The copilot answers questions about the deal on screen — “what are the likely deal-specific exclusions?”, “which quote is best and why?”, “what will underwriters ask on the call?” — by reading the deal through tools, and citing what it read.
Seven read-only tools
| Tool | Reads | Opens |
|---|---|---|
deal_overview | The deal, its stage, what blocks the next, the narrative | The deal |
reps_and_coverage | Each rep, qualifiers, schedules, diligence behind it | Reps |
findings | Findings by category and severity, with citations | Findings |
data_room_search | Passages that answer a question, cited to the page | Documents |
quotes_and_markets | The engine’s comparison and market fit | Markets |
company_intelligence | The target’s company 360 from InsightPulse | Findings |
decision_record | Controls that fired, prepared decisions, sign-offs | Decision |
sequenceDiagram
participant B as Broker
participant CP as Copilot
participant M as Claude Sonnet
participant T as Tools
B->>CP: Which quote is best and why?
CP->>CP: Is this deal in the asker's book?
CP->>M: Plan
M->>T: quotes_and_markets, findings
T-->>M: Rows and findings, each with a source
M-->>CP: Answer citing S1, S2
CP->>CP: Every figure checked against tool results
CP-->>B: Answer with source chips
- Scope. A deal opens only if it is in the asker’s book; each tool re-checks.
- Grounded. Every figure in the answer is checked against what the tools returned.
- Reads, never acts. Suggested actions are links to the workspace’s governed actions.
- Without a model it answers with the deal’s facts, and says so.
flowchart LR
Q["Question"] --> S{"Deal in
your book?"} S -->|no| N["Refused"] S -->|yes| T["Tools"] --> A["Answer"] --> C{"Figures
grounded?"} C -->|yes| OUT["Shown with sources"] C -->|no| FIX["Replaced with
not in the record"]
your book?"} S -->|no| N["Refused"] S -->|yes| T["Tools"] --> A["Answer"] --> C{"Figures
grounded?"} C -->|yes| OUT["Shown with sources"] C -->|no| FIX["Replaced with
not in the record"]

