CommonCompute
Log inGet started
Log in
Get startedDownload the Mac app
Roadmap

What we're building, what's next, what we're not doing.

We try to keep the roadmap honest — including the "not doing" column, because saying no publicly is the trust signal we'd want from a tool we're evaluating. Source of truth is ROADMAP.md on GitHub. Last updated 2026-05-13.

Shipping now

Now

Currently shipping or queued for the next release.

  • Notarized installs end-to-end. Sign with Developer ID, notarize, staple, publish via Sparkle. Removes the Gatekeeper warning on first launch. Tracking via v1.6.1.
  • In-app feedback with auto-attached context. Replace mailto:support@ with a sheet that bundles app version, chip, telemetry, and recent logs so the first reply isn't "what version are you on?". Shipped to main, releases with v1.7.0.
  • Accessibility floor. VoiceOver labels on the dashboard, sidebar, menu-bar icon, and login fields. Releases with v1.7.0.
  • Public changelog page at commoncompute.ai/changelog so every shipped thing is searchable. Live now.
Up next

Next

Started or planned for the following release.

  • Per-version Sparkle adoption dashboard. Admin-only view of { version: count } from the heartbeat stream so we can spot a bad release before it phases out.
  • Reserved capacity tier. Pre-buy provider hours for predictable throughput on schedule-bound workloads. Pricing tier between batch and on-demand.
  • One-click "report a bad task" inside the Jobs view. Job-level feedback that pre-attaches the receipt, the inputs, and the runner log — closing the loop on tasks that ran but produced bad output.
  • Real-time mode for the request API — sub-second dispatch with a small premium, for interactive workloads that don't tolerate batch latency.
Thinking about

Later

Things we want to do but aren't sized yet. Order is roughest here.

  • First-party SDKs for Python and TypeScript with the same surface as the LangChain / LlamaIndex / Vercel AI integrations.
  • Multi-Mac fleet view for organisations. When one user runs Common Compute on 30 Macs across the office, give them one screen.
  • Linux providers. ARM and x86, gated behind capability profiles so the workload catalog stays honest about where things actually run well.
  • GPU-attached provider pools. Discrete Apple GPUs (eGPU) and external NVIDIA cards on supported hosts. Same envelope shape.
  • Per-task carbon receipts. Heartbeat already carries chip + thermal + duration — render the kWh and gCO₂eq alongside the price.
Not on the list

What we're not doing

These come up often enough to be worth saying out loud.

  • A general-purpose serverless GPU API. We don't have the hardware to compete with Modal / Together / Fireworks on chat. We're a catalog of specific workloads that Apple Silicon serves well at honest cost. We'd rather refuse a workload than overpromise on it.
  • A consumer compute marketplace. Providers contribute Macs they own; we don't intermediate hardware purchases or financing.
  • Closed-source clients. The Mac app, the SDKs, and the router-side verification code are open and stay that way.