🚀 Trial v0.30.0 is live! The latest release is available for trial today — get started in 5 minutes →
GPU & AI infrastructure fabric for Kubernetes — autonomous placement, FinOps, Zeus AI ops engineer.
Public issue tracker and community feedback space for Forge by Zyvor AI Labs. The source code is maintained in a private repository. This repo is for bug reports, feature requests, UX feedback, and community discussion.
- Zeus AI approval gate GA — Zeus proposes remediation actions, operators approve with one click
- FinOps chargeback export — export per-team GPU cost reports to CSV, Slack, or S3
- RAG studio v2 — visual pipeline builder for retrieval-augmented generation workflows
- GPU scheduler preemption — high-priority training jobs can preempt lower-priority inference
- Multi-cluster federation v0.30 — unified placement and cost view across up to 20 clusters
| Problem | Forge answer |
|---|---|
| GPU clusters are opaque — no visibility without DCGM expertise | Live inventory, DCGM telemetry, MIG lifecycle, scheduling pools — out of the box |
| kubectl is too slow for GPU fleet ops | K8s console with SSE logs, WebSocket exec, drain progress, fleet health in one UI |
| AI infra spans too many disconnected tools | Model catalog, RAG studio, inference gateway, training factory — one control plane |
| Incidents need context across 6 tabs | Zeus AI embedded everywhere — grounded answers, approval-gated remediation |
| GPU cost and security are afterthoughts | FinOps chargebacks, budget guardian, SIEM export, threat hunting built in |
┌─────────────────────────────────────────────────────────────────────┐
│ Clients Web UI · Rust CLI · Python/Go SDK │
├─────────────────────────────────────────────────────────────────────┤
│ AI Layer Zeus · Model Router · Embedding · RAG Ingest │
├─────────────────────────────────────────────────────────────────────┤
│ Intelligence Placement · FinOps · RCA · Orchestrator · Topology │
├─────────────────────────────────────────────────────────────────────┤
│ Control Plane 7 Operators · 60+ CRDs · GPU scheduler · eBPF │
└─────────────────────────────────────────────────────────────────────┘
# Standalone install (any Kubernetes cluster — EKS, GKE, AKS, k3s, kubeadm)
curl -fsSL https://releases.hypersdk.dev/forge/install.sh | FORGE_VERSION=0.30.0 bash
# Helm
helm repo add hypersdk https://charts.hypersdk.dev
helm upgrade --install forge hypersdk/forge-core --version 0.30.0 \
--namespace forge-system --create-namespaceRequirements: Kubernetes 1.28+, NVIDIA GPU nodes with DCGM (for GPU features), 8 GB RAM for control plane
Include: what you did, what happened, what you expected, your environment, screenshots or logs (redact secrets).
Use GitHub Discussions for open-ended questions, ideas, and roadmap conversations.
Do not report security vulnerabilities publicly. Email security@zyvor.dev — see SECURITY.md.
- Source code, internal configs, or architecture details
- API tokens, license keys, or credentials
- Private logs with sensitive data
- Security vulnerabilities — email security@zyvor.dev instead
⭐ Star this repo · 💬 Open a Discussion · 🐛 Report a Bug · 📧 hello@zyvor.dev
Maintained by Zyvor AI Labs