The multimodal warehouse for AI

    Turn any data
    into answers.

    Your AI agents need to understand more than text. Mixpeek turns video, images, audio, and documents into searchable intelligence, with one pipeline instead of many.

    Try it on your files Explore templates

    Index once. Query every way.

    1. 01 Connect
      Your files
    2. 02 Understand
      Every modality
    3. 03 Ask
      Anything
    10 videos indexed
    00:12.4
    Library
    10 videos · 977 moments
    Featured case study

    A performance-video agency maps the creative DNA of its entire ad library: search any moment, cluster what works, and reuse the winners. See how

    Use Mixpeek from your agent
    Files Understand Structure Use

    Every file becomes data you can query.

    Mixpeek finds what's in each file and stores it as fields your software and agents can search.

    Person walking past a storefront, with face, body, and handbag detectedPerson · 0.97Face · 0.95Handbag · 0.92
    Features extracted
    objects
    person 0.97 · handbag 0.92
    faces
    1 face · 0.95
    scene_caption
    A woman carrying a tan tote bag walks past a red storefront.
    transcript
    “…meet me at the corner in five.”
    a woman walking past a storefront carrying a bag
    24 results · 38msillustrative
    • street_cam_04.mp4 · 00:04:120.94

      A woman carrying a tan tote passes a red storefront.

      captionhandbagface
    • promo_reel_12.mp4 · 00:01:370.89

      Pedestrian with a tan tote bag passing a storefront.

      handbagcaption
    • bodycam_2207.mp4 · 00:00:480.85

      Single person walking, face visible in profile.

      faceperson
    Try it with a ready-made template
    See every extractor
    Start from a template

    Pick a job. Deploy it in one click.

    Each template is a complete pipeline for one job: sources, collections and search in one manifest. Deploy it to Studio, then point it at your own bucket.

    See all templates
    A camera operator silhouetted against a lit setScene · 0.91Camera · 0.89Video · footage.webp

    Find the take in an archive nobody has watched end to end.

    1.27M
    scenes in the reference corpus
    5 → 3
    sources into clean collections
    Deploy this templateHow it works Opens in Studio
    Discovered structure

    "What concepts exist in my data that nobody has labeled yet?"

    Mixpeek groups your files with no labels and builds the taxonomy for you. Every search and correction sharpens it.

    Explore clustering
    Your files, unlabeled
    Hands unboxing a small item from a package with crinkle paperHands opening a white package to reveal the contentsOpening a cardboard box with tissue-wrapped contents
    Cross-modal joins

    One query across every modality.

    Mixpeek lines up what's seen, said and shown at the same moment, so an agent gets the exact clip. Switch a feature off to see what each one adds.

    How joins work
    “Find the moment our CEO said guidance while the slide read Q4 outlook”
    ⨝ joined
    earnings_call.mp4 · 00:12:041 exact match

    The CEO says “guidance” as the slide behind her reads “Q4 outlook.”

    Quickstart

    One install. Two paths.

    Connect your files and Mixpeek builds the index. Already have embeddings? Bring them instead.

    Install
    pip install mixpeek
    Managed quickstart

    Connect your files

    Start with your files Read Managed docs
    # pip install mixpeek
from mixpeek import Mixpeek

mp = Mixpeek(api_key="YOUR_KEY")

collection = mp.collections.create(
    name="media_objects",
    source="s3://media/**/*",
    feature_extractors=[
        {"type": "visual_embedding"},
        {"type": "transcription"},
        {"type": "ocr"}
    ]
)

retriever = mp.retrievers.create(
    name="media_search",
    collections=[collection.id],
    stages=[{"type": "hybrid_search"}, {"type": "rerank"}]
)

results = retriever.search(
    "Find clips where customers discuss pricing"
)
    01Your bucket
    02Managed extraction
    03Agent retrieval
    For coding agents

    Using Mixpeek from an agent.

    Mixpeek reads your files where they already live, in S3, GCS or any bucket, and turns what's in them into data your software can search, classify and moderate. This is what an agent needs to call it.

    Install

    curl -fsSL https://mixpeek.com/install | sh

    Adds the Mixpeek MCP server to Claude Code, Codex, Gemini CLI and Cursor.

    claude mcp add --scope user --transport http mixpeek https://mcp.mixpeek.com/mcp --header "Authorization: Bearer YOUR_API_KEY"

    The same MCP server, added to Claude Code by hand.

    pip install mixpeek

    Python SDK.

    POST https://api.mixpeek.com/v1/signup/agent

    Gets an API key without a browser, for an agent acting for a person. Read the agent signup guide.

    Good to know

    • Send Authorization: Bearer <api_key> on every call, and X-Namespace: <namespace> on every call inside a namespace.
    • The parameters each retriever stage accepts are listed at GET /v1/retrievers/stages.
    • POST /v1/retrievers/{retriever_id}/execute refuses collection_identifiers, because a stored retriever already knows its collections. Only the ad-hoc POST /v1/retrievers/execute takes them.
    • A published retriever's stages can't be edited. Clone it, edit the clone, then point the listing at the clone with PATCH /v1/marketplace/listings/{listing_id}.
    • Document lists return a total only when you send include_total=true, and offset pagination stops at 10,000. Use next_cursor past that.

    Use it for

    • Searching video, images, audio and documents by what is in them, while the files stay in your S3, GCS or other bucket.
    • Classifying, tagging and moderating a library of files with one pipeline.
    • Giving an agent tools to search a file library, through MCP or the API.
    • Searching embeddings you already have, stored on object storage (MVS).

    Not for

    • Generating or editing images, video or audio.
    • A few hundred files you can process once with a script.
    • Rows of structured data with no files behind them. Use a regular database.
    Measured

    What we measured, and how.

    Numbers from runs on production, with the method under each one. The raw results are published here.

    Accuracy

    Finds the exact item.

    50 of 50 ranked first

    We picked 50 artworks at random from the 118,000 in the National Gallery of Art listing and searched with each one's image. Every one came back as the top result.

    Top-1 matches50 / 50

    Image-to-image search with SigLIP on production, November 3, 2026. Each query used a stored artwork as the seed (input_mode document), top 10 results, against all 118,000 artworks. 50 of 50 ranked their own artwork first. A different collection or model will give different numbers.

    Speed

    The index answers in milliseconds.

    34 ms median search

    The vector search step over 118,000 images took 34 ms at the median and 284 ms at p95. Turning the query image into an embedding took longer, 830 ms at the median.

    Same 50 queries, read from each response's stage timing breakdown (search_ms and embedding_ms). The full request took 925 ms at the median and 3.7 s at p95, most of it embedding.

    Agents

    One call to a working key.

    1 call to an API key

    POST /v1/signup/agent returns a key, a namespace with a bucket, a collection and a retriever, and the calls that take a file to a search result.

    From the endpoint's contract. The key starts with 1,000 credits and expires after 72 hours unless the person whose email the agent sent confirms it. Signups are limited to 5 an hour per IP address.

    Founded by engineers from
    MongoDBBerkeleyNVIDIAEtsyAmazon Web ServicesEquinixIAB Tech Lab
    Enterprise-ready
    SOC 2-ready Encrypted at rest BYO-Cloud Self-hosted option Audit trails SSO / SAML
    Pricing

    From $25/mo. Usage-based everything.

    Two products, one model: a flat monthly plan fee, with usage billed on top at the same published rates. MVS is priced by the vector, Managed by the unit of content.

    MVS
    $25/mo plus usage

    Bring your own embeddings and pay by the vector. Dense, sparse, and BM25 search on your own object storage. Build starts at $25/mo with up to 1M vectors; Scale ($250/mo) covers 25M.

    Start with MVS
    Managed
    $25/mo plus usage

    Bring raw files and pay per unit of content, such as video by the minute and images by the thousand. Build includes $10 of processing a month and covers 100K objects; Scale ($250/mo) includes $100 and covers 1M.

    Start with Managed
    Enterprise
    Custom

    Dedicated infrastructure, self-hosted options, SSO, SLA, security reviews, and hands-on architecture support.

    Talk to us
    See full pricing

    Signing up from a coding agent? Read the agent signup guide.

    FAQ

    Common questions.

    What does Mixpeek do?

    Mixpeek understands what's inside your files and turns that understanding into data your software can use. It reads video, images, audio and documents where they already live, in S3, GCS or any S3-compatible bucket, and extracts faces, scenes, objects, speech, on-screen text and document layout. Your application then searches, classifies, filters, ranks and moderates that data through one API.

    Do I have to move my data?

    No. Mixpeek reads from your existing S3, GCS, R2, Azure, or S3-compatible bucket. Your storage stays the system of record, and nothing leaves your cloud.

    How fast is retrieval?

    The search step is fast; turning the query into an embedding takes most of the time. On November 3, 2026, image-to-image search over 118,000 artworks took 34 ms at the median and 284 ms at p95 for the search step, with vectors on object storage. Embedding the query image took 830 ms at the median, and the full request 925 ms.

    Do I need embeddings to start?

    No. Bring your own vectors with MVS, or point Managed at raw files and it generates embeddings and features for you.

    What can Managed extract?

    Faces, scenes, transcripts, OCR, labels, and embeddings from video, images, audio, PDFs, and documents, all indexed at the object level.

    Can I self-host?

    Yes. Deploy in your own cloud (BYO-Cloud) with encryption, role-based access, SSO, audit trails, and namespaces. We hold no SOC 2 or HIPAA certification today; see mixpeek.com/trust.

    How does pricing work?

    Both MVS and Managed start at a $25 monthly plan fee, with usage billed on top. MVS bills vectors stored, storage, writes and queries. Managed includes $10 of processing a month on Build, then bills extraction, embedding, indexing and retrieval per unit of content.