veri

ROADMAP / CHANGELOG

What we are shipping for faster model iteration.

Follow the work moving Veri from training experiments to reliable production infrastructure: orchestration, observability, compute routing, deployment, and billing.

Start training
Now

Production-ready training loops

Tightening the core path from dataset upload to trained artifact.

  • End-to-end production training smoke tests
  • Live reward, KL, throughput, and GPU utilization traces
  • Dataset and evaluation create flows
  • Dashboard polling and settings backed by real API data
Next

Provider-aware compute

Making GPU selection cheaper, faster, and easier to reason about.

  • Lambda, RunPod, and SF Compute adapter validation
  • Provider auto-selection by GPU type and availability
  • Encrypted SSH key management for VM providers
  • Clearer job status across Modal and worker agents
Later

Deployment and billing maturity

Polishing the production surface teams need before scaling usage.

  • Stripe webhook verification and billing event history
  • Credential rotation workflows
  • Rollbackable model artifacts
  • OpenAI-compatible deployment endpoint hardening

CHANGELOG

Recent platform updates.

Latest

Production API hardening

ECS Fargate deployment, production config validation, CORS allowlists, webhook signature enforcement, and CI security scanning.

Recent

Multi-method training

GRPO and video generation SFT support, method-specific schemas, model filters, and training job API filtering.

Recent

Multi-provider compute

Lambda Cloud, RunPod, and SF Compute adapters now sit beside Modal with worker-agent status sync.

Recent

Model catalog and dashboard

95 curated models, Video Gen filters, LoRA pills, TanStack Query dashboard polling, and consolidated settings pages.