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

# FAQ

> Frequently asked questions about Pureframe AI.

## Getting started

**Do I need to create a collection before uploading?**

Yes. Every video must belong to a collection. Create one with `POST /v1/collections` first, then pass its `id` as `collection_id` in the upload request.

**How do I get an API key?**

Go to [platform.pureframe.ai/settings/api-keys](https://platform.pureframe.ai/settings/api-keys), click **New Key**, and copy it. Keys are only shown once.

***

## Upload and processing

**What video formats are supported?**

MP4, MOV, AVI, WebM, MPEG, and MKV. Maximum file size is 5 GB per upload. To increase the upload limit, [contact us](https://pureframe.ai/contact).

**How long does indexing take?**

Typically 1–3 minutes per 10 minutes of video, depending on server load. Poll `GET /v1/jobs/{job_id}` to track progress — see [Processing Status](/video-processing/processing-status).

**Can I search a video before it finishes processing?**

No. The video is only searchable once its job status reaches `done`. Searching before that returns no results for that video.

**What happens if processing fails?**

The job status becomes `failed`. Re-upload the video to create a new job. If failures persist, contact support.

***

## Search

**Can I search across all my videos at once?**

For image search, yes — omit `collection_id` to search your entire library. Text search currently requires `collection_id`.

**What does `thumbnail_base64` contain?**

A base64-encoded JPEG image of the matched video frame. Vision-capable models (Claude, GPT-4o, Gemini) can consume it directly in the same API call without fetching a URL. It's designed for agent integrations where the model needs to see the frame immediately.

**Why are presigned URLs only valid for \~1 hour?**

`video_url` and `thumbnail_url` are short-lived presigned URLs for security. Don't store them — store the `video_id` and re-fetch the URL when you need to display or stream it.

**What signal types can a search match?**

Four: `frame` (visual content), `transcript` (spoken words), `scene` (auto-detected scene labels, Pay as you go/Enterprise), and `ocr` (text visible in the frame). The `modes` request parameter controls video, transcript, and scene; OCR matches surface independent of `modes`. See [Combined Queries](/search/combined-queries).

***

## Agent Vision

**Which AI clients support the Pureframe AI MCP server?**

Claude Desktop, Claude Code, Claude.ai (via remote MCP), Cursor, VS Code, Codex, and OpenCode. Any client that supports MCP should work.

**Do I need to install anything for remote MCP?**

No. The remote MCP server at `https://mcp.pureframe.ai` runs on Pureframe AI's infrastructure. Just add the URL and your API key to your client config.

***

## Billing

**How are credits consumed?**

Each search request consumes 1 credit on Pay as you go accounts. Video processing costs $0.038 per minute of video, plus $0.005/minute if captioning is enabled. Free accounts have a daily search quota of 30 and 500 MB of storage — no credits needed. See [Usage and Billing](/production/usage-and-billing).

**What happens when I run out of credits?**

Searches return a `402` with code `INSUFFICIENT_CREDITS`. Add credits from the [Billing page](https://platform.pureframe.ai/billing) or enable auto top-up.