Skip to content

Commit

Permalink
Allow research mode and other conversation commands in automations
Browse files Browse the repository at this point in the history 
Previously we couldn't enable research mode or use other slash
commands in automated tasks.

This change separates determining if a chat query is triggered via
automated task from the (other) conversation commands to run the query
with.

This unlocks the ability to enable research mode in automations
apart from other variations like /image or /diagram etc.
  • Loading branch information
@debanjum
debanjum committed Dec 17, 2024
1 parent 88aa8c3 commit 20eafa9
Show file tree
Hide file tree
Showing 3 changed files with 39 additions and 34 deletions.
8 changes: 4 additions & 4 deletions src/interface/web/app/automations/page.tsx
View file
Original file line number Diff line number Diff line change
Expand Up @@ -165,7 +165,7 @@ const suggestedAutomationsMetadata: AutomationsData[] = [
{
subject: "Weekly Newsletter",
query_to_run:
"Compile a message including: 1. A recap of news from last week 2. An at-home workout I can do before work 3. A quote to inspire me for the week ahead",
"/research Compile a message including: 1. A recap of news from last week 2. An at-home workout I can do before work 3. A quote to inspire me for the week ahead",
schedule: "9AM every Monday",
next: "Next run at 9AM on Monday",
crontime: "0 9 * * 1",
Expand All @@ -185,7 +185,7 @@ const suggestedAutomationsMetadata: AutomationsData[] = [
{
subject: "Front Page of Hacker News",
query_to_run:
"Summarize the top 5 posts from https://news.ycombinator.com/best and share them with me, including links",
"/research Summarize the top 5 posts from https://news.ycombinator.com/best and share them with me, including links",
schedule: "9PM on every Wednesday",
next: "Next run at 9PM on Wednesday",
crontime: "0 21 * * 3",
Expand All @@ -195,7 +195,7 @@ const suggestedAutomationsMetadata: AutomationsData[] = [
{
subject: "Market Summary",
query_to_run:
"Get the market summary for today and share it with me. Focus on tech stocks and the S&P 500.",
"/research Get the market summary for today and share it with me. Focus on tech stocks and the S&P 500.",
schedule: "9AM on every weekday",
next: "Next run at 9AM on Monday",
crontime: "0 9 * * *",
Expand All @@ -214,7 +214,7 @@ const suggestedAutomationsMetadata: AutomationsData[] = [
{
subject: "Round-up of research papers about AI in healthcare",
query_to_run:
"Summarize the top 3 research papers about AI in healthcare that were published in the last week. Include links to the full papers.",
"/research Summarize the top 3 research papers about AI in healthcare that were published in the last week. Include links to the full papers.",
schedule: "9AM every Friday",
next: "Next run at 9AM on Friday",
crontime: "0 9 * * 5",
Expand Down
14 changes: 7 additions & 7 deletions src/khoj/routers/api_chat.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,7 @@
gather_raw_query_files,
generate_excalidraw_diagram,
generate_summary_from_files,
get_conversation_command,
get_conversation_commands,
is_query_empty,
is_ready_to_chat,
read_chat_stream,
Expand Down Expand Up @@ -724,7 +724,7 @@ def collect_telemetry():
yield result
return

conversation_commands = [get_conversation_command(query=q, any_references=True)]
conversation_commands, is_automated_task = get_conversation_commands(query=q)

conversation = await ConversationAdapters.aget_conversation_by_user(
user,
Expand Down Expand Up @@ -757,11 +757,8 @@ def collect_telemetry():
location = None
if city or region or country or country_code:
location = LocationData(city=city, region=region, country=country, country_code=country_code)

user_message_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

meta_log = conversation.conversation_log
is_automated_task = conversation_commands == [ConversationCommand.AutomatedTask]

researched_results = ""
online_results: Dict = dict()
Expand All @@ -778,7 +775,7 @@ def collect_telemetry():
generated_excalidraw_diagram: str = None
program_execution_context: List[str] = []

if conversation_commands == [ConversationCommand.Default] or is_automated_task:
if conversation_commands == [ConversationCommand.Default]:
chosen_io = await aget_data_sources_and_output_format(
q,
meta_log,
Expand All @@ -799,7 +796,10 @@ def collect_telemetry():
async for result in send_event(ChatEvent.STATUS, f"**Selected Tools:** {conversation_commands_str}"):
yield result

for cmd in conversation_commands:
cmds_to_rate_limit = conversation_commands
if is_automated_task:
cmds_to_rate_limit += [ConversationCommand.AutomatedTask]
for cmd in cmds_to_rate_limit:
try:
await conversation_command_rate_limiter.update_and_check_if_valid(request, cmd)
q = q.replace(f"/{cmd.value}", "").strip()
Expand Down
51 changes: 28 additions & 23 deletions src/khoj/routers/helpers.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -231,34 +231,39 @@ def get_next_url(request: Request) -> str:
return urljoin(str(request.base_url).rstrip("/"), next_path)


def get_conversation_command(query: str, any_references: bool = False) -> ConversationCommand:
def get_conversation_commands(query: str) -> tuple[list[ConversationCommand], bool]:
# Intialize variables
commands: list[ConversationCommand] = []
is_automated_task = False

# Automated Task can be mixed with other conversation commands
if query.startswith("/automated_task"):
is_automated_task = True
query = query.replace("/automated_task", "").lstrip()

if query.startswith("/notes"):
return ConversationCommand.Notes
commands += [ConversationCommand.Notes]
elif query.startswith("/help"):
return ConversationCommand.Help
commands += [ConversationCommand.Help]
elif query.startswith("/general"):
return ConversationCommand.General
commands += [ConversationCommand.General]
elif query.startswith("/online"):
return ConversationCommand.Online
commands += [ConversationCommand.Online]
elif query.startswith("/webpage"):
return ConversationCommand.Webpage
commands += [ConversationCommand.Webpage]
elif query.startswith("/image"):
return ConversationCommand.Image
elif query.startswith("/automated_task"):
return ConversationCommand.AutomatedTask
commands += [ConversationCommand.Image]
elif query.startswith("/summarize"):
return ConversationCommand.Summarize
commands += [ConversationCommand.Summarize]
elif query.startswith("/diagram"):
return ConversationCommand.Diagram
commands += [ConversationCommand.Diagram]
elif query.startswith("/code"):
return ConversationCommand.Code
commands += [ConversationCommand.Code]
elif query.startswith("/research"):
return ConversationCommand.Research
# If no relevant notes found for the given query
elif not any_references:
return ConversationCommand.General
commands += [ConversationCommand.Research]
else:
return ConversationCommand.Default
commands += [ConversationCommand.Default]
return commands, is_automated_task


async def agenerate_chat_response(*args):
Expand Down Expand Up @@ -411,14 +416,14 @@ async def aget_data_sources_and_output_format(
selected_sources = [q.strip() for q in response.get("source", []) if q.strip()]
selected_output = response.get("output", "text").strip() # Default to text output

if not isinstance(selected_sources, list) or not selected_sources or len(selected_sources) == 0:
if is_none_or_empty(selected_sources) or not isinstance(selected_sources, list):
raise ValueError(
f"Invalid response for determining relevant tools: {selected_sources}. Raw Response: {response}"
)

result: Dict = {"sources": [], "output": None if not is_task else ConversationCommand.AutomatedTask}
result: Dict = {"sources": [], "output": ConversationCommand.Text}
for selected_source in selected_sources:
# Add a double check to verify it's in the agent list, because the LLM sometimes gets confused by the tool options.
# Verify selected data sources are enabled for the agent, as the LLM can sometimes get confused by the tool options.
if (
selected_source in source_options.keys()
and isinstance(result["sources"], list)
Expand All @@ -427,12 +432,12 @@ async def aget_data_sources_and_output_format(
# Check whether the tool exists as a valid ConversationCommand
result["sources"].append(ConversationCommand(selected_source))

# Add a double check to verify it's in the agent list, because the LLM sometimes gets confused by the tool options.
# Verify selected output mode is enabled for the agent, as the LLM can sometimes get confused by the tool options.
if selected_output in output_options.keys() and (len(agent_outputs) == 0 or selected_output in agent_outputs):
# Check whether the tool exists as a valid ConversationCommand
result["output"] = ConversationCommand(selected_output)

if is_none_or_empty(result):
if is_none_or_empty(result) or is_none_or_empty(result.get("sources")):
if len(agent_sources) == 0:
result = {"sources": [ConversationCommand.Default], "output": ConversationCommand.Text}
else:
Expand Down Expand Up @@ -1686,7 +1691,7 @@ def scheduled_chat(
last_run_time = datetime.strptime(last_run_time, "%Y-%m-%d %I:%M %p %Z").replace(tzinfo=timezone.utc)

# If the last run time was within the last 6 hours, don't run it again. This helps avoid multithreading issues and rate limits.
if (datetime.now(timezone.utc) - last_run_time).total_seconds() < 21600:
if (datetime.now(timezone.utc) - last_run_time).total_seconds() < 6 * 60 * 60:
logger.info(f"Skipping scheduled chat {job_id} as the next run time is in the future.")
return

Expand Down

0 comments on commit 20eafa9

Please sign in to comment.