The Short Answer
iconik search matches metadata: titles, custom fields, tags and transcripts. To find a moment by what happens on screen, such as "the wide shot of the empty stadium at dusk", a face, a logo or a line someone said, you need an index built from the footage itself.
In November 2026 there are three ways to get one:
This guide compares the three and walks through connecting iconik to Mixpeek.
What iconik searches on its own
According to iconik's AI page, the built-in features are:
Administrators switch these on, and visual analysis runs on proxies, so full-resolution originals are not sent to AI providers. iconik's IBC 2026 preview (8 September 2026) added AI metadata enrichment, where a person reviews suggested field values, and natural language search, labelled beta, in plain English and more than 70 other languages.
Tags and transcripts answer "which assets contain a stadium" and "where did someone say kickoff". A moment described in a sentence, like "the coach walks off before the final whistle", needs search over the pictures themselves. The next two options add that.
iconik and TwelveLabs
TwelveLabs named iconik its first Premier Tier ecosystem partner on 10 September 2026. Its announcement says iconik customers already use TwelveLabs semantic search through custom actions and browser-based tooling, and that the native connector, which puts semantic search into iconik's search next to its lexical and metadata search, is in early access. You get on the list through your iconik account manager or TwelveLabs sales. Results come back as clips with timecodes inside iconik.
The same week, AWS named iconik among the first customers using TwelveLabs Marengo through Amazon Bedrock Knowledge Bases. Marengo 3.0 on Amazon Bedrock covers what that managed path does with the vectors.
Mixpeek: an index of your iconik library you can query from anywhere
Mixpeek connects to iconik with an App ID and an Auth Token. It checks for new and changed assets on a schedule (every 5 minutes by default), downloads each asset's proxy, and runs the extractors you choose: scene embeddings, face identity, object detection, speech transcription and on-screen text (OCR). Every result carries the iconik asset ID, title, collection path and your custom metadata fields, so a search can filter on them and link straight back to the asset.
You query the index through the API, the MCP server for AI agents, or Studio. The connector reads from iconik and does not write tags back into it.
Choose this path when you need one or more of these:
Which option to choose
| iconik built-in AI | iconik with TwelveLabs | Mixpeek on iconik | |
|---|---|---|---|
| Where you search | iconik interface | iconik interface | API, MCP, Studio or your own app |
| What gets indexed | Transcripts, faces, objects and scenes as metadata | Video embeddings from TwelveLabs models | Scene embeddings, faces, speech, on-screen text and objects, with iconik metadata |
| Status, November 2026 | Available; natural language search in beta | Custom actions today; native connector in early access | Connector available |
| Other storage in the same index | iconik only | iconik only | iconik plus S3, GCS, Azure Blob, Box, SharePoint and more |
| Suits | Editors who search in iconik | iconik teams who want moment search in the iconik interface | Teams building search into a product, an agent or a pipeline |
You can run more than one. The Mixpeek sync only reads assets and proxies, so it does not interfere with iconik's own analysis or a TwelveLabs connector on the same library.
Steps to connect iconik to Mixpeek
1. Create an iconik application: in iconik go to Admin > Integrations > Applications, click Create Application, and copy the App ID and Auth Token. An application token keeps working when the person who made it leaves. 2. Create the connection in Mixpeek: in Studio open Settings > Storage Connections > Add Connection and pick iconik, or call the API with
provider_type set to iconik and test_before_save on.
3. Create a bucket with a sync from that connection, and scope it with provider_filters: collection IDs, asset status, media type, or path_patterns globs on the collection path such as *UGC*.
4. Attach a collection with the extractors you need. The footage intelligence template sets up scene search for video libraries.
5. Search with a retriever: a sentence, an image or a short clip, filtered on iconik fields such as directory_path or a custom metadata field.
6. Optionally set a webhook_secret on the connection and add an iconik webhook, so deletes and updates reach Mixpeek straight away. Without it, each sync reconciles deletes on its next pass.from mixpeek import Mixpeek
client = Mixpeek(api_key="YOUR_API_KEY")
connection = client.organizations.connections.create(
name="iconik Production",
provider_type="iconik",
provider_config={
"credentials": {
"auth_token": "YOUR_ICONIK_AUTH_TOKEN",
"app_id": "YOUR_ICONIK_APP_ID"
}
},
test_before_save=True
)The full field reference is in the iconik integration docs.
The sync downloads proxies, never originals. Each new or changed asset takes three iconik API calls: one for the asset record, one for custom metadata and one for the proxy URL. Unchanged assets are skipped, assets your filters exclude are dropped before any per-asset call, and the sync slows down and resumes when iconik returns rate-limit responses.
Indexing is billed per minute of video for each feature. At the published rate of $0.05 a minute for scene search, 1,000 hours of footage costs $3,000 to index once. Faces, transcription and OCR are priced separately, and you only pay for the ones you turn on.
iconik announced natural language search as a beta feature in September 2026, in plain English and more than 70 other languages. Semantic search over the video itself comes through its TwelveLabs partnership, where the native connector was in early access as of November 2026. An external index such as Mixpeek can also sync from iconik and search the footage.
Yes. TwelveLabs named iconik its first Premier Tier ecosystem partner in September 2026. Customers use TwelveLabs search through custom actions and browser tooling, and a native connector inside iconik's search is in early access through iconik account managers.
iconik's built-in face recognition finds where a person appears across the library. With Mixpeek you can also search by a reference photo you supply and combine the face match with iconik metadata filters, such as a collection path or a custom field, in the same query.
iconik's documented search works from text and metadata. A Mixpeek retriever over your iconik library takes an image or a short clip as the query and returns the source asset and the timestamp. How to find the full video from a clip explains how that matching works.
No. The connector reads assets, metadata and proxies from iconik. Each search result carries the iconik asset ID, so your application can link to the asset or write fields back through the iconik API itself.
No. It downloads the proxy iconik already generates for each asset, which keeps transfers small and leaves your originals where they are.
The matching Mixpeek object and everything extracted from it are removed. With an iconik webhook configured this happens when the delete occurs; without one, the next sync reconciles it.
iconik integration: what the connector does, end to end
Best video search tools: hosted video search options compared
Best AI digital asset management tools: DAM and MAM platforms with AI search
Media asset management: what a MAM does and how it differs from a DAM
Semantic video search: how search over footage works
Cost and load on your iconik account
The sync downloads proxies, never originals. Each new or changed asset takes three iconik API calls: one for the asset record, one for custom metadata and one for the proxy URL. Unchanged assets are skipped, assets your filters exclude are dropped before any per-asset call, and the sync slows down and resumes when iconik returns rate-limit responses.
Indexing is billed per minute of video for each feature. At the published rate of $0.05 a minute for scene search, 1,000 hours of footage costs $3,000 to index once. Faces, transcription and OCR are priced separately, and you only pay for the ones you turn on.
Frequently Asked Questions
Does iconik have semantic or natural language search?
iconik announced natural language search as a beta feature in September 2026, in plain English and more than 70 other languages. Semantic search over the video itself comes through its TwelveLabs partnership, where the native connector was in early access as of November 2026. An external index such as Mixpeek can also sync from iconik and search the footage.
Does iconik use TwelveLabs?
Yes. TwelveLabs named iconik its first Premier Tier ecosystem partner in September 2026. Customers use TwelveLabs search through custom actions and browser tooling, and a native connector inside iconik's search is in early access through iconik account managers.
Can I search iconik by face?
iconik's built-in face recognition finds where a person appears across the library. With Mixpeek you can also search by a reference photo you supply and combine the face match with iconik metadata filters, such as a collection path or a custom field, in the same query.
Can I find a clip in iconik from a screenshot or a short clip?
iconik's documented search works from text and metadata. A Mixpeek retriever over your iconik library takes an image or a short clip as the query and returns the source asset and the timestamp. How to find the full video from a clip explains how that matching works.
Does Mixpeek write tags back into iconik?
No. The connector reads assets, metadata and proxies from iconik. Each search result carries the iconik asset ID, so your application can link to the asset or write fields back through the iconik API itself.
Does Mixpeek download my original files from iconik?
No. It downloads the proxy iconik already generates for each asset, which keeps transfers small and leaves your originals where they are.
What happens when an asset is deleted in iconik?
The matching Mixpeek object and everything extracted from it are removed. With an iconik webhook configured this happens when the delete occurs; without one, the next sync reconciles it.