Status: accepted working plan (discussed with the owner; refines with each era). Scope: from the current state to the full purpose of the platform. One roadmap per repository — this file. Companion documents: QUALITY.md (quality charter), docs/rfc/ (design records). Previous internal plans are superseded by this roadmap.
A lifelong cloud: a provider of one's own, free from vendor lock-in and jurisdiction dependence, developed and operated by a single engineer with a team of AI agents; full self-service for users; third parties extend the platform through the Open Cloud Standard (OCS) without modifying the core. Strategic horizon: a decentralized, unkillable network of clouds. Personally: number-one engineering quality and the right to stop.
The purpose is achieved when all of the following hold:
| # | Dimension | Verifiable criterion |
|---|---|---|
| P1 | Freedom | an installation ports across substrates (second adapter passes the same acceptance) and providers in different jurisdictions; user export leaves no residue |
| P2 | One engineer + agents | ≤30 min/day × 14 days of upkeep; install ≤2 h of attention; a second person repeats operations from public docs |
| P3 | Self-service | the full user and administrator journey (login, programmatic control, zero-residue lifecycle) proven on every release |
| P4 | Unkillability | ≥3 independent owners/jurisdictions at the network level; the federation runs with no central point |
| P5 | OCS is alive | ≥2 independent implementations of the contract (CNCF Sandbox will expect 3+); a third-party product runs without core changes |
| P6 | Quality #1 | comparable to or better than the leaders on every dimension; stage invariants enforced on every release |
| P7 | A decade | the domain core carries no substrate types (substrate rotation every 5–10 years); standard stewardship institutionalized |
I1 production-grade at every stage · I2 code is the complete source of truth · I3 reliable updates (pinned artifacts, rollback, deprecation policy) · I4 no operator IT competence required · I5 modern architecture and process.
Strategic mandates: the core is Rust; the UI is Rust with an extension host from day one; OCS fully implemented including the SDK — the foundation of the product; OCS never restricts the language of connected services; docs and licenses stay current; the runtime substrate is swappable on a ten-year horizon.
Deliver from code alone (parity), enforce process gates (CI tiers, regression gate, coverage floor, e2e and upgrade tests, SLSA L2), land OCS 1.0 + Rust SDK + conformance, the first Rust vertical slice under the M1 behavioral acceptance, the Rust UI with an extension host, finish M2/M3, the GUI installer, and the second substrate adapter. Exit: the platform deploys and updates from code alone on a clean cluster with zero manual steps; the behavioral acceptance is green; an external user lives on it.
The full mandatory catalog (Network, Volume, Image, Compute, Kubernetes, S3, Backup, Access, Support — each proven on a real backend including failures). A second independently managed site; a second engineer repeats operations from the docs; the 1.0 measurements (regional RPO 0 / RTO ≤5 min; off-cell RPO ≤15 min / RTO ≤60 min; ≥99.95% availability over 24 h). A release train with n-2 support, canaries, and rollback. One license policy. Exit: 1.0 released; entry into Era II at ≥3 external independent installations (proposed threshold — the owner confirms or sets their own).
The standard lives outside one repository: specification separate from implementation; ≥3 independent implementations (including ≥1 not ours); a conformance program with a public registry (CNCF model); a marketplace with mandatory signing. First third-party products in production at external users. Stewardship: a technical committee, a public standard roadmap, an RFC process. Exit: OCS is a language ≥3 independent teams speak without our participation.
The public cloud + partner installations; the community machine (good-first-issues, mentorship programs, multi-organization maintainers — the strongest survival predictor); CNCF Sandbox (once 2–3 independent implementations exist) → Incubating (CNCF criterion: 3+ independent dev/test adopters; our bar: production adopters) → Graduated (maintainers from ≥2 organizations). "Thousands of companies" follows the Kubernetes adoption curve (83% production share four years from announcement under the foundation model). Exit: the platform is a top-3 choice for a new private/sovereign cloud; the standard is de facto.
Provider federation: the P2P bus, cross-cloud connect, settlement (per the transformation plan, its 18–36-month horizon); ≥3 jurisdictions at the network level; no jurisdiction, company, or single operator can switch the network off (P4). Decentralized governance of the standard (foundation/association). Exit: purpose dimensions P1–P7 are met and confirmed by external audits.
- Era 0: IP held by the founder, Iurii Trukhin; a dedicated vehicle is the intended future owner ("ethereum foundation" model) — transfer not yet executed; stewardship form is an Era II decision.
- The team: one engineer + an agent team; "one engineer" is an operating invariant, not an organizational cap; the hiring trigger (proposed: when P2 genuinely breaks) is the owner's call — the sources do not set one.
- Every era has entry/exit criteria above; invariants I1–I5 hold on every release.
- Track life metrics (cores, installations, external products), not hype metrics (stars, summits) — the OpenStack lesson.
- Stop-losses: N iterations without progress → stop, re-plan, escalate to the owner.
Era durations are orientation from precedents (Linux 26 years; Kubernetes 4 years from announcement; the OpenStack hype/life divergence). No public case exists of an agent team running a production OSS platform for years — our compass is behavioral acceptance plus gates, not trust. Public clouds die in 2–6 years against hyperscalers: the speed of Eras 0–II is survival, not a race.