PostHog AI modes

PostHog AI uses specialized modes to handle different types of analysis tasks. Each mode is optimized with specific tools and capabilities for its domain.

When you ask a question, the agent selects the most appropriate mode based on your request, or you can be in a specific mode based on the page you're viewing. The agent can switch between modes automatically based on your needs, and context is preserved when switching.

How mode switching works

The agent can switch modes in two ways:

1. Automatic switching - When your request requires tools from a different mode, the agent will switch automatically. For example, if you're in Product analytics mode and ask to "run a custom SQL query", the agent will switch to SQL mode.

2. Manual switching - You can explicitly select a mode with the mode selector above the chat input.

When switching modes:

• Your conversation history is preserved
• The agent gains access to the new mode's specialized tools
• The agent maintains context about what you were working on

Available modes

Product analytics

The default mode for creating insights and dashboards.

Best for:

• Creating trend, funnel, and retention insights
• Building dashboards
• Analyzing user behavior patterns

Mode-specific tools:

• create_insight - Create or edit insights
• create_dashboard / upsert_dashboard - Create and manage dashboards
• upsert_alert - Create and edit alerts on trend insights

SQL

A specialized mode for executing HogQL queries against your data.

Best for:

• Complex queries requiring custom SQL
• Exploring your data warehouse sources
• Ad-hoc data analysis
• Finding specific data that can be used in other modes

Mode-specific tools:

• execute_sql - Write and execute HogQL queries

Session replay

A mode for analyzing session recordings and user behavior.

Best for:

• Finding recordings matching specific criteria
• Understanding user behavior patterns
• Analyzing sessions with errors or specific events
• Getting AI summaries of session content

Mode-specific tools:

• filter_session_recordings - Filter recordings by properties and events
• summarize_sessions - Get AI-powered summaries of sessions

Error tracking

A mode for investigating errors and exceptions in your product.

Best for:

• Finding and analyzing error patterns
• Understanding error frequency and impact
• Investigating stack traces and root causes

Mode-specific tools:

• search_error_tracking_issues - Search and filter error tracking issues

LLM analytics

A mode for searching and analyzing LLM traces to understand model usage, costs, latency, and errors.

Best for:

• Finding and filtering LLM traces by date range, model, or error status
• Analyzing LLM usage costs and latency
• Investigating specific traces in detail
• Monitoring LLM error patterns

Mode-specific tools:

• search_llm_traces - Search for traces based on date range, model, error status, and other properties

Survey

A mode for creating, editing, and analyzing surveys.

Best for:

• Creating NPS, feedback, and custom surveys
• Targeting surveys by URL, user properties, or feature flags
• Analyzing survey responses for themes and sentiment
• Editing and managing existing surveys

Mode-specific tools:

• create_survey - Create new surveys with targeting rules
• edit_survey - Modify, stop, or archive existing surveys
• analyze_survey_responses - Extract themes, sentiment, and insights from responses

Feature flags

A mode for creating and managing feature flags and experiments.

Best for:

• Creating feature flags with rollout percentages
• Setting up property-based targeting rules
• Creating A/B test experiments with multivariate flags
• Analyzing experiment results

Mode-specific tools:

• create_feature_flag - Create feature flags with targeting and rollout settings
• create_experiment - Set up A/B tests linked to feature flags
• experiment_results_summary - Analyze experiment results with statistical summaries
• experiment_session_replays_summary - Summarize session recordings across experiment variants

Default tools (available in all modes)

Regardless of which mode is active, the agent always has access to: