Roadmap

What's shipping next, by quarter. This is a forecast, not a contract — dates slip; priorities reshuffle. The substance of the bets stays stable.

Shipped (the launch window)

• OpenAI-compatible Inference, S3-compatible Object Store, Postgres-wire JouleDB, Compute, Functions, Meet
• $5-hold-then-joules billing, with Stripe, USDC, and reward-point top-ups
• FROST 3-of-5 ed25519 signed receipts on every API call
• Docs at docs.greenjoules.cloud — 55+ pages including 9 SDK quickstarts, 11 migration guides, 3 tutorials, 3 deep dives
• OpenAPI 3.1 spec + interactive Swagger UI at /api-explorer
• MCP server for Claude Desktop / Cursor / Zed
• Webhooks (HMAC-signed) for account events
• 16-node backbone fleet across Hetzner / Vultr / Latitude

Q3 2026 — reliability + comfort

• SOC 2 Type 2 audit field-work complete (attestation Q4)
• Multi-region region = "auto" default for Compute — SLA jumps from 99.5% to 99.9% with no config change
• Provisioned concurrency UI for Functions — visual scale.min tuning
• Per-account tier calibration — the router learns your traffic's thresholds over 2 weeks
• JouleDB read replicas across regions, replication scheduled on grid-cheap-hour where latency permits
• Speculative decoding at the gateway for L3 reasoning — ~2× cost cut on long generations
• Audit log surface in the portal with diff-style change tracking on workloads
• Per-customer-tag billing reports — one-click split of monthly bill across your X-Customer-Tag values

Q4 2026 — the certifications quarter

• SOC 2 Type 2 attestation issued
• HIPAA BAA stable, US region available for protected health workloads
• EU Cloud Code of Conduct adherence declared, with auditor-signed evidence pack
• FedRAMP Moderate baseline mapping published (full attestation 2027 via Oracle partnership)
• Encrypted object-store SSE-KMS with HSM-backed customer key option
• Joule-budget-as-code — energy_budget as a first-class spec on every resource type (workload, function, db, bucket)
• Real cert-pinning fingerprints published at /.well-known/cert-pins.txt

Q1 2027 — broader compute

• Raw GPU access for non-inference workloads — H100 / H200 / MI300 by the hour, joule-priced
• Spot tier for batch compute — preemptible at 30% joule discount, snapshot-restore semantics
• Edge runtime expansion to 12 cities (currently 9 Fly.io PoPs)
• JouleDB Postgres 17 wire-level features
• Object Store SelectObjectContent + lifecycle policy transitions
• Cron job scheduler primitive, separate from Functions (multi-trigger orchestration)

Q2 2027 — the platform leans in

• FedRAMP Moderate attestation via Oracle partnership
• ISO 27001 certified
• Customer-side substrate — install jc-agent on your own bare metal, register into the mesh, your hardware shows up as cheapest-capable silicon for your own workloads
• Joule language v1 targets-stable release
• Disaggregated Inference — prefill / decode split across silicon types

What we have decided NOT to build

To keep the substrate honest, a short list of common requests we've declined:

• A managed Kubernetes service. The mesh IS the cluster. Bring your container, declare it. We don't want to be in the cluster-management business.
• Per-second VM billing. The whole point is per-joule.
• A Vercel-style preview-deploy-per-PR magic loop. We provide deploy primitives; CI tools do the magic. We don't want to own your CI.
• A "free tier". The $5 hold is the floor; it covers a meaningful amount of real work. Free tiers attract abuse and degrade per-customer reliability.
• Reserved instances / committed-spend discounts. No reservations means no monetisation of variance. The price is the price.

How requests get on the roadmap

The fastest path: be a paying customer with the bill open, or be a contributor sending PRs against the source-available repos. Both routes go in the founder's inbox; both have shaped the roadmap above. Write to [email protected] with the workload shape; if it fits, we'll work backwards from your launch date.