Skip to content
AM

The conversation theatre

The Portal root page is your AI COO’s operating surface. A briefing can be pinned above the chat, the conversation stays in the main transcript, and the bottom dock switches between chat, code, and audit modes.

What this is

The Conversation Theatre is implemented by AiCooPage and TheatreLayout. In the Portal app it is mounted at /, and the command palette action “Open AI COO” navigates there.

The page has three modes:

  • Chat - the regular assistant transcript, with pinned content rendered above the message stream.
  • Code - the code-mode transcript for code-mode planning and execution output.
  • Audit - an AuditCore placeholder panel for provenance-oriented review.

The page-level briefing fetch has explicit states: loading renders BriefingSkeletonCard, failure renders AiCooErrorMessage, empty renders no pinned content, and ready renders BriefingCard.

When you’d use it

  • First thing in the morning - review the pinned briefing when commerce dashboard data is available.
  • When you need the agent to investigate or do work - ask questions, request analysis, or start an action that may create a pending action.
  • When you need evidence - ask why something happened. The agent can use provenance tooling and return citation handles that the Portal renders as chips.
  • When you need code-mode output - switch to Code mode to review the code-mode transcript separately from ordinary chat.

Walkthrough

The briefing. The current briefing source is the commerce dashboard stats query for the last 30 days. The UI maps those stats into a pinned briefing card when data is ready. The card is not a standalone MCP tool; it is composed by the Portal page from product data.

The transcript. Chat mode renders the normal AI COO transcript. The agent only receives the tools that pass tenant, role, entitlement, feature, GL-layer, and policy visibility checks. See Pillars, tiers, and tool visibility.

Mode switching. The bottom dock exposes Chat, Code, and Audit modes. Switching modes changes the content in the main region without leaving the Portal shell.

Citation chips. When an answer includes a supported citation handle, Portal renders it as a chip and deep-links to the relevant product surface:

  • [auditcore:event:<id>] - Accounting agent activity.
  • [workflow:<id>] - Accounting agent activity.
  • [rule:<id-or-version>] - Portal rule package surface.
  • [principal:<id>] - Portal team or role surface.

Unsupported or malformed handles remain plain text. If provenance evidence is incomplete, the explanation contract requires the assistant to surface the limitation instead of inventing a causal chain.

What changes by tier

Tool visibility changes by plan, tenant features, role entitlements, and agent policy. At a high level:

TierWhat the agent can see
CoreAccounting, reporting, and portal surfaces that pass role and policy checks
GrowthAdds commerce and CX surfaces when enabled
Scale / EnterpriseAdds the broader pillar set, including Studio where enabled
PracticePractice/accounting scope unless additional entitlements are granted

See Plans, tiers, and entitlements for the canonical plan table.

Programmatic access

  • Agent contract: external and internal agents use MCP for AI COO tools. See The InBlack agentic surface and Quickstart: direct HTTP MCP.
  • Provenance explanations: inblack_explain_provenance returns the structured explanation result used for grounded “why” answers.
  • Briefing: there is no single morning-briefing MCP tool in the current source. The Portal page composes the pinned briefing from product queries.
  • Agent policy: the operator policy controls are documented on Agent activity timeline.