:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
from typing import cast
from posthog.schema import AgentMode
from posthog.models import Team, User
from ee.hogai.chat_agent.prompt_builder import ChatAgentPlanPromptBuilder, ChatAgentPromptBuilder
from ee.hogai.chat_agent.toolkit import (
ChatAgentPlanToolkit,
ChatAgentToolkit,
ChatAgentToolkitManager,
PlanModeSwitchAgentToolkit,
)
from ee.hogai.context import AssistantContextManager
from ee.hogai.core.agent_modes.factory import AgentModeDefinition
from ee.hogai.core.agent_modes.mode_manager import AgentModeManager
from ee.hogai.core.agent_modes.presets.ai_observability import (
ai_observability_agent,
chat_agent_plan_ai_observability_agent,
)
from ee.hogai.core.agent_modes.presets.customer_analytics import customer_analytics_agent
from ee.hogai.core.agent_modes.presets.error_tracking import chat_agent_plan_error_tracking_agent, error_tracking_agent
from ee.hogai.core.agent_modes.presets.flags import chat_agent_plan_flags_agent, flags_agent
from ee.hogai.core.agent_modes.presets.product_analytics import (
chat_agent_plan_product_analytics_agent,
product_analytics_agent,
subagent_product_analytics_agent,
)
from ee.hogai.core.agent_modes.presets.session_replay import chat_agent_plan_session_replay_agent, session_replay_agent
from ee.hogai.core.agent_modes.presets.sql import chat_agent_plan_sql_agent, sql_agent
from ee.hogai.core.agent_modes.presets.survey import chat_agent_plan_survey_agent, subagent_survey_agent, survey_agent
from ee.hogai.core.agent_modes.prompt_builder import AgentPromptBuilder
from ee.hogai.core.agent_modes.toolkit import AgentToolkit, AgentToolkitManager
from ee.hogai.utils.feature_flags import has_customer_analytics_mode_feature_flag, has_plan_mode_feature_flag
from ee.hogai.utils.types.base import AssistantState, NodePath
# Execution and plan mode definitions - fictitious modes used to trigger transition in and out of plan mode
execution_agent = AgentModeDefinition(
mode=AgentMode.EXECUTION,
mode_description="Switch to this mode when the user has approved your plan to proceed with execution.",
toolkit_class=PlanModeSwitchAgentToolkit,
)
plan_agent = AgentModeDefinition(
mode=AgentMode.PLAN,
mode_description="Switch to this mode when you need to plan a complex task that requires multiple steps and approvals.",
toolkit_class=PlanModeSwitchAgentToolkit,
)
# Default mode registry for normal chat agent operation
DEFAULT_CHAT_AGENT_MODE_REGISTRY: dict[AgentMode, AgentModeDefinition] = {
AgentMode.PRODUCT_ANALYTICS: product_analytics_agent,
AgentMode.SQL: sql_agent,
AgentMode.SESSION_REPLAY: session_replay_agent,
AgentMode.ERROR_TRACKING: error_tracking_agent,
AgentMode.FLAGS: flags_agent,
AgentMode.SURVEY: survey_agent,
AgentMode.LLM_ANALYTICS: ai_observability_agent,
}
DEFAULT_CHAT_AGENT_PLAN_MODE_REGISTRY: dict[AgentMode, AgentModeDefinition] = {
AgentMode.PRODUCT_ANALYTICS: chat_agent_plan_product_analytics_agent,
AgentMode.SQL: chat_agent_plan_sql_agent,
AgentMode.SESSION_REPLAY: chat_agent_plan_session_replay_agent,
AgentMode.EXECUTION: execution_agent,
AgentMode.ERROR_TRACKING: chat_agent_plan_error_tracking_agent,
AgentMode.FLAGS: chat_agent_plan_flags_agent,
AgentMode.SURVEY: chat_agent_plan_survey_agent,
AgentMode.LLM_ANALYTICS: chat_agent_plan_ai_observability_agent,
}
SUBAGENT_CHAT_AGENT_MODE_REGISTRY: dict[AgentMode, AgentModeDefinition] = {
AgentMode.PRODUCT_ANALYTICS: subagent_product_analytics_agent,
AgentMode.SQL: sql_agent,
AgentMode.SESSION_REPLAY: session_replay_agent,
AgentMode.ERROR_TRACKING: error_tracking_agent,
AgentMode.FLAGS: chat_agent_plan_flags_agent,
AgentMode.SURVEY: subagent_survey_agent,
AgentMode.LLM_ANALYTICS: chat_agent_plan_ai_observability_agent,
}
def get_plan_mode_registry(team: Team, user: User) -> dict[AgentMode, AgentModeDefinition]:
return dict(DEFAULT_CHAT_AGENT_PLAN_MODE_REGISTRY)
def get_execution_mode_registry(team: Team, user: User) -> dict[AgentMode, AgentModeDefinition]:
"""Build the mode registry for execution mode (non-plan, non-subagent).
This is the registry that will be available after transitioning out of plan mode.
"""
registry = dict(DEFAULT_CHAT_AGENT_MODE_REGISTRY)
if has_customer_analytics_mode_feature_flag(team, user):
registry[AgentMode.CUSTOMER_ANALYTICS] = customer_analytics_agent
return registry
class ChatAgentModeManager(AgentModeManager):
def __init__(
self,
*,
team: Team,
user: User,
node_path: tuple[NodePath, ...],
context_manager: AssistantContextManager,
state: AssistantState,
):
self._is_subagent = context_manager.is_subagent
# Set _mode and _supermode before super().__init__ because the parent
# calls self.mode_registry which accesses these attributes.
if state.agent_mode == AgentMode.PLAN:
self._supermode: AgentMode | None = AgentMode.PLAN
self._mode = AgentMode.PRODUCT_ANALYTICS
else:
self._supermode = cast(AgentMode | None, state.supermode)
self._mode = state.agent_mode or AgentMode.PRODUCT_ANALYTICS
super().__init__(
team=team,
user=user,
node_path=node_path,
context_manager=context_manager,
state=state,
)
# Validate mode is in registry, fall back if the feature flag is off.
if self._mode not in self.mode_registry:
self._mode = AgentMode.PRODUCT_ANALYTICS
@property
def mode_registry(self) -> dict[AgentMode, AgentModeDefinition]:
if self._is_subagent:
return self._subagent_mode_registry
if self._supermode == AgentMode.PLAN:
return get_plan_mode_registry(self._team, self._user)
registry = get_execution_mode_registry(self._team, self._user)
if has_plan_mode_feature_flag(self._team, self._user):
registry[AgentMode.PLAN] = plan_agent
return registry
@property
def _subagent_mode_registry(self) -> dict[AgentMode, AgentModeDefinition]:
return dict(SUBAGENT_CHAT_AGENT_MODE_REGISTRY)
@property
def prompt_builder_class(self) -> type[AgentPromptBuilder]:
if self._supermode == AgentMode.PLAN:
return ChatAgentPlanPromptBuilder
return ChatAgentPromptBuilder
@property
def toolkit_class(self) -> type[AgentToolkit]:
if self._supermode == AgentMode.PLAN:
return ChatAgentPlanToolkit
return ChatAgentToolkit
@property
def toolkit_manager_class(self) -> type[AgentToolkitManager]:
return ChatAgentToolkitManager