# Knowledge

Source: https://agentboxd.com/docs/knowledge

> Documents your agent can search, and that reply drafts cite: policies, shipping times, how you write.

Knowledge is reference text for your agent: a refund policy, shipping times, opening hours, the tone you want. A document belongs to one inbox or, without an `inbox_id`, to all inboxes in the workspace.

## Write and search

knowledge.ts:

```ts
// Workspace-wide (no inbox_id) or for one inbox.
await mr.knowledge.create({
  title: 'Shipping times',
  body: '- Germany: 1–2 working days (DHL)
- EU: 3–5 working days',
  inbox_id: inbox.id,
});

// Ranked full-text search. With inbox_id, workspace-wide documents are included too.
const { data: hits } = await mr.knowledge.search('how long does shipping to France take', { inbox_id: inbox.id });
// [{ id, title: 'Shipping times', inbox_id, rank: 0.61, snippet: '…EU: 3–5 working days…' }]

```

knowledge.py:

```python
mr.knowledge.create(
    title="Shipping times",
    body="- Germany: 1–2 working days (DHL)
- EU: 3–5 working days",
    inbox_id=inbox["id"],
)

hits = mr.knowledge.search("how long does shipping to France take", inbox_id=inbox["id"])
for h in hits["data"]:
    print(h["title"], h["rank"], h["snippet"])

```

Search is Postgres full-text search, ranked, with a `snippet` around the matching words. It matches words, not meaning, so “refund” finds “refunds” but not “money back”. When you pass `inbox_id`, results include that inbox’s documents and the workspace-wide ones.

## Routes

| Method & path | Notes |
| --- | --- |
| `GET /v1/knowledge` | `?inbox_id=&cursor=&limit=`. List with an `excerpt` instead of the body. |
| `POST /v1/knowledge` | `{ title, body, inbox_id? }`. Plain text or markdown, up to 200,000 characters. |
| `GET /v1/knowledge/:id` · `PATCH /v1/knowledge/:id` · `DELETE /v1/knowledge/:id` | Read, change (`{ title?, body?, inbox_id? }`) or delete one document. |
| `GET /v1/knowledge/search` | `?q=&inbox_id=&limit=` → `{ data: [{ id, title, inbox_id, rank, snippet }] }`. |

## Where it’s used

- The `search_knowledge` MCP tool and `mr.knowledge.search` in the SDKs.
- [Reply drafts](https://agentboxd.com/docs/reply-drafts) use the top 5 matches for the latest inbound message and list them as `citations`.
- The [Knowledge](https://agentboxd.com/app/knowledge) page in the dashboard, with search, scope and a character count.

> **Note:** Knowledge is your text, so your agent can trust it. Keep it to what you’d tell a new colleague on day one; secrets and internal credentials don’t belong here, especially with AI processing set to `full`.