Intelligence

The Intelligence area gives operators visibility into how the platform is used: what it costs, what users are asking, and where the knowledge base falls short. It lives under /admin/console/intelligence.

Usage & cost

At /admin/console/intelligence/usage, a platform-wide view of AI consumption and its cost.

• A period selector scopes the view to a month.
• Platform totals summarize overall usage.
• A usage-by-organization table lists each tenant's plan, runs used against the run cap, tokens in/out, estimated cost, an MRR proxy, and the cost-to-revenue ratio. Columns are sortable, and rows where cost runs high relative to revenue are highlighted.
• A near-cap watchlist surfaces organizations approaching their monthly run cap.

Backed by operatorIntelligence.usage.getPlatformDashboard, usage.listByOrg, and usage.listNearCap.

Conversation oversight

At /admin/console/intelligence/conversations, operators can review advisor conversations across tenants.

• Filter by organization, channel (web / whatsapp), and whether the conversation contains a refusal; search by title or query.
• The list (cursor-paginated) shows each conversation's organization, channel, last activity, and refusal/message counts.
• Opening a conversation shows the full message timeline, the citations behind each answer, and any refusal indicators.

Backed by operatorIntelligence.conversations.list and conversations.get. Access to these views is itself audited.

Knowledge gaps

At /admin/console/intelligence/gaps, the Knowledge Gap Radar ranks the topics the advisor couldn't answer, derived from refused turns.

• Ranked by refusal count, each row shows the topic, a sample question, how many refusals it drew (as a bar), which organizations were affected, and when it last occurred.
• An optional date filter limits the window.
• Each gap links to the Global Library so an operator can add content that closes it.

Backed by operatorIntelligence.gaps.listRanked, which reads from the refusal index on advisor messages described in the data model.

AI models

At /operator/ai-model-config, assign per-task model overrides. Every AI task — extraction, the Advisor, gap review, report narrative, embedding — runs on its default model unless an override is set here, so a stronger model can be used for written narrative and a cheaper one for routine work.

Model Lab

At /operator/model-lab, run the Advisor across two to four models side by side on the same question. Use it to compare candidates before committing a model to a task under AI models above.

KPI Reference Library

At /operator/kpi-library, manage the global master KPI list used to match LP template fields and to supply aggregation rules and methodology prose to funds' dictionaries.

See usage & audit for the review routine.