Skip to content

roadmap.md

Latest commit

 

History

History
114 lines (84 loc) · 4.66 KB

roadmap.md

File metadata and controls

114 lines (84 loc) · 4.66 KB
Raw

MatrixHub roadmap

This page describes the public, high-level direction for MatrixHub. It is maintained in the repo alongside other contributor docs; timelines shift with releases and community feedback.

For day-to-day work, see GitHub issues, releases, and the documentation site at matrixhub.ai.

Current focus

  • Reliability and performance for large-model distribution and Hugging Face–compatible access
  • Kubernetes and Helm deployment ergonomics
  • Operational completeness for the four enterprise workflows listed below

Key workflows we target

  • Intranet inference acceleration — pull-once, serve-all caching for large model fan-out on GPU clusters
  • Air-gapped model transfer — controlled export and import of approved models into isolated networks
  • Enterprise model artifact governance — tag locking, promotion, audit, and CI/CD-friendly access
  • Cross-region distribution — policy-driven, resumable replication between data centers

Scope

V0 must-have

  • Hugging Face–compatible API subset required by vLLM, SGLang, and common HF clients
  • Model repository creation, upload, download, delete, and visibility controls
  • Large-file storage on local filesystem, NFS, and S3-compatible backends
  • Proxy cache mode for public Hugging Face sources
  • Basic Web UI for repository browsing and administration
  • Token-based access control and project / namespace isolation
  • Basic audit logging
  • Export and import flow for air-gapped environments
  • Replication foundation with chunked transfer, resume, and retry
  • Deployment paths for Docker Compose and Kubernetes Helm

V1 should-have

  • Dataset repository support
  • Role-based access control with more granular permissions
  • Storage quotas and cleanup policies
  • LDAP, OIDC, or SSO integration
  • Access statistics and usage trends
  • Security scanning for malicious model content
  • Model signing and signature verification
  • Release management concepts such as tag locking and promotion workflow
  • CDN-friendly download acceleration

Later or exploratory

  • OCI artifact packaging for models
  • P2P distribution for startup storms
  • Direct-to-GPU loading patterns (NetLoader-style)
  • Kubernetes-native acceleration components for vLLM and SGLang
  • Automatic upstream mirror selection based on geography or latency
  • Deeper integration with inference-serving ecosystems
  • ModelScope compatibility where strategically useful

Milestones

M0 — Private hub baseline

  • Basic repository CRUD
  • Local and S3-compatible storage
  • API token authentication
  • Minimal Web UI
  • Hugging Face–compatible read path for core clients

M1 — Enterprise distribution baseline

  • Proxy cache mode
  • Project isolation
  • Audit logging
  • Air-gapped export and import
  • Initial replication worker with chunked transfer and resume

M2 — Production governance

  • Stronger RBAC
  • Tag locking or release promotion
  • Quotas and cleanup
  • LDAP or OIDC integration
  • Security scanning and signing foundation

M3 — Inference-native acceleration

  • Distribution optimizations for startup storms
  • Deeper vLLM and SGLang integration
  • Evaluation of P2P, OCI packaging, and direct-load acceleration

Non-goals

To stay focused on the workflows above, MatrixHub explicitly does not aim to be:

  • A general-purpose MLOps platform
  • A training orchestration or experiment tracking system
  • A public community model-sharing platform
  • Fully feature-equivalent to Hugging Face on day one
  • A platform pursuing broad multi-platform compatibility that does not materially help vLLM, SGLang, or core enterprise workflows

Success criteria

MatrixHub is succeeding when:

  • An enterprise team can switch inference clients to MatrixHub with minimal client-side changes
  • A large internal cluster can fan out one large model without repeated public downloads
  • An air-gapped organization can move approved models through a controlled import and export process
  • A production team can treat models as governed release artifacts rather than loose files
  • Cross-region replication is good enough to become part of normal operations instead of an exceptional manual task
  • The open-source project is seen as a complete self-hosted option for this category

How we plan

  • This page describes direction, not commitments to specific dates.
  • Concrete work is tracked in GitHub issues and shipped via releases.
  • Feedback and proposals are welcome via issues or GitHub Discussions.