Seiri · AI cataloguing
Every hour of your archive, described. Nobody has to watch it.
Seiri watches and listens to every video, audio file and image, and returns what an archivist would take hours to log: who is speaking, what they say, what is on screen and what it is about. Timecoded and ready to search.
- 01
From viewing to searching
Stop playing tapes to find out what is on them. Type what you need and jump to the exact second.
- 02
An archive that gets reused
What is described gets found, and what gets found is broadcast again, licensed, or feeds a new story.
- 03
Inside your MAM
Metadata reaches your asset manager over the API. Your team keeps working where it already works.
Reference project · RTVE
The RTVE archive, described at scale.
Spain’s public broadcaster awarded Crosspoint the automatic metadata service for its archive, with Seiri as the technology core: transcription, entities, OCR, face and scene recognition, and translation of Spain’s co-official languages into Spanish.

28,000h/year
of archive analysed automatically by the service.
198,220h
from regional centres that RTVE has set out to catalogue.
Everything it does
Everything in the picture and in the sound.
Three layers that build on each other, the same three you choose in your plan. Every piece of data comes with its timecode.
Voice
What is said.
Transcription
Speech-to-text with punctuation and capitalisation, ready to read and to search.
00:12:47 · “It was a bold, provocative move…”Speakers
Diarisation and recognition of who is speaking at every moment.
SPK #6 · 00:12:47 → 00:13:23Language and translation
Identifies the language and translates the transcript.
EN → ES · CA · FRNamed entities
People, places and organisations, with how often they appear.
Kamala Harris ×6 · California ×5Summary
What the asset is about, in a paragraph that follows your editorial criteria.
Summary · 5 linesClassification
Content type and topic according to your own taxonomy.
NEWS · PoliticsSemantic segmentation
Splits a programme by story or subject, not by minutes.
12 segments · 19:52Vision
What is seen.
Keyframes
A storyboard to scan an hour at a glance.
00:00:03 · 00:00:18 · 00:01:02Faces
Public figures and the people you train from your own portal.
David Muir · on screenOn-screen text (OCR)
Captions, credits and headlines, exactly as they read.
“BREAKING NEWS”Logos and objects
Brands and relevant objects that appear in shot.
person ×46 · airplane ×10Scenes and image
Scene type and a description of what is seen, useful for accessibility.
Studio · “anchor in front of a screen”Your own training
Teach Seiri your own presenters’ faces and voices, without depending on anyone.
Training PortalMind
What you can ask of it.
Chat with your content
Ask in plain language and get the answer with its exact moment.
Talk to SeiriAI agents
Hand over whole tasks: a selection, a sequence, a correction.
Agentic modeAgent mode
Hand it the whole task.
Some jobs are not one search: they are twenty. Seiri’s agent mode plans the task, shows you every step it takes and stops to ask for confirmation before changing anything. If it is unsure, it asks.
- Searches, compares and selects across your whole archive
- Prepares sequences and cut lists, multicam included
- Corrects transcripts and speaker names
- Every action is in plain sight; you approve or reject
Workflows, API, plugins and webhooks
If it repeats, automate it.
You write in plain language what should happen and the platform turns it into a workflow with a trigger, conditions and actions: a complete process, with no programming. You review and refine it in a visual editor.
- Workflows: triggered when a video is processed or edited, with AI-evaluated conditions
- REST API: your applications ask the platform for what they need, when they need it
- Plugins: run an action when a system event occurs
- Webhooks: the platform notifies your systems as soon as a process finishes
How it works
From ingest to catalogue, without touching your workflow.
01Ingest
Upload files or connect your storage: local, Samba, FTP, S3 or LTO tape.
02Analysis
Seiri processes each asset with the models in your package, in Amplify’s cloud or on your machines.
03Review
Your team corrects, validates and extends the fields. The AI proposes; your archivists decide.
04Delivery
Metadata to your MAM over the REST API, or exported as CSV, Excel, JSON and text.
Demo
Seiri in action.
Pricing
From 349 a month. Two choices: hours and layer.
You pay for the hours you analyse and the layer you switch on. The entry tier exists so a local station — or an organisation that is not a media company — can afford it too.
Subscription (OpEx)
from349USD/EUR per month
SaaS model. Start today and scale as volume grows.
Perpetual licence (CapEx)
from9,990USD/EUR
A licence you own, to install on your infrastructure.
1. Choose your workload
| Tier | Monthly usage per node | Recommended for |
|---|---|---|
| Entry-Level | 150 hours | Local TV stations and organisations in other sectors |
| Mid-Market | 300 hours | Mid-sized broadcasters and production companies |
| Enterprise | Unlimited | Large broadcasters and operators, OTTs, newsrooms |
2. Choose your layer
| Functionality | Voice | Vision | Mind |
|---|---|---|---|
| Transcription with diarisation | Included | Included | Included |
| Speaker recognition | Included | Included | Included |
| Language identification | Included | Included | Included |
| Translation | Included | Included | Included |
| Content summary and classification | Included | Included | Included |
| Semantic segmentation | Included | Included | Included |
| Named entity recognition | Included | Included | Included |
| Smart keyframes | Not included | Included | Included |
| Face recognition | Not included | Included | Included |
| Image description | Not included | Included | Included |
| Text (OCR), logo and object recognition | Not included | Included | Included |
| Scene type recognition | Not included | Included | Included |
| Built-in chat grounded in your content | Not included | Not included | Included |
| Tasks handled by AI agents | Not included | Not included | Included |
“From” prices, excluding taxes. We prepare a fixed proposal based on hours, package and deployment.
Which plan fits you?
Your starting point
Entry-Level · Voice
Up to 150 hours a month per node. Designed for local TV stations and organisations in other sectors.
Get your proposal- Input
- High-resolution originals or proxies from your MAM or PAM; horizontal 16:9 and vertical 9:16
- Languages
- Transcription in 90+ languages
- Deployment
- Amplify’s cloud, hybrid, on-premise or air-gapped; Docker microservices
- Integration
- REST API and Developer Portal, LLM endpoint, MCP and webhooks; event-driven plugins; 15+ systems across MAM, editing, storage and CMS
- Storage
- Local, Samba, FTP, S3, HTTP and LTO; online and nearline
- Access
- Multi-organisation, LDAP / Active Directory, two-factor login, API tokens; audit log
- Export
- JSON, XML, CSV, Excel, SRT, VTT, text transcript and EDL
- Licence
- Per NVIDIA GPU: perpetual with annual support, or subscription
- Minimum on-premise hardware
- NVIDIA GPU with 24 GB of VRAM (an L4, for example), 64 GB of RAM, an 8-core CPU and Ubuntu 24
- Certified
- NVIDIA DGX Spark and GB10 systems
- Interface
- 8 languages: English, Spanish, Catalan, French, German, Portuguese, Italian and Russian
FAQ
Frequently asked questions
Does my content leave my infrastructure?
Only if you want it to. Seiri installs on your servers with local models, or runs as a service in Amplify’s cloud. Hybrid deployments are available too.
Does it work with my MAM?
Seiri integrates with third-party management systems over its REST API. It already runs in production alongside TMD’s Mediaflex, and we work with Knox Media Hub and ELEMENTS.
Can I teach it to recognise my own people?
Yes. The training portal lets you add your own faces and voices — presenters, contributors, local politicians — without depending on us.
How much does it cost?
We publish entry prices: subscription from 349 a month or a perpetual licence from 9,990. The rest depends on hours and package.
What if the AI gets it wrong?
Every piece of data carries its timecode: it is checked in one click and corrected before it reaches the MAM.
Does it replace the archive team?
No. It removes the viewing and the typing; the judgement stays with your team, who validate, correct and train the models on their taxonomy.
Can I take my metadata with me?
Yes: it leaves whenever you want as CSV, Excel, JSON or over the API.
Next step
Show us your content. We’ll show you what it’s worth.
A 30-minute demo on your own footage, with someone on our team who has worked in television.