Skip to content

developer_plane_plan.md

Latest commit

 

History

History
36 lines (25 loc) · 2.07 KB

developer_plane_plan.md

File metadata and controls

36 lines (25 loc) · 2.07 KB

Developer Access Plane: The "Ollama for Enterprise Compute"

Vision

To create a seamless, "local-feeling" experience where open-source contributors have instant access to enterprise-grade compute (e.g., H100 clusters). It should feel as simple as running Ollama, but instead of running on your laptop, it transparently offloads to powerful infrastructure subsidized by providers.

Core Philosophy

  • Merit-Based Access: Compute is not bought; it is earned. "Unlimited" enterprise-grade resources are available to those whose open-source contributions qualify them.
  • "Ollama-like" Simplicity: No complex cloud consoles or provisioning tickets. You just run a command, and the compute is there.
  • Invisible Infrastructure: The developer doesn't "request" a server; they just run their workload. The system handles the federation of resources from providers (NVIDIA, etc.) in the background.

Architecture

1. The Access Plane (The "Brain")

  • Merit Engine: Verifies contributor status and authorizes access based on public impact.
  • Transparent Broker: Routes workloads to the best available provider without user intervention.

2. Provider Interface (The "Muscle")

  • Federated Resource Pool: Providers (NVIDIA, Academic Clusters, Clouds) plug in their idle or subsidized capacity.
  • Standardized Runtime: Ensures workloads run consistently regardless of the backing hardware.

3. Developer Interface (The "Magic")

  • Unified CLI: A single tool (e.g., cc run) that mimics local execution.
  • Seamless Tunneling: Ports and logs are forwarded instantly, making remote H100s feel like localhost.

Roadmap

Phase 1: The "Merit" Proof

  • Implement the logic to validate open-source contributions (e.g., GitHub/GitLab activity).

Phase 2: The "Ollama" Experience

  • Build the CLI that abstracts the remote execution.
  • cc run llama3 should feel identical to ollama run llama3, but running on a DGX Cloud instance.

Phase 3: Provider Federation

  • Connect the first major provider (e.g., NVIDIA) to back the experience.