Store AI assets, training files and briefs
Connect
SPIRITT
ReplicateReplicate
Models into products
SPIRITT turns Replicate models into working features and production workflows. Run predictions, compare versions, train custom models, and preserve the outputs your business needs.
What you can do with Replicate.
Useful AI features
Build a product around a model with validated inputs, visible prediction status, and saved results rather than a fragile one-off request.
Evidence before rollout
Compare model versions on your own examples and document where output quality changes before choosing a release.
Owned execution
Track predictions and training jobs, cancel unwanted runs, and move valuable outputs into durable storage.
Works with your other apps.
Get started in three steps.
01
Open a workspace
Create a SPIRITT workspace for the systems you want built and the recurring work you want owned.
02
Connect Replicate
Connect Replicate with an API key for the account that owns your models, files, predictions, and deployments.
03
Hand over a mission
Describe the feature or production batch, provide sample inputs and evaluation rules, and let SPIRITT build and run it.
Hand over a mission.
Choose a mission. Make it yours.
Put Replicate to work.
Put your models to work
Questions
Replicate, with SPIRITT.
01Can SPIRITT run both public models and deployments?+
Yes. Official models can run by owner and name, other predictions can target a version ID, and deployment predictions target your dedicated deployment. SPIRITT selects the right route and checks the relevant input schema instead of treating these as interchangeable.
02What access does Replicate need?+
Connect an API key for the intended account. Model ownership, visibility, and account permissions determine the files, training jobs, and deployments available to the mission; hardware and model choices remain part of its configuration.
03Can it train a custom model?+
SPIRITT can create a destination model and start a training job against a version that supports training, then list or cancel training jobs. Training inputs and parameters depend on that version, so it checks them before submitting your dataset.
04Can it build a customer-facing AI feature?+
Yes. A custom app can collect schema-valid inputs, create predictions, show their status, and save successful results. That app supplies the customer experience while Replicate performs model execution.
05How do results reach our other tools?+
SPIRITT can move completed assets into Google Drive, store job records in Supabase, and send Slack completion notices. Needed files should be persisted promptly because retrieved Replicate file URLs may be short-lived.
06What makes a model workflow reliable?+
Version-specific schemas and example inputs prevent malformed requests. Prediction retrieval, cancellation, saved artifacts, and verified webhook notifications provide the execution controls around the model rather than assuming every run finishes immediately.