Turn your systems into training data
Connect a source once and the training file keeps itself current. SaaS platforms, databases, APIs and documents become training-ready datasets, and a fine-tuning job reads the file by ID.
Bring the systems you already run
Databases, APIs and document stores connect from the catalogue, and custom integration support covers the rest.
One file per source, kept current
Each connected source writes a JSONL training file, and the sync refreshes it on a schedule.
Validated before you spend GPU time
Every file is checked line by line, and a rejected line comes back with its number and the reason.
Every step is an API call
Upload data, fine-tune, serve and export the weights through one documented API.
Every source, filterable by kind
SaaS platforms, databases, APIs and documents sit in one catalogue. Filter by kind or by what the system holds, pick a source and connect it in place. Custom integration support covers the systems you already run.
Results from LLMs powered by Nucleus
monthly revenue misattributed
Ad performance tracking
systems on one thread
Real-time field data collection
sweep of the tracking feed
Shipment delay management
downtime prevented this month
Predictive maintenance
Workflows that read several systems at once
$73.4K a month was being credited to the wrong channel.
Google Ads, Facebook Ads, Matomo, Stripe, Shopify
Cable survey readings processed while the crew was still on site.
AutoCAD, Google Maps, Google Drive, Asana, Slack, HubSpot, MQTT
Every shipment two days past ETA, found before the customer called.
Salesforce, MySQL, Slack, Gmail
What the connectors give you
Sources
- SaaS platforms
- Databases
- REST, GraphQL and webhook APIs
- Document stores
- CSV files
- Custom integration support
The file
- JSONL, one object per line
- Chat format with system, user and assistant roles
- Purposes: fine-tune, evaluation and retrieval
- Attached to a project
Validation and refresh
- Every line checked
- Errors carry the line number and the reason
- A new version reaches ready before a job can use it
- Listed newest first
Control
- Keys bound to another project cannot read a file
- Delete honours data-erasure requests
- An Idempotency-Key guards a retried upload
- The usage API reports storage by day and project
Storage
Datasets, checkpoints and exported weights are metered per GB-hour, so you only pay while you keep them.
Training tokens are metered separately when a job runs.
More in Train
Fine-tuning
Managed training on 25 open models
Training loop
DocsRL, DPO and distillation as API calls
Models
Open bases across six modalities
Read on
Connect your first source
Pick a source from the catalogue, or tell us which systems you run.