> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kb2b.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Scout — your first AI agent

> Scout reads your POT facts, fetches external web context, and synthesizes insights with verifiable provenance. When to use it, how to read it, and its limits.

## What Scout does

**Scout** is an AI agent inside kb2b that combines three things you'd do by hand if you had time:

1. **Reads your POT** — pulls the relevant facts already in the account.
2. **Fetches external web context** (via Tavily) — finds news, releases, recent posts about the people, companies, or topics mentioned in those facts.
3. **Synthesizes insights** with Claude — combines both sides and produces **3-5 insight cards** with inline citations to your facts (`[fact:X]`) and external sources (`[src:N]`).

Each insight tells you something actionable: a connection you hadn't seen, a relevant shift in the customer's context, a question worth asking in the next meeting.

## When to use it

* **Before a meeting** with a customer: arrive with fresh context beyond what's already in the POT.
* **After a customer event** (funding round, announcement, executive departure): to understand what's changing.
* **In quarterly reviews**: as input for discussion about what's happening around the account.

**Not for** factual questions about the POT — use chat for that. Scout is for **discovery**, not lookup.

## When NOT to use it

* If your POT has **fewer than 5 facts**, Scout can't run — there isn't enough material to derive useful queries.
* If the account is **very generic or private** (not findable online), web searches won't return anything relevant.
* If you need **closed** information (internal customer data that isn't public), Scout won't find it — use chat over the loaded documents.

## How to run it

1. Navigate to `/dashboard/pot/agents` in kb2b.
2. Click **"Run Scout"**.
3. Scout kicks off a background run and shows a screen with the **thinking log** live:
   * Phase 1: reading POT facts
   * Phase 2: generating search queries
   * Phase 3: searching the web
   * Phase 4: synthesizing insights
4. When done (typically 30-90 seconds), an **insight card carousel** appears.

Each card shows:

* The insight in a short sentence
* The cited POT facts (with their POT Score visible)
* The cited external sources (with link)
* A CTA **"Ask your knowledge base"** that opens the chat with the context pre-loaded

## History

Every run is saved in **History** inside `/dashboard/pot/agents`. You can:

* Reopen previous runs
* Compare what changed between two runs (manually — automatic diff coming in Phase 2)
* Export the result (Phase 2)

## Limits

Scout has rate limits to protect your token quota and the Tavily API:

| Limit                        | Value |
| ---------------------------- | ----- |
| Minimum POT facts            | 5     |
| Runs per hour (organization) | 5     |
| Runs per user in 10 minutes  | 2     |

If you hit a limit, Scout tells you **when you can run again**. No tokens are charged if a run doesn't complete.

## Token cost

Each run consumes plan tokens:

* **Query generation** (Claude Haiku): low cost, \~5-10K tokens
* **Insight synthesis** (Claude Sonnet): main cost, \~30-50K tokens per run depending on POT size

If your workspace is near the token limit, Scout warns you before running. More detail in [Token limits](/en/admin/token-limits).

## Why is it called "Scout" and not "Researcher"?

Internally the feature ships as `agent_kind = "research"` because it's designed as the first of several exploration tools. **Scout** is the product name — the idea is a quick explorer that **brings things to you**, not a deep research assistant. If it surfaces something interesting, you decide whether it's worth digging deeper (with chat, another meeting, additional documents).

*Page in progress Phase 1 — insight card and thinking log screenshots coming in Phase 2.*