# Feature: Client Memory System (Persistent Intelligence per Client)

[nerkat]

## Goal

Add a **Client Memory system** that maintains persistent structured intelligence about a client across multiple calls.

Each new call analysis should generate a **memory patch** that updates the client's memory snapshot.

This allows the system to provide:

- Client context
- Recurring risks
- Decision style
- Budget signals
- Timeline signals
- Key people
- Briefing bullets for "Prep for Call"

This memory becomes the foundation for **deal intelligence across time**, not just per-call analysis.

---

# Architecture

We will introduce two data layers:

### 1. Call Analysis (already exists)

Each call already stores structured AI output.

We will reuse this.

---

### 2. Client Memory (new)

Single document per client.

Collection:

clientMemory/{clientId}

Structure:

```json
{
  "clientId": "string",
  "memoryTags": ["string"],
  "decisionStyle": "string",
  "budgetSignals": "string",
  "timelineSignals": "string",
  "recurringRisks": ["string"],
  "keyPeople": [
    {
      "name": "string",
      "role": "string | null",
      "notes": "string | null"
    }
  ],
  "briefingBullets": ["string"],
  "lastUpdatedAt": "timestamp"
}

---

Memory Update Flow

When a call analysis completes:

1. Load existing client memory
2. Generate a memory patch from the call analysis
3. Merge patch into client memory
4. Save updated memory document

---

Memory Patch Schema

AI should generate a patch object like:

{
  "newTags": ["string"],
  "riskSignals": ["string"],
  "peopleUpdates": [
    {
      "name": "string",
      "role": "string | null",
      "notes": "string | null"
    }
  ],
  "budgetSignal": "string | null",
  "timelineSignal": "string | null",
  "briefingUpdates": ["string"]
}

---

Merge Logic

When applying a patch:

Tags

Add new tags if not already present.

Risks

Append new recurring risks if relevant.

People

Merge by name:

• update role/notes if provided
• avoid duplicates

Budget / Timeline

If patch provides updated signal → replace previous.

Briefing Bullets

Keep max 5–8 bullets.

Replace older ones if new signal is stronger.

---

API

Add endpoint:

POST /api/update-client-memory

Input:

{
  clientId: string,
  callAnalysisId: string
}

Steps:

1. Load analysis
2. Load existing memory
3. Generate memory patch (AI)
4. Merge
5. Save

Return updated memory.

---

UI Changes

On Client page:

Add Client Memory section showing:

• Memory tags
• Decision style
• Recurring risks
• Key people
• Briefing bullets

Add button:

Refresh Memory

Triggers memory update.

---

Future Use

Client memory will power:

• Prep for Call
• Story So Far
• Deal trajectory insights

---

Acceptance Criteria

• Client memory document created automatically
• Memory updates after each analyzed call
• Patch merge prevents duplicates
• Client page shows memory snapshot
• No existing features break