AI/ML

Replicate Product Plan

ML model hosting platform. Run open-source models via API with auto-scaling and pay-per-use pricing.

Use this Product Plan

Open the editable plan instantly, or carry its requirements into a generated system Architecture.

Run models via API

Auto-scaling GPU instances

Custom model deployment (Cog)

Streaming predictions

Webhook callbacks

File inputs/outputs

Model versioning

Community models

Fine-tuning support

Pay-per-second billing

As a developer, I can run Stable Diffusion via API in seconds

As a team, I can deploy custom models with Cog

As a developer, I can chain models in prediction pipelines

As a developer, I can get webhooks when predictions complete

As a startup, I can scale without managing GPUs

PythonGoCog (Docker)NVIDIA GPUsKubernetesPostgreSQLRedisS3gRPC

Category	Requirement	Target
Performance	Cold start	< 10s for first prediction (GPU warm-up)
Scalability	GPU scaling	Auto-scale from 0 to 100+ GPUs per model
Availability	API	99.9% for predictions
Security	Model isolation	Container-level isolation per prediction

MVPRun Models

API predictionsCommunity modelsFile I/OWebhooks

v1.0Custom Models

Cog packagingAuto-scalingStreamingFine-tuningVersioning

v2.0Platform

Model marketplaceTraining APIDeployments (dedicated GPU)Batch predictions

Models available: > 50K community models

Daily predictions: > 10M

Cold start time: < 10s median

Developer activation: first prediction < 5 minutes

Real-time ride-sharing platform with driver matching, GPS tracking, surge pricing, and payments.

Video streaming platform with personalized recommendations, adaptive bitrate streaming, and content management.

Team messaging platform with channels, direct messages, file sharing, integrations, and real-time search.

Open the editable Product Plan first, then carry it into Architecture and Plan & Ship when it is ready.