Outlit

Use Outlit tools to ground customer intelligence work in customer data. Outlit joins product activity, conversations, billing, and web signals into a unified customer context graph and timeline for agents.

Use the registered outlit_* tools as the interface. Do not tell the user to install the Outlit CLI or configure MCP from inside Pi unless they explicitly ask about a different agent environment.

Do not invent customer state when Outlit can answer it. Call out sparse or messy data instead of overstating confidence.

Tool Choice

• Use outlit_list_customers to discover customers for portfolio, segment, risk, revenue, trial, or account-health questions.
• Use outlit_list_users for user-level questions or when a customer answer depends on individual users.
• Use outlit_get_customer before deep analysis of a named customer or account.
• Use outlit_get_timeline when order, recency, activity sequence, meetings, messages, product usage, support, or billing chronology matters.
• Use outlit_list_facts to browse structured account facts, known signals, open issues, health indicators, relationship notes, activation, billing, or renewal context. Narrow with status, sourceTypes, and factTypes when you know what evidence class you need.
• Use outlit_get_fact when you already have a fact id and need the canonical fact payload.
• Use outlit_search_customer_context for fuzzy or thematic questions such as pricing concern, blocked integration, not using, renewal, champion left, negative sentiment, expansion, implementation, or support escalation.
• Use outlit_get_source when a fact or search result needs stronger evidence from the underlying source artifact.

Use customer lookups before SQL. SQL is for aggregates, joins, cohorts, time-series analysis, and custom reporting.

Facts vs Search vs Timeline

• Use outlit_list_facts to list what Outlit already knows about an account.
• Use factTypes for specific extracted fact classes such as CHURN_RISK, EXPANSION, SENTIMENT, BUDGET, REQUIREMENTS, PRODUCT_USAGE, or CHAMPION_RISK when those are relevant. Do not request anomaly detector fact types such as CORE_ACTION_DECAY, CADENCE_BREAK, QUIET_ACCOUNT, ACTIVATION_RATE_DROP, or FUNNEL_DROPOFF as filters.
• Use outlit_get_fact when you already have a fact id and need that exact fact.
• Use outlit_search_customer_context for a specific question or theme, including cross-customer questions.
• Use outlit_get_source when you need the exact email, call, calendar event, ticket, or other source artifact behind a fact or search hit.
• Use outlit_get_timeline when order and sequence matter.

SQL Tools

Only use SQL if outlit_schema and outlit_query are available in the current Pi session.

• Call outlit_schema before writing SQL.
• Prefer customer lookups, facts, timeline, and search before SQL for account-specific analysis.
• Use SQL for aggregates, cohorts, joins, custom reporting, and questions that need exact counts or revenue math.
• Use the query patterns in the SQL reference. Do not assume another database's date or JSON helpers.
• Add explicit time filters for event queries.
• Use LIMIT.
• Divide money fields in cents by 100 for display.
• Request only the fields needed for the answer.

For supported query patterns, read references/sql-reference.md.

Working Rules

• Start with the highest-level tool that can answer the question.
• Gather evidence before drawing conclusions.
• Separate evidence from interpretation in the final answer.
• Cite the kind of evidence used, such as timeline event, fact, search result, source, customer record, user record, or SQL result.
• If tools return empty or inconsistent data, say what is missing and how that affects confidence.
• Do not send messages, create tasks, update CRM records, or take external actions unless the user explicitly asks and the necessary tools are available.
• If an Outlit tool errors because OUTLIT_API_KEY is missing, tell the user to set OUTLIT_API_KEY and retry.

Output

Keep answers concise and evidence-backed. Results often include timestamps and source attribution when available; use those details in the final answer.

For analyses, include:

• the scope you reviewed
• the most important signals
• the supporting evidence
• the confidence level
• any open questions or missing data

Common Prompts

• "What changed for this customer this week?"
• "Who is paying but inactive for 30 days?"
• "What pricing objections show up in conversations?"
• "Which channels are driving revenue?"