Merge branch 'main' into feat/talk-reactions-polls-files

Author: Pavlinchen

.github/workflows/integration_test.yml

@@ -150,22 +150,31 @@ jobs:
           sudo apt-get update
           sudo apt install pipx
           pipx install poetry
+          mkdir -p ../llm2-persistent_storage
           poetry install
 
+      - name: Cache llm2 models
+        uses: actions/cache/restore@v5
+        id: cache-llm2-models-restore
+        env:
+          cache-name: cache-llm2-models
+        with:
+          path: llm2-persistent_storage/
+          key: ${{ runner.os }}-llm2-models-${{ env.cache-name }}-${{ hashFiles('llm2/lib/main.py') }}
+
       - name: Init llm2
         working-directory: llm2/lib
         env:
           APP_ID: llm2
           APP_PORT: 9080
           APP_VERSION: ${{ fromJson(steps.llm2_appinfo.outputs.result).version }}
         run: |
-          poetry run python3 main.py > ../logs 2>&1 &
+          APP_PERSISTENT_STORAGE="$(pwd)/../../llm2-persistent-storage/" poetry run python3 main.py > ../logs 2>&1 &
 
       - name: Register backend
         run: |
           ./occ app_api:app:register llm2 manual_install --json-info "{\"appid\":\"llm2\",\"name\":\"Local large language model\",\"daemon_config_name\":\"manual_install\",\"version\":\"${{ fromJson(steps.llm2_appinfo.outputs.result).version }}\",\"secret\":\"12345\",\"port\":9080,\"scopes\":[\"AI_PROVIDERS\", \"TASK_PROCESSING\"],\"system_app\":0}" --force-scopes --wait-finish
 
-
       - name: Install context_agent app
         working-directory: ${{ env.APP_NAME }}
         run: |
@@ -244,6 +253,14 @@ jobs:
           [ "$STATUS" = "200" ]
           echo "$BODY"
           echo "$BODY" | grep -q 'list_calendars'
+  
+      - name: Cache llm2 models
+        uses: actions/cache/save@v5
+        env:
+          cache-name: cache-llm2-models
+        with:
+          path: llm2-persistent_storage/
+          key: ${{ steps.cache-llm2-models-restore.outputs.cache-primary-key }}
 
       - name: Run task
         env:
@@ -273,7 +290,7 @@ jobs:
         run: |
           tail data/nextcloud.log
 
-      - name: Show context_chat logs
+      - name: Show context_agent logs
         if: always()
         run: |
           [ -f ${{ env.APP_NAME }}/logs ] && cat ${{ env.APP_NAME }}/logs || echo "No context_chat logs"

ex_app/lib/agent.py

@@ -127,7 +127,7 @@ async def call_model(
 At the end of each message to the user, if you have carried out a task or answered a question, suggest up to three actions for things you can do for the user based on the tools you have available and details of the previous task. For example: If the user wants to know the weather for some location, they might be planning an event, you can suggest to create an event for them, or if they searched for a file, they may want to share it with others, suggest to create a share link for them, if they want a summary of something, you can suggest them to send the summary to somebody.
 """
 		if tool_enabled("duckduckgo_results_json"):
-			system_prompt_text += "Only use the duckduckgo_results_json tool if the user explicitly asks for a web search.\n"
+			system_prompt_text += "Use the duckduckgo_results_json tool if the user explicitly asks for a web search or you don't know about a topic or concept that the user is referencing.\n"
 		if tool_enabled("list_talk_conversations"):
 			system_prompt_text += "Use the list_talk_conversations tool to check which conversations exist.\n"
 		if tool_enabled("list_calendars"):
@@ -137,11 +137,13 @@ async def call_model(
 		if tool_enabled("find_person_in_users"):
 			system_prompt_text += "Use the find_person_in_users tool to find a person's userId and user details.\n"
 		if tool_enabled("find_details_of_current_user"):
-			system_prompt_text += "Use the find_details_of_current_user tool to find the current user's location.\n"
+			system_prompt_text += "Use the find_details_of_current_user tool to find the current user's location and timezone.\n"
 		if tool_enabled("list_mails"):
 			system_prompt_text += "Always check for the mail account id before requesting a folder list.\n"
+		if tool_enabled("web_fetch"):
+			system_prompt_text += "Use the web_fetch tool to fetch web content. You can fetch the complete page content of a duckduckgo search result using web_fetch as well.\n"
 
-		if task['input'].get('memories', None) is not None:
+		if task['input'].get('memories', None) is not None and task['input'].get('memories', None) is not []:
 			system_prompt_text += "You can remember things from other conversations with the user. If relevant, take into account the following memories:\n\n" + "\n".join(task['input']['memories']) + "\n\n"
 		# this is similar to customizing the create_react_agent with state_modifier, but is a lot more flexible
 		system_prompt = SystemMessage(

ex_app/lib/all_tools/assignments.py

@@ -0,0 +1,92 @@
+# SPDX-FileCopyrightText: 2025 Nextcloud GmbH and Nextcloud contributors
+# SPDX-License-Identifier: AGPL-3.0-or-later
+import datetime
+
+import pytz
+from langchain_core.tools import tool
+from nc_py_api import AsyncNextcloudApp
+
+from ex_app.lib.all_tools.lib.decorator import dangerous_tool, safe_tool
+
+
+async def get_tools(nc: AsyncNextcloudApp):
+
+	@tool
+	@dangerous_tool
+	async def create_scheduled_task(title: str, prompt: str, recurrence_rule: str, timezone: str|None = None, starts_at: None|str = None):
+		"""
+		Create a Scheduled Task for the assistant that will be carried out autonomously.
+		The user will still have to approve sensitive actions.
+		For example, the user could ask to transcribe new files in a certain folder every hour. Then the
+		prompt argument for this tool would be "Transcribe new files in folder /Audio" and the recurrence_rule would be "FREQ=HOURLY".
+		After having created the Scheduled Task, let the user know that the Scheduled Task will run in a newly created chat session.
+		:param title: A title for the Scheduled Task, e.g. "Transcribe audio files" -- This is only for the user's reference and has no effect on the execution of the Scheduled Task.
+		:param prompt: The instructions for the AI carrying out the Scheduled Task
+		:param recurrence_rule: An RRule compliant with RFC 5545 that defines the recurrence rule for the Scheduled Task. For example "FREQ=DAILY;INTERVAL=1" to run the Scheduled Task every day, an empty string as the recurrence_rule means the task does not repeat.
+		:param starts_at: A date time string in ISO 8601 format that defines when the Scheduled Task should start. For example "2025-01-01T09:00:00Z". If not provided, the Scheduled Task will start immediately. Make sure to use the user's timezone for this, obtainable with find_details_of_current_user
+		:param timezone: Timezone (e.g., 'America/New_York') defaults to the user's current time zone
+		:return:
+		"""
+
+		await nc.ocs('POST', f'/ocs/v2.php/apps/assistant/assignments', json={
+			'title': title,
+			'prompt': prompt,
+			'recurrence': recurrence_rule,
+			'startsAt': int(datetime.datetime.fromisoformat(starts_at.replace('Z', '+00:00')).timestamp()) if starts_at is not None else datetime.datetime.now(datetime.UTC).timestamp(),
+			'timezone': pytz.timezone(timezone).zone if timezone is not None else None
+		})
+
+		return True
+
+	@tool
+	@safe_tool
+	async def list_scheduled_tasks():
+		"""
+		List all assistant Scheduled Tasks by the current user.
+		:return:
+		"""
+
+		return await nc.ocs('GET', f'/ocs/v2.php/apps/assistant/assignments')
+
+	@tool
+	@dangerous_tool
+	async def update_scheduled_task(id: int, prompt: None|str = None, recurrence_rule: None|str = None, timezone: str|None = None, starts_at: None|str = None):
+		"""
+		Update a assistant Scheduled Task
+		:param id: The ID of the Scheduled Task to update, you can obtain this from the list_scheduled_tasks tool
+		:param prompt: The instructions for the AI carrying out the Scheduled Task. Pass `None` to leave this unchanged.
+		:param recurrence_rule: An RRule compliant with RFC 5545 that defines the recurrence rule for the Scheduled Task. For example "FREQ=DAILY;INTERVAL=1" to run the Scheduled Task every day. An empty string means, it does not repeat. Pass `None` to leave this unchanged.
+		:param timezone A timezone for the scheduled task, set to None to leave this as is.
+		:param starts_at: A date time string in ISO 8601 format that defines when the Scheduled Task should start. For example "2025-01-01T09:00:00Z". If not provided, the Scheduled Task will start immediately. Pass `None` to leave this unchanged.
+		:return:
+		"""
+
+		return await nc.ocs('PATCH', f'/ocs/v2.php/apps/assistant/assignments/{id}', json={
+			'prompt': prompt,
+			'recurrence': recurrence_rule,
+			'startsAt': int(datetime.datetime.fromisoformat(starts_at.replace('Z', '+00:00')).timestamp()) if starts_at is not None else None,
+			'timezone': pytz.timezone(timezone).zone if timezone is not None else None
+		})
+
+	@tool
+	@dangerous_tool
+	async def delete_scheduled_task(id: int):
+		"""
+		Delete a recurring Assistant Scheduled Task
+		:param id: The ID of the Scheduled Task to delete, you can obtain this from the list_scheduled_tasks tool
+		:return:
+		"""
+		return await nc.ocs('DELETE', f'/ocs/v2.php/apps/assistant/assignments/{id}')
+
+	return [
+		create_scheduled_task,
+		list_scheduled_tasks,
+		update_scheduled_task,
+		delete_scheduled_task,
+	]
+
+def get_category_name():
+	return "Assistant Scheduled Tasks"
+
+async def is_available(nc: AsyncNextcloudApp):
+	return True

ex_app/lib/all_tools/calendar.py

@@ -165,7 +165,7 @@ def find_free_time_slot_in_calendar_sync(participants: list[str], slot_duration:
 																	   headers={
 																		   "Content-Type": "text/calendar; charset=utf-8",
 																		   "Depth": "0",
-																	   }, content=freebusyRequest)
+																	   }, data=freebusyRequest)
 			print(freebusyRequest)
 			print(response.text)
 

ex_app/lib/all_tools/contacts.py

@@ -70,7 +70,7 @@ async def find_person_in_contacts(name: str) -> list[dict[str, typing.Any]]:
 			response = await nc._session._create_adapter(True).request('REPORT', f"{nc.app_cfg.endpoint}{link}", headers={
 				"Content-Type": "application/xml; charset=utf-8",
 				"Depth": "1",
-			}, content=xml_body)
+			}, data=xml_body)
 
 			if response.status_code != 207:  # Multi-Status
 				raise Exception(f"Error: {response.status_code} - {response.reason_phrase}")
@@ -98,7 +98,7 @@ async def find_person_in_contacts(name: str) -> list[dict[str, typing.Any]]:
 	@safe_tool
 	async def find_details_of_current_user() -> dict[str, typing.Any]:
 		"""
-		Find the current user's personal information
+		Find the current user's personal information, such as name, location, timezone, language
 		:return: a dictionary with the person's personal information
 		"""
 

ex_app/lib/all_tools/files.py

@@ -15,7 +15,7 @@ async def get_tools(nc: AsyncNextcloudApp):
 	@safe_tool
 	async def get_file_content(file_path: str):
 		"""
-		Get the content of a file
+		Get the content of a nextcloud-internal file of the current user
 		:param file_path: the path of the file
 		:return:
 		"""
@@ -33,7 +33,7 @@ async def get_file_content(file_path: str):
 	async def get_file_content_by_file_link(file_url: str):
 		"""
 		Get the content of a file given an internal Nextcloud link (e.g., https://host/index.php/f/12345)
-		:param file_url: the internal file URL
+		:param file_url: the nextcloud-internal file URL
 		:return: text content of the file
 		"""
 

ex_app/lib/all_tools/web.py

@@ -0,0 +1,31 @@
+# SPDX-FileCopyrightText: 2025 Nextcloud GmbH and Nextcloud contributors
+# SPDX-License-Identifier: AGPL-3.0-or-later
+import niquests
+from langchain_core.tools import tool
+from nc_py_api import AsyncNextcloudApp
+
+from ex_app.lib.all_tools.lib.decorator import safe_tool
+
+
+async def get_tools(nc: AsyncNextcloudApp):
+
+	@tool
+	@safe_tool
+	async def web_fetch(url: str) -> str:
+		"""
+		Get the contents of a web page via HTTP
+		:param url: The HTTP URL to the web page (e.g. https://nextcloud.com/team/ )
+		:return: the web page content
+		"""
+		res = await niquests.get(url)
+		return res.text()
+
+	return [
+		web_fetch,
+	]
+
+def get_category_name():
+	return "Web access"
+
+async def is_available(nc: AsyncNextcloudApp):
+	return True

ex_app/lib/mcp_server.py

@@ -8,6 +8,7 @@
 from fastmcp.server.dependencies import get_context
 from nc_py_api import AsyncNextcloudApp, NextcloudApp
 from fastmcp.server.middleware import Middleware, MiddlewareContext, CallNext
+from fastmcp.server.dependencies import get_http_headers
 from fastmcp.tools import Tool
 from mcp import types as mt
 from ex_app.lib.tools import get_tools
@@ -30,7 +31,7 @@ def get_user(authorization_header: str, nc: AsyncNextcloudApp) -> str:
 class UserAuthMiddleware(Middleware):
 	async def on_message(self, context: MiddlewareContext, call_next):
 		# Middleware stores user info in context state
-		authorization_header = context.fastmcp_context.request_context.request.headers.get("Authorization")
+		authorization_header = get_http_headers().get("authorization")
 		if authorization_header is None:
 			raise Exception("Authorization header is missing/invalid")
 		nc = AsyncNextcloudApp()

ex_app/lib/nc_model.py

@@ -37,7 +37,7 @@ class ChatWithNextcloud(BaseChatModel):
 	tools: Sequence[
 		Union[typing.Dict[str, Any], type, Callable, BaseTool]] = []
 	TIMEOUT: int = 60 * 30 # 30 minutes
-	MAX_MESSAGE_HISTORY: int = 13
+	MAX_MESSAGE_HISTORY: int = 42
 
 	def _generate(self, messages: list[BaseMessage], stop: Optional[list[str]] = None, run_manager: Optional[CallbackManagerForLLMRun] = None, **kwargs: Any):
 		raise Exception("Use _agenerate instead")