unique_toolkit 1.70.3__tar.gz → 1.70.5__tar.gz

{unique_toolkit-1.70.3 → unique_toolkit-1.70.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-toolkit
-Version: 1.70.3
+Version: 1.70.5
 Summary: 
 Author: Cedric Klinkert, Martin Fadler, Sadique Sheik, Fabian Schläpfer, Pascal Hauri, Ejiroghene Oki-Peter, Rami Azouz, Ricardo Visini, Sanzio Monti
 Author-email: Cedric Klinkert <cedric.klinkert@unique.ch>, Martin Fadler <martin.fadler@unique.ch>, Sadique Sheik <sadique@unique.ch>, Fabian Schläpfer <fabian@unique.ch>, Pascal Hauri <pascal@unique.ch>, Ejiroghene Oki-Peter <oki-peter.ejiroghene@unique.ai>, Rami Azouz <rami.ext@unique.ch>, Ricardo Visini <ricardo.visini@unique.ch>, Sanzio Monti <sanzio.monti@unique.ai>

{unique_toolkit-1.70.3 → unique_toolkit-1.70.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_toolkit"
-version = "1.70.3"
+version = "1.70.5"
 description = ""
 readme = "README.md"
 requires-python = ">=3.12"

{unique_toolkit-1.70.3 → unique_toolkit-1.70.5}/unique_toolkit/agentic/tools/openai_builtin/code_interpreter/postprocessors/generated_files.py

@@ -587,7 +587,12 @@ class DisplayCodeInterpreterFilesPostProcessor(
                 missed_files.append(filename)
 
             is_html_rendered = is_html
-            if replaced and not (is_image or is_html_rendered or fence_ff_on):
+            has_superscript = f"<sup>{ref_number}</sup>" in loop_response.message.text
+            if (
+                replaced
+                and has_superscript
+                and not (is_image or is_html_rendered or fence_ff_on)
+            ):
                 loop_response.message.references.append(
                     ContentReference(
                         sequence_number=ref_number,
@@ -613,7 +618,16 @@ class DisplayCodeInterpreterFilesPostProcessor(
                 len(code_blocks),
                 [f.filename for b in code_blocks for f in b.files],
             )
-            _warn_unmatched_code_blocks(self._content_map, code_blocks)
+            unmatched_blocks = _warn_unmatched_code_blocks(
+                self._content_map, code_blocks
+            )
+            if unmatched_blocks:
+                self._log.info(
+                    "Fence path: %d file(s) have no code-block match and will appear "
+                    "as plain download links: %s",
+                    len(unmatched_blocks),
+                    list(unmatched_blocks),
+                )
             text_before = loop_response.message.text
             loop_response.message.text = _inject_code_execution_fences(
                 loop_response.message.text,
@@ -1409,7 +1423,7 @@ def _replace_dangling_sandbox_links(text: str, error_message: str) -> tuple[str,
 def _warn_unmatched_code_blocks(
     content_map: dict[str, str | None],
     code_blocks: list[CodeInterpreterBlock],
-) -> None:
+) -> dict[str, str]:
     """Warn for files that were uploaded but could not be matched to any code block.
 
     When the fence feature flag is on, every uploaded file should map to a code block
@@ -1417,8 +1431,13 @@ def _warn_unmatched_code_blocks(
     matched (e.g. the LLM used a variable for the output path rather than a literal
     string) it falls back to a plain unique://content/ link with no code context.
     The user can still download it, but the frontend artifact UI will not be shown.
+
+    Returns:
+        Mapping of filename → content_id for every file that could not be matched
+        to a code block.  Callers may use this to inject additional fallback links.
     """
     fenced_filenames = {f.filename for block in code_blocks for f in block.files}
+    unmatched: dict[str, str] = {}
     for filename, content_id in content_map.items():
         if content_id is None:
             continue
@@ -1434,6 +1453,8 @@ def _warn_unmatched_code_blocks(
                 content_id,
                 filename,
             )
+            unmatched[filename] = content_id
+    return unmatched
 
 
 def _get_next_ref_number(references: list[ContentReference] | None) -> int:
@@ -1474,11 +1495,12 @@ def _replace_container_image_citation(
     if not re.search(image_markdown, text):
         logger.warning(
             "No sandbox link found for image '%s' (content_id=%s); "
-            "file was uploaded but the LLM did not reference it — it will not be displayed.",
+            "file was uploaded but the LLM did not reference it — appending fallback link.",
             filename,
             content_id,
         )
-        return text, False
+        fallback = f"\n\n📎 [{filename}](unique://content/{content_id})"
+        return text + fallback, True
 
     logger.info("Inserting image citation for '%s'", filename)
     return re.sub(
@@ -1497,11 +1519,12 @@ def _replace_container_html_citation(
     if not re.search(html_markdown, text):
         logger.warning(
             "No sandbox link found for HTML file '%s' (content_id=%s); "
-            "file was uploaded but the LLM did not reference it — it will not be displayed.",
+            "file was uploaded but the LLM did not reference it — appending fallback link.",
             filename,
             content_id,
         )
-        return text, False
+        fallback = f"\n\n📎 [{filename}](unique://content/{content_id})"
+        return text + fallback, True
 
     logger.info("Inserting HTML rendering block for '%s'", filename)
     block = f"```HtmlRendering\n800px\n600px\n\nunique://content/{content_id}\n\n```"
@@ -1554,11 +1577,12 @@ def _replace_container_file_citation(
     if not re.search(file_markdown, text):
         logger.warning(
             "No sandbox link found for file '%s' (content_id=%s); "
-            "file was uploaded but the LLM did not reference it — it will not be displayed.",
+            "file was uploaded but the LLM did not reference it — appending fallback link.",
             filename,
             content_id,
         )
-        return text, False
+        fallback = f"\n\n📎 [{filename}](unique://content/{content_id})"
+        return text + fallback, True
 
     logger.info("Displaying file %s", filename)
     replacement = (

{unique_toolkit-1.70.3 → unique_toolkit-1.70.5}/unique_toolkit/agentic/tools/openai_builtin/code_interpreter/service.py

@@ -6,6 +6,12 @@ from typing import Any, override
 from openai import AsyncOpenAI, BaseModel, NotFoundError
 from openai.types.responses import ResponseCodeInterpreterToolCall, ResponseIncludable
 from openai.types.responses.tool_param import CodeInterpreter
+from tenacity import (
+    AsyncRetrying,
+    before_sleep_log,
+    stop_after_attempt,
+    wait_exponential,
+)
 
 from unique_toolkit import ContentService, ShortTermMemoryService
 from unique_toolkit.agentic.feature_flags.feature_flags import feature_flags
@@ -26,6 +32,25 @@ from unique_toolkit.content.schemas import (
 
 logger = logging.getLogger(__name__)
 
+_UPLOAD_MAX_RETRIES = 2
+_UPLOAD_RETRY_BASE_DELAY = 0.5
+
+
+def _build_upload_retry() -> AsyncRetrying:
+    """Exponential-backoff retry policy for transient upload/download failures.
+
+    Matches the pattern used in the ``DisplayCodeInterpreterFilesPostProcessor``
+    so that every outbound I/O call gets the same behaviour: up to
+    ``_UPLOAD_MAX_RETRIES`` extra attempts, doubling the wait each time,
+    with a WARNING log before each sleep.
+    """
+    return AsyncRetrying(
+        stop=stop_after_attempt(1 + _UPLOAD_MAX_RETRIES),
+        wait=wait_exponential(multiplier=_UPLOAD_RETRY_BASE_DELAY),
+        before_sleep=before_sleep_log(logger, logging.WARNING),
+        reraise=True,
+    )
+
 
 _SHORT_TERM_MEMORY_NAME = "container_code_execution"
 
@@ -127,17 +152,34 @@ async def _upload_files_to_container(
 
         if upload:
             logger.info(
-                "Uploding file %s to container %s", file.id, memory.container_id
+                "Uploading file %s (%s) to container %s",
+                file.id,
+                file.key,
+                memory.container_id,
             )
-            file_content = await content_service.download_content_to_bytes_async(
-                content_id=file.id, chat_id=chat_id
+            file_content = await _build_upload_retry()(
+                content_service.download_content_to_bytes_async,
+                content_id=file.id,
+                chat_id=chat_id,
             )
-
-            openai_file = await client.containers.files.create(
+            logger.info(
+                "Downloaded %d bytes for file %s; uploading to container %s",
+                len(file_content),
+                file.id,
+                container_id,
+            )
+            openai_file = await _build_upload_retry()(
+                client.containers.files.create,
                 container_id=container_id,
                 file=(file.key, file_content),
             )
             memory.file_ids[file.id] = openai_file.id
+            logger.info(
+                "File %s successfully uploaded as OpenAI file %s in container %s",
+                file.id,
+                openai_file.id,
+                container_id,
+            )
 
     return memory