feat(max): deep research template system #39104

tatoalo · 2025-10-03T14:50:53Z

Problem

For Max Deep Research, one possible pain point during the onboarding phase is for the user to go through receiving a long list of questions from Max that needs to be answered. This can feel tiresome and make the experience not particularly great.

Idea: Have a list of curated + custom templates (notebooks) that the user can compile in their own time and then load as onboarding step when starting Deep Research.

This PR builds on top of the foundational piece for notebooks tags (#39101) and add backend support for Max and Notebooks API.

How did you test this code?

manual tests
unit tests

deep_research_templates_e2e.mp4

feat: deep research templates chore: mypy

greptile-apps

Additional Comments (1)

ee/hogai/graph/deep_research/onboarding/nodes.py, line 77 (link)

logic: Resetting current_run_notebooks to empty list may conflict with the skip_onboarding flow that expects notebooks to persist

_{10 files reviewed, 5 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2025-10-03T14:55:54Z

ee/hogai/assistant/assistant.py

+            assistant_initial_state = None
+            if initial_state is not None:
+                if isinstance(initial_state, (AssistantState | PartialAssistantState)):
+                    assistant_initial_state = initial_state


style: This code duplicates the type checking logic from the ASSISTANT mode. Consider extracting this pattern into a helper method to avoid repetition.

Prompt To Fix With AI

This is a comment left during a code review. Path: ee/hogai/assistant/assistant.py Line: 55:58 Comment: **style:** This code duplicates the type checking logic from the ASSISTANT mode. Consider extracting this pattern into a helper method to avoid repetition. How can I resolve this? If you propose a fix, please make it concise.

greptile-apps · 2025-10-03T14:55:55Z

ee/hogai/graph/deep_research/onboarding/nodes.py

 class DeepResearchOnboardingNode(DeepResearchNode):
    @property
    def node_name(self) -> MaxNodeName:
        return DeepResearchNodeName.NOTEBOOK_PLANNING


logic: The node_name returns NOTEBOOK_PLANNING but the class is DeepResearchOnboardingNode - this seems inconsistent and could be confusing

Prompt To Fix With AI

This is a comment left during a code review. Path: ee/hogai/graph/deep_research/onboarding/nodes.py Line: 20:20 Comment: **logic:** The node_name returns NOTEBOOK_PLANNING but the class is DeepResearchOnboardingNode - this seems inconsistent and could be confusing How can I resolve this? If you propose a fix, please make it concise.

greptile-apps · 2025-10-03T14:55:57Z

ee/hogai/graph/deep_research/notebook/nodes.py

        if not isinstance(last_message, HumanMessage):
            raise ValueError("Last message is not a human message.")

+        human_content = last_message.content if last_message else ""


logic: Redundant condition - last_message is guaranteed to exist due to the check on line 42

Suggested change

human_content = last_message.content if last_message else ""

human_content = last_message.content

Prompt To Fix With AI

This is a comment left during a code review. Path: ee/hogai/graph/deep_research/notebook/nodes.py Line: 45:45 Comment: **logic:** Redundant condition - `last_message` is guaranteed to exist due to the check on line 42 ```suggestion human_content = last_message.content ``` How can I resolve this? If you propose a fix, please make it concise.

greptile-apps · 2025-10-03T14:55:57Z

ee/hogai/graph/deep_research/notebook/nodes.py

+        if not (
+            state.template_notebook_short_id and not state.template_markdown and not state.has_emitted_template_loaded
+        ):


style: Complex nested boolean logic makes this condition hard to understand. Consider extracting to a helper method with descriptive name like should_retrieve_template_from_db.

Prompt To Fix With AI

This is a comment left during a code review. Path: ee/hogai/graph/deep_research/notebook/nodes.py Line: 107:109 Comment: **style:** Complex nested boolean logic makes this condition hard to understand. Consider extracting to a helper method with descriptive name like `should_retrieve_template_from_db`. How can I resolve this? If you propose a fix, please make it concise.

kappa90

Recap:

I think notebooks should be sent as UI context, not as a custom field, and only as an ID, not the whole notebook, this would allow us to reuse this logic to support them in base Max; if we move to the notebook template model, it can still be just an ID as UI context maybe with a template: True field, so we trigger the template flow rather than passing it as context, just an idea
Template management/streaming should be in a separate node, and I don't think the frontend needs to send a tag to say "create a custom template", it's the backend that should do that when no template is provided
There's some unused state variables and I couldn't really understand the entire flow, so some more comments would have helped

kappa90 · 2025-10-06T13:15:05Z

posthog/notebooks/curated_templates.py

+
+# Curated deep-research notebook templates stored as ProseMirror JSON
+
+CURATED_DEEP_RESEARCH_NOTEBOOKS: list[dict] = [


Ok for now, we can then move these to a model, as discussed.

kappa90 · 2025-10-06T13:16:02Z

posthog/api/notebook.py


        for key in filters:
            value = filters.get(key, None)
+            if key == "tags" and isinstance(value, str):


internal_tags to align with the other PR.

kappa90 · 2025-10-06T13:20:44Z

ee/hogai/assistant/deep_research_assistant.py

+        # and stream it immediately by setting _latest_message.
+        message_for_state = self._latest_message
+        if not self._latest_message and self._deep_research_template:
+            from uuid import uuid4


import at the top

kappa90 · 2025-10-06T13:21:18Z

ee/hogai/assistant/deep_research_assistant.py

+            if isinstance(self._deep_research_template, dict):
+                title = self._deep_research_template.get("notebook_title")
+
+            content = f"Load template: {title}" if title else "Load template"


Why do we need this message?

kappa90 · 2025-10-06T13:22:37Z

ee/hogai/assistant/deep_research_assistant.py

+            messages=[message_for_state] if message_for_state else [],
+            start_id=message_for_state.id if message_for_state else None,
+            graph_status=None,
+            notebook_short_id=None,


notebook_short_id should be removed, it's not on DeepResearchState anymore, we have add the other two notebook fields as None

kappa90 · 2025-10-06T13:50:24Z

ee/hogai/graph/deep_research/notebook/nodes.py

+            if content:
+                try:
+                    nb_json = ProsemirrorJSONContent.model_validate(notebook.content)
+                    from ee.hogai.notebook.notebook_serializer import NotebookSerializer


import at the top if possible

kappa90 · 2025-10-06T13:51:02Z

ee/hogai/graph/deep_research/notebook/nodes.py

+
+    async def _retrieve_template_markdown(self, state: DeepResearchState) -> str | None:
+        if not (
+            state.template_notebook_short_id and not state.template_markdown and not state.has_emitted_template_loaded


if not state.template_markdown -> return state.template_markdown?

why?

Also state.markdown is never set anywhere? I really don't understand the logic.

kappa90 · 2025-10-06T13:52:23Z

ee/hogai/graph/deep_research/onboarding/nodes.py


    def should_run_onboarding_at_start(self, state: DeepResearchState) -> Literal["onboarding", "planning", "continue"]:
+        # Skipping onboarding when provided a template
+        if state.skip_onboarding:


Why do we need skip_onboarding when we also have template_id in the state?

kappa90 · 2025-10-06T14:03:13Z

ee/hogai/graph/deep_research/notebook/nodes.py

+        human_content = last_message.content if last_message else ""
+        # If a template was provided, emit a synthetic "loaded notebook" message once
+        pre_messages: list = []
+        if template_markdown and not state.has_emitted_template_loaded:


Is all of this needed to create custom templates?

It seems extremely convoluted. This is the flow I would expect:

If no template is selected, the node streams the default custom template, or better, the agent writes a new draft template notebook to fill up

If template is selected, the node simply clones and streams the templated notebook to fill up

In both cases it seems like you would need an additional node that only does this, and this node should then be relegated to just the planning.

This would also help us with separation of concerns when we want to move to the Google DR flow.

kappa90 · 2025-10-06T14:06:47Z

ee/api/conversation.py

    trace_id = serializers.UUIDField(required=True)
    session_id = serializers.CharField(required=False)
    deep_research_mode = serializers.BooleanField(required=False, default=False)
+    deep_research_template = serializers.JSONField(required=False)


We're sending the whole notebook here, it's overkill

posthog-bot · 2025-10-14T07:32:29Z

This PR hasn't seen activity in a week! Should it be merged, closed, or further worked on? If you want to keep it open, post a comment or remove the stale label – otherwise this will be closed in another week. If you want to permanentely keep it open, use the waiting label.

posthog-bot · 2025-10-21T07:32:58Z

This PR was closed due to lack of activity. Feel free to reopen if it's still relevant.

feat: deep research template system

94843d4

feat: deep research templates chore: mypy

tatoalo self-assigned this Oct 3, 2025

tatoalo mentioned this pull request Oct 3, 2025

feat(max): frontend support for deep research templates #39105

Closed

1 task

tatoalo marked this pull request as ready for review October 3, 2025 14:54

tatoalo requested a review from a team October 3, 2025 14:54

tatoalo added the feature/posthog-ai label Oct 3, 2025

github-project-automation bot added this to PostHog AI Oct 3, 2025

greptile-apps bot reviewed Oct 3, 2025

View reviewed changes

tatoalo mentioned this pull request Oct 3, 2025

feat(max): notebook tags foundation #39101

Closed

kappa90 requested changes Oct 6, 2025

View reviewed changes

posthog-bot added the stale label Oct 14, 2025

posthog-bot closed this Oct 21, 2025

github-project-automation bot moved this to Done in PostHog AI Oct 21, 2025

	human_content = last_message.content if last_message else ""
	human_content = last_message.content


		# Curated deep-research notebook templates stored as ProseMirror JSON

		CURATED_DEEP_RESEARCH_NOTEBOOKS: list[dict] = [

feat(max): deep research template system #39104

feat(max): deep research template system #39104

Uh oh!

Conversation

tatoalo commented Oct 3, 2025

Problem

How did you test this code?

Uh oh!

greptile-apps bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Additional Comments (1)

Uh oh!

greptile-apps bot Oct 3, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Oct 3, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Oct 3, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Oct 3, 2025

Choose a reason for hiding this comment

Uh oh!

kappa90 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

posthog-bot commented Oct 14, 2025

Uh oh!

posthog-bot commented Oct 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

greptile-apps bot left a comment •

edited

Loading