Merge branch 'main' into feature/add-unstructured-api-provider

Author: sahana-sreeram

.github/workflows/publish-openapi-sdk.yml

@@ -124,7 +124,7 @@ jobs:
           UV_PUBLISH_TOKEN: ${{ secrets.PYPI_API_TOKEN }}
 
       - name: Upload build artifacts
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v4.6.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v4.6.0
         with:
           name: python-sdk-dist
           path: client-sdks/openapi/sdks/python/dist/

ARCHITECTURE.md

@@ -115,7 +115,7 @@ The full list of auto-routed pairs is defined in `builtin_automatically_routed_a
 
 The `ogx_api` package defines all public-facing types and protocols:
 
-- **Protocols** -- Python `Protocol` classes like `Inference`, `Responses` that define the API contract. HTTP routes are defined via FastAPI routers in `fastapi_routes.py` modules.
+- **Protocols** -- Python `Protocol` classes like `Inference`, `Responses`, `Skills` that define the API contract. HTTP routes are defined via FastAPI routers in `fastapi_routes.py` modules.
 - **Data Types** -- Pydantic models for requests, responses, and resources (e.g., `Model`, `VectorStore`, `ChatCompletionRequest`).
 - **Provider Specs** -- `InlineProviderSpec`, `RemoteProviderSpec`, and related types that define how providers are declared.
 - **Internal utilities** -- KVStore and SqlStore abstract interfaces live here so third-party providers can use them without depending on the full server.
@@ -152,7 +152,7 @@ storage:
 | PostgreSQL| `PostgresKVStoreConfig`  | Production deployments |
 | MongoDB   | `MongoDBKVStoreConfig`   | Document-oriented      |
 
-Used by: distribution registry, quota tracking, provider state.
+Used by: distribution registry, quota tracking, provider state, skills metadata.
 
 ### SqlStore
 

README.md

@@ -41,6 +41,7 @@ response = client.chat.completions.create(
 - **Responses API** — server-side agentic orchestration with tool calling, MCP server integration, and built-in file search (RAG) in a single API call ([learn more](https://ogx-ai.github.io/docs/api-openai))
 - **Vector Stores & Files** — `/v1/vector_stores` and `/v1/files` for managed document storage and search
 - **Batches** — `/v1/batches` for offline batch processing
+- **Skills** — `/v1alpha/skills` for managing versioned skill bundles (zip archives with SKILL.md manifests) that agents can invoke
 - **[Open Responses](https://www.openresponses.org/) conformant** — the Responses API implementation passes the Open Responses conformance test suite
 - **Multi-SDK support** — use the [Anthropic SDK](https://docs.anthropic.com/en/api/messages) (`/v1/messages`) or [Google GenAI SDK](https://ai.google.dev/gemini-api/docs/interactions) (`/v1alpha/interactions`) natively alongside the OpenAI API
 

pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools>=61.0", "setuptools-scm>=8.0"]
 build-backend = "setuptools.build_meta"
 
 [tool.setuptools_scm]
-fallback_version = "1.0.3.dev0"
+fallback_version = "1.1.1.dev0"
 
 [tool.uv]
 required-version = ">=0.7.0"
@@ -80,7 +80,7 @@ dependencies = [
 
 [project.optional-dependencies]
 client = [
-    "ogx-client>=1.0.2", # Optional for library-only usage
+    "ogx-client>=1.1.0", # Optional for library-only usage
 ]
 starter = [
     "aiohttp",
@@ -145,7 +145,7 @@ dev = [
     "pre-commit>=4.4.0",
     "ruamel.yaml",                   # needed for openapi generator
     "openapi-spec-validator>=0.9.0",
-    "ogx-client>=1.0.2",
+    "ogx-client>=1.1.0",
     "boto3>=1.43.18",
     "torch>=2.6.0",
 ]
@@ -184,7 +184,7 @@ type_checking = [
     "langchain-openai>=1.2.2",
     "langchain-core",
     "langgraph",
-    "ogx-client>=1.0.2",
+    "ogx-client>=1.1.0",
 ]
 test-common = [
     "aiohttp",

src/ogx/cli/stack/lets_go.py

@@ -248,6 +248,7 @@ def _add_file_search_and_responses(run_config: StackConfig) -> None:
         ("remote::brave-search", "brave-search"),
         ("remote::tavily-search", "tavily-search"),
         ("remote::bing-search", "bing-search"),
+        ("remote::nimble-search", "nimble-search"),
     ]
     tool_runtime_registry = get_provider_registry().get(Api.tool_runtime, {})
     existing_web_search: set[str] = {

src/ogx/providers/inline/README.md

@@ -14,6 +14,7 @@ inline/
   tool_runtime/        # Tool runtime (RAG context retrieval)
   files/               # File storage and management
   file_processor/      # File processing (text extraction, etc.)
+  skills/              # Skills API (versioned skill bundle management)
   __init__.py
 ```
 
@@ -29,3 +30,4 @@ Their factory function is typically named `get_provider_impl()` and returns an i
 - **`inference/sentence_transformers`** -- Runs embedding models using the sentence-transformers library.
 - **`inference/transformers`** -- Runs Llama models locally using the transformers library.
 - **`vector_io/sqlite_vec`** -- SQLite-based vector storage using the sqlite-vec extension.
+- **`skills/builtin`** -- Manages versioned skill bundles (zip archives with SKILL.md manifests). Stores bundles via the Files API and metadata in KVStore.

src/ogx/providers/inline/file_processor/docling/docling.py

@@ -4,8 +4,10 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import asyncio
 import os
 import tempfile
+import threading
 import time
 import uuid
 from typing import Any
@@ -49,6 +51,7 @@ def __init__(self, config: DoclingFileProcessorConfig, files_api=None) -> None:
         self.converter = DocumentConverter(
             format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
         )
+        self._converter_lock = threading.Lock()
 
     async def process_file(
         self,
@@ -80,13 +83,25 @@ async def process_file(
             )
             content = content_response.body
 
+        return await asyncio.to_thread(self._process_content, content, filename, file_id, chunking_strategy, start_time)
+
+    def _process_content(
+        self,
+        content: bytes,
+        filename: str,
+        file_id: str | None,
+        chunking_strategy: VectorStoreChunkingStrategy | None,
+        start_time: float,
+    ) -> ProcessFileResponse:
+        """Convert and chunk file content. Runs in a thread."""
         # Preserve original file extension so DocumentConverter can detect the format
         suffix = os.path.splitext(filename)[1] or ".bin"
         with tempfile.NamedTemporaryFile(suffix=suffix, delete=True) as tmp:
             tmp.write(content)
             tmp.flush()
 
-            result = self.converter.convert(tmp.name)
+            with self._converter_lock:
+                result = self.converter.convert(tmp.name)
 
         doc = result.document
         page_count = doc.num_pages()

src/ogx/providers/inline/file_processor/markitdown/markitdown_processor.py

@@ -4,8 +4,10 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import asyncio
 import os
 import tempfile
+import threading
 import time
 import uuid
 from typing import Any
@@ -40,6 +42,7 @@ def __init__(self, config: MarkItDownFileProcessorConfig, files_api) -> None:
         self.config = config
         self.files_api = files_api
         self.converter = MarkItDown()
+        self._converter_lock = threading.Lock()
 
     async def process_file(
         self,
@@ -69,13 +72,25 @@ async def process_file(
             )
             content = content_response.body
 
+        return await asyncio.to_thread(self._process_content, content, filename, file_id, chunking_strategy, start_time)
+
+    def _process_content(
+        self,
+        content: bytes,
+        filename: str,
+        file_id: str | None,
+        chunking_strategy: VectorStoreChunkingStrategy | None,
+        start_time: float,
+    ) -> ProcessFileResponse:
+        """Convert and chunk file content. Runs in a thread."""
         suffix = os.path.splitext(filename)[1] or ".bin"
         with tempfile.NamedTemporaryFile(suffix=suffix, delete=True) as tmp:
             tmp.write(content)
             tmp.flush()
 
             try:
-                result = self.converter.convert(tmp.name)
+                with self._converter_lock:
+                    result = self.converter.convert(tmp.name)
             except Exception as e:
                 raise HTTPException(
                     status_code=422,

src/ogx/providers/inline/file_processor/pypdf/pypdf.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import asyncio
 import io
 import mimetypes
 import time
@@ -74,9 +75,11 @@ async def process_file(
         mime_category = mime_type.split("/")[0] if (mime_type and "/" in mime_type) else None
 
         if mime_type == "application/pdf":
-            return self._process_pdf(content, filename, file_id, chunking_strategy, start_time)
+            return await asyncio.to_thread(self._process_pdf, content, filename, file_id, chunking_strategy, start_time)
         elif mime_category == "text":
-            return self._process_text(content, filename, file_id, chunking_strategy, start_time)
+            return await asyncio.to_thread(
+                self._process_text, content, filename, file_id, chunking_strategy, start_time
+            )
         else:
             raise HTTPException(
                 status_code=422,

src/ogx/providers/inline/interactions/impl.py

@@ -298,7 +298,6 @@ async def _passthrough_request(
                 logger.error(
                     "Passthrough request failed",
                     status_code=resp.status_code,
-                    response_body=resp.text[:500],
                     url=url,
                 )
                 return JSONResponse(content=resp.json(), status_code=resp.status_code)