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

# Prepare and import App context

> Create a provenance-safe FloKit context package using the accepted version 1 format.

FloKit can start from customer text and files even when a website is unfinished.
The package is uploaded through the authenticated dashboard; never publish
customer materials, tokens, private account IDs or customer lists in a public
page or URL.

## Workflow

1. Select the App and open **Asset Library → Context**.
2. Open **Import an agent-prepared context package**.
3. Give the agent only permitted customer materials and public sources.
4. Paste the JSON package and choose **Validate and import for review**.
5. Correct any validation errors, then review product context, audiences and
   assets before approval.

Replacing non-empty active context requires an explicit
[context restart](/guides/restart-context). Imported content is inert data and
is never executed.

## Accepted package format

The importer accepts only `schemaVersion: 1` and `containsPrivateData: false`.
Every asserted fact carries `sourceIds` naming entries in `sources`, plus an
`evidenceState`: `verified`, `user_confirmed`, `inferred`, `missing`,
or `conflicting`.

Top-level fields are:

| Field                        | Meaning                                                                           |
| ---------------------------- | --------------------------------------------------------------------------------- |
| `generatedAt`                | ISO timestamp for package creation                                                |
| `sources`                    | Customer file/text, website, store, provider, or other source references          |
| `product`                    | One-liner, description, use cases, features, benefits, and jobs to be done        |
| `audiences`                  | Named audiences with needs, motivations, objections, and provenance               |
| `messaging`                  | Value proposition, positioning, approved/forbidden claims, proof, hooks, and tone |
| `market`                     | Countries, locales, compliance facts, and product constraints                     |
| `brand`                      | Confirmed colors, typography, and voice                                           |
| `pricingFacts`               | Read-only pricing facts; never instructions to edit provider prices               |
| `media`                      | Permitted HTTP(S) media references and explicit rights confirmation               |
| `missingFacts` / `conflicts` | Explicit unknown or contradictory information                                     |

### Valid synthetic example

```json theme={null}
{
  "schemaVersion": 1,
  "containsPrivateData": false,
  "generatedAt": "2026-09-09T10:00:00.000Z",
  "sources": [
    { "sourceId": "brief-1", "type": "customer_file", "label": "Approved product brief" }
  ],
  "product": {
    "oneLiner": {
      "value": "Guided focus sessions for busy professionals.",
      "sourceIds": ["brief-1"],
      "evidenceState": "user_confirmed"
    },
    "features": {
      "values": ["Timed focus sessions", "Weekly progress review"],
      "sourceIds": ["brief-1"],
      "evidenceState": "user_confirmed"
    }
  },
  "audiences": [
    {
      "name": "Busy professionals",
      "painPoints": ["Frequent interruptions"],
      "motivations": ["Finish priority work"],
      "objections": [],
      "jobsToBeDone": ["Protect focused time"],
      "sourceIds": ["brief-1"],
      "evidenceState": "user_confirmed"
    }
  ],
  "messaging": {
    "approvedClaims": [],
    "forbiddenClaims": [],
    "proofPoints": [],
    "hooks": []
  },
  "market": {},
  "brand": {},
  "pricingFacts": [],
  "media": [],
  "missingFacts": ["Approved outcome claim"],
  "conflicts": []
}
```

Arrays omitted from the example are optional only when the schema gives them a
default. Unknown top-level fields, unknown source IDs, non-HTTP(S) URLs, and
secret-shaped content are rejected. Bundled text is always treated as inert
customer data and never executed as instructions.