling PipelineDeployment flows via Prefect."""
+
+import asyncio
+from collections.abc import Awaitable, Callable, Coroutine
+from functools import wraps
+from typing import Any, TypeVar, cast
+from uuid import UUID
+
+from prefect import get_client
+from prefect.client.orchestration import PrefectClient
+from prefect.client.schemas import FlowRun
+from prefect.context import AsyncClientContext
+from prefect.deployments.flow_runs import run_deployment
+from prefect.exceptions import ObjectNotFound
+
+from ai_pipeline_core.deployment import DeploymentContext, DeploymentResult, PipelineDeployment
+from ai_pipeline_core.documents import Document
+from ai_pipeline_core.logging import get_pipeline_logger
+from ai_pipeline_core.observability.tracing import TraceLevel, set_trace_cost, trace
+from ai_pipeline_core.pipeline.options import FlowOptions
+from ai_pipeline_core.settings import settings
+
+logger = get_pipeline_logger(__name__)
+
+TOptions = TypeVar("TOptions", bound=FlowOptions)
+TResult = TypeVar("TResult", bound=DeploymentResult)
+
+ProgressCallback = Callable[[float, str], Awaitable[None]]
+"""Signature for remote deployment progress callbacks: (fraction, message) -> None."""
+
+
+def _is_already_traced(func: Callable[..., Any]) -> bool:
+    """Check if function or its __wrapped__ has __is_traced__ attribute."""
+    if getattr(func, "__is_traced__", False):
+        return True
+    wrapped = getattr(func, "__wrapped__", None)
+    return getattr(wrapped, "__is_traced__", False) if wrapped else False
+
+
+_POLL_INTERVAL = 5.0
+
+
+async def _poll_remote_flow_run(
+    client: PrefectClient,
+    flow_run_id: UUID,
+    deployment_name: str,
+    poll_interval: float = _POLL_INTERVAL,
+    on_progress: ProgressCallback | None = None,
+) -> Any:
+    """Poll a remote flow run until final, invoking on_progress callback with progress.
+
+    Reads the remote flow run's progress labels on each poll cycle and calls
+    on_progress(fraction, message) if provided. Without a callback, no progress
+    is reported. Only sends 1.0 on successful completion (not failure).
+    """
+    last_fraction = 0.0
+
+    while True:
+        try:
+            flow_run = await client.read_flow_run(flow_run_id)
+        except Exception:
+            logger.warning("Failed to poll remote flow run %s", flow_run_id, exc_info=True)
+            await asyncio.sleep(poll_interval)
+            continue
+
+        state = flow_run.state
+        if state and state.is_final():
+            if on_progress and state.is_completed():
+                await on_progress(1.0, f"[{deployment_name}] Completed")
+            return await state.result()  # type: ignore[union-attr]
+
+        if on_progress:
+            labels: dict[str, Any] = flow_run.labels or {}
+            progress_val = labels.get("progress.progress")
+
+            if progress_val is not None:
+                fraction = max(float(progress_val), last_fraction)
+                last_fraction = fraction
+                flow_name = str(labels.get("progress.flow_name", ""))
+                message = str(labels.get("progress.message", ""))
+                display = f"[{deployment_name}] {flow_name}: {message}" if flow_name else f"[{deployment_name}] Running"
+                await on_progress(fraction, display)
+            else:
+                await on_progress(last_fraction, f"[{deployment_name}] Waiting to start")
+
+        await asyncio.sleep(poll_interval)
+
+
+async def run_remote_deployment(
+    deployment_name: str,
+    parameters: dict[str, Any],
+    on_progress: ProgressCallback | None = None,
+) -> Any:
+    """Run a remote Prefect deployment with optional progress callback.
+
+    Creates the remote flow run immediately (timeout=0) then polls its state,
+    invoking on_progress(fraction, message) on each poll cycle if provided.
+    """
+
+    async def _create_and_poll(client: PrefectClient, as_subflow: bool) -> Any:
+        fr: FlowRun = await run_deployment(
+            client=client,
+            name=deployment_name,
+            parameters=parameters,
+            as_subflow=as_subflow,
+            timeout=0,
+        )  # type: ignore
+        return await _poll_remote_flow_run(client, fr.id, deployment_name, on_progress=on_progress)
+
+    async with get_client() as client:
+        try:
+            await client.read_deployment_by_name(name=deployment_name)
+            return await _create_and_poll(client, True)  # noqa: FBT003
+        except ObjectNotFound:
+            pass
+
+    if not settings.prefect_api_url:
+        raise ValueError(f"{deployment_name} not found, PREFECT_API_URL not set")
+
+    async with PrefectClient(
+        api=settings.prefect_api_url,
+        api_key=settings.prefect_api_key,
+        auth_string=settings.prefect_api_auth_string,
+    ) as client:
+        try:
+            await client.read_deployment_by_name(name=deployment_name)
+            ctx = AsyncClientContext.model_construct(client=client, _httpx_settings=None, _context_stack=0)
+            with ctx:
+                return await _create_and_poll(client, False)  # noqa: FBT003
+        except ObjectNotFound:
+            pass
+
+    raise ValueError(f"{deployment_name} deployment not found")
+
+
+def remote_deployment(
+    deployment_class: type[PipelineDeployment[TOptions, TResult]],
+    *,
+    deployment_name: str | None = None,
+    name: str | None = None,
+    trace_level: TraceLevel = "always",
+    trace_cost: float | None = None,
+) -> Callable[[Callable[..., Any]], Callable[..., Coroutine[Any, Any, TResult]]]:
+    """Decorator to call PipelineDeployment flows remotely with automatic serialization.
+
+    The decorated function's body is never executed — it serves as a typed stub.
+    The wrapper enforces the deployment contract: (project_name, documents, options, context).
+    """
+
+    def decorator(func: Callable[..., Any]) -> Callable[..., Coroutine[Any, Any, TResult]]:
+        fname = getattr(func, "__name__", deployment_class.name)
+
+        if _is_already_traced(func):
+            raise TypeError(f"@remote_deployment target '{fname}' already has @trace")
+
+        @wraps(func)
+        async def _wrapper(
+            project_name: str,
+            documents: list[Document],
+            options: TOptions,
+            context: DeploymentContext | None = None,
+            *,
+            on_progress: ProgressCallback | None = None,
+        ) -> TResult:
+            parameters: dict[str, Any] = {
+                "project_name": project_name,
+                "documents": documents,
+                "options": options,
+                "context": context if context is not None else DeploymentContext(),
+            }
+
+            full_name = f"{deployment_class.name}/{deployment_name or deployment_class.name.replace('-', '_')}"
+
+            result = await run_remote_deployment(full_name, parameters, on_progress=on_progress)
+
+            if trace_cost is not None and trace_cost > 0:
+                set_trace_cost(trace_cost)
+
+            if isinstance(result, DeploymentResult):
+                return cast(TResult, result)
+            if isinstance(result, dict):
+                return cast(TResult, deployment_class.result_type(**cast(dict[str, Any], result)))
+            raise TypeError(f"Expected DeploymentResult, got {type(result).__name__}")
+
+        traced_wrapper = trace(
+            level=trace_level,
+            name=name or deployment_class.name,
+        )(_wrapper)
+
+        return traced_wrapper
+
+    return decorator

{ai_pipeline_core-0.4.5 → ai_pipeline_core-0.4.7}/ai_pipeline_core/llm/ai_messages.py

@@ -38,6 +38,34 @@ def _ensure_llm_compatible_image(content: bytes, mime_type: str) -> tuple[bytes,
     return buf.getvalue(), "image/png"
 
 
+def _looks_like_text(content: bytes) -> bool:
+    """Check if content is valid UTF-8 text (not binary).
+
+    Uses heuristics: must decode as UTF-8 and have no null bytes.
+    Null bytes are common in binary files but rare in text.
+    """
+    if not content:
+        return True
+    # Null bytes indicate binary content
+    if b"\x00" in content:
+        return False
+    try:
+        content.decode("utf-8")
+        return True
+    except UnicodeDecodeError:
+        return False
+
+
+def _has_pdf_signature(content: bytes) -> bool:
+    """Check if content starts with PDF magic bytes (%PDF-).
+
+    Real PDFs start with %PDF- (possibly after whitespace).
+    This prevents false positives when a real PDF happens to be
+    partly UTF-8 decodable (e.g., ASCII-heavy PDF metadata).
+    """
+    return content.lstrip().startswith(b"%PDF-")
+
+
 AIMessageType = str | Document | ModelResponse
 """Type for messages in AIMessages container.
 
@@ -350,7 +378,7 @@ class AIMessages(list[AIMessageType]):  # noqa: PLR0904
         return count
 
     @staticmethod
-    def document_to_prompt(document: Document) -> list[ChatCompletionContentPartParam]:  # noqa: PLR0912, PLR0914
+    def document_to_prompt(document: Document) -> list[ChatCompletionContentPartParam]:  # noqa: C901, PLR0912, PLR0914, PLR0915
         """Convert a document to prompt format for LLM consumption.
 
         Renders the document as XML with text/image/PDF content, followed by any
@@ -368,8 +396,15 @@ class AIMessages(list[AIMessageType]):  # noqa: PLR0904
         description = f"<description>{document.description}</description>\n" if document.description else ""
         header_text = f"<document>\n<id>{document.id}</id>\n<name>{document.name}</name>\n{description}"
 
+        # Check if "PDF" is actually text (misnamed file from URL ending in .pdf)
+        # Real PDFs start with %PDF- magic bytes; if missing and content is UTF-8, it's text
+        is_text = document.is_text
+        if not is_text and document.is_pdf and _looks_like_text(document.content) and not _has_pdf_signature(document.content):
+            is_text = True
+            logger.debug(f"Document '{document.name}' has PDF extension but contains text content - sending as text")
+
         # Handle text documents
-        if document.is_text:
+        if is_text:
             text_content = document.content.decode("utf-8")
             content_text = f"{header_text}<content>\n{text_content}\n</content>\n"
             prompt.append({"type": "text", "text": content_text})
@@ -407,8 +442,16 @@ class AIMessages(list[AIMessageType]):  # noqa: PLR0904
             desc_attr = f' description="{att.description}"' if att.description else ""
             att_open = f'<attachment name="{att.name}"{desc_attr}>\n'
 
-            if att.is_text:
-                prompt.append({"type": "text", "text": f"{att_open}{att.text}\n</attachment>\n"})
+            # Check if "PDF" attachment is actually text (same logic as document)
+            att_is_text = att.is_text
+            if not att_is_text and att.is_pdf and _looks_like_text(att.content) and not _has_pdf_signature(att.content):
+                att_is_text = True
+                logger.debug(f"Attachment '{att.name}' has PDF extension but contains text content - sending as text")
+
+            if att_is_text:
+                # Use content.decode() directly - att.text property raises ValueError if is_text is False
+                att_text = att.content.decode("utf-8")
+                prompt.append({"type": "text", "text": f"{att_open}{att_text}\n</attachment>\n"})
             elif att.is_image or att.is_pdf:
                 prompt.append({"type": "text", "text": att_open})
 

{ai_pipeline_core-0.4.5 → ai_pipeline_core-0.4.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ai-pipeline-core"
-version = "0.4.5"
+version = "0.4.7"
 description = "Core utilities for AI-powered processing pipelines using prefect"
 readme = "README.md"
 license = {text = "MIT"}
@@ -142,7 +142,7 @@ convention = "google"
 ".vulture_whitelist.py" = ["B018", "E402", "F401", "D"]
 "ai_pipeline_core/__init__.py" = ["RUF067"]
 "ai_pipeline_core/deployment/base.py" = ["T20", "C901", "PLR0912", "PLR0914", "PLR0915", "PLR0917", "PLR6301", "PLC0415", "UP046"]
-"ai_pipeline_core/deployment/deploy.py" = ["T20", "PLC2701", "S404", "S602", "S603"]
+"ai_pipeline_core/deployment/deploy.py" = ["T20", "PLC2701", "S404", "S602", "S603", "ASYNC"]
 "ai_pipeline_core/deployment/remote.py" = ["UP047"]
 "ai_pipeline_core/deployment/progress.py" = ["RUF029"]
 "ai_pipeline_core/documents/document.py" = ["PLR0904"]

ai_pipeline_core-0.4.5/ai_pipeline_core/deployment/remote.py

@@ -1,116 +0,0 @@
-"""Remote deployment utilities for calling PipelineDeployment flows via Prefect."""
-
-import inspect
-from collections.abc import Callable
-from functools import wraps
-from typing import Any, ParamSpec, TypeVar, cast
-
-from prefect import get_client
-from prefect.client.orchestration import PrefectClient
-from prefect.client.schemas import FlowRun
-from prefect.context import AsyncClientContext
-from prefect.deployments.flow_runs import run_deployment
-from prefect.exceptions import ObjectNotFound
-
-from ai_pipeline_core.deployment import DeploymentContext, DeploymentResult, PipelineDeployment
-from ai_pipeline_core.observability.tracing import TraceLevel, set_trace_cost, trace
-from ai_pipeline_core.pipeline.options import FlowOptions
-from ai_pipeline_core.settings import settings
-
-P = ParamSpec("P")
-TOptions = TypeVar("TOptions", bound=FlowOptions)
-TResult = TypeVar("TResult", bound=DeploymentResult)
-
-
-def _is_already_traced(func: Callable[..., Any]) -> bool:
-    """Check if function or its __wrapped__ has __is_traced__ attribute."""
-    if getattr(func, "__is_traced__", False):
-        return True
-    wrapped = getattr(func, "__wrapped__", None)
-    return getattr(wrapped, "__is_traced__", False) if wrapped else False
-
-
-async def run_remote_deployment(deployment_name: str, parameters: dict[str, Any]) -> Any:
-    """Run a remote Prefect deployment, trying local client first then remote."""
-
-    async def _run(client: PrefectClient, as_subflow: bool) -> Any:
-        fr: FlowRun = await run_deployment(client=client, name=deployment_name, parameters=parameters, as_subflow=as_subflow)  # type: ignore
-        return await fr.state.result()  # type: ignore
-
-    async with get_client() as client:
-        try:
-            await client.read_deployment_by_name(name=deployment_name)
-            return await _run(client, True)  # noqa: FBT003
-        except ObjectNotFound:
-            pass
-
-    if not settings.prefect_api_url:
-        raise ValueError(f"{deployment_name} not found, PREFECT_API_URL not set")
-
-    async with PrefectClient(
-        api=settings.prefect_api_url,
-        api_key=settings.prefect_api_key,
-        auth_string=settings.prefect_api_auth_string,
-    ) as client:
-        try:
-            await client.read_deployment_by_name(name=deployment_name)
-            ctx = AsyncClientContext.model_construct(client=client, _httpx_settings=None, _context_stack=0)
-            with ctx:
-                return await _run(client, False)  # noqa: FBT003
-        except ObjectNotFound:
-            pass
-
-    raise ValueError(f"{deployment_name} deployment not found")
-
-
-def remote_deployment(
-    deployment_class: type[PipelineDeployment[TOptions, TResult]],
-    *,
-    deployment_name: str | None = None,
-    name: str | None = None,
-    trace_level: TraceLevel = "always",
-    trace_cost: float | None = None,
-) -> Callable[[Callable[P, TResult]], Callable[P, TResult]]:
-    """Decorator to call PipelineDeployment flows remotely with automatic serialization."""
-
-    def decorator(func: Callable[P, TResult]) -> Callable[P, TResult]:
-        fname = getattr(func, "__name__", deployment_class.name)
-
-        if _is_already_traced(func):
-            raise TypeError(f"@remote_deployment target '{fname}' already has @trace")
-
-        @wraps(func)
-        async def _wrapper(*args: P.args, **kwargs: P.kwargs) -> TResult:
-            sig = inspect.signature(func)
-            bound = sig.bind(*args, **kwargs)
-            bound.apply_defaults()
-
-            # Pass parameters with proper types - Prefect handles Pydantic serialization
-            parameters: dict[str, Any] = {}
-            for pname, value in bound.arguments.items():
-                if value is None and pname == "context":
-                    parameters[pname] = DeploymentContext()
-                else:
-                    parameters[pname] = value
-
-            full_name = f"{deployment_class.name}/{deployment_name or deployment_class.name.replace('-', '_')}"
-
-            result = await run_remote_deployment(full_name, parameters)
-
-            if trace_cost is not None and trace_cost > 0:
-                set_trace_cost(trace_cost)
-
-            if isinstance(result, DeploymentResult):
-                return cast(TResult, result)
-            if isinstance(result, dict):
-                return cast(TResult, deployment_class.result_type(**cast(dict[str, Any], result)))
-            raise TypeError(f"Expected DeploymentResult, got {type(result).__name__}")
-
-        traced_wrapper = trace(
-            level=trace_level,
-            name=name or deployment_class.name,
-        )(_wrapper)
-
-        return traced_wrapper  # type: ignore[return-value]
-
-    return decorator