Files
archy/.planning/PROJECT.md
T

116 lines
6.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Archipelago
## What This Is
Archipelago is a self-hosted personal-server platform: a Rust daemon (workspace at `core/`)
plus a Vue 3 frontend (`neode-ui/`, built to `web/dist/neode-ui/`) running on Debian nodes
with rootless Podman, managing ~40 declarative, manifest-driven apps (Bitcoin, Lightning,
mesh/LoRa, federation, media, and more). It ships as OTA-updated releases to a live fleet
and is actively shipping v1.7.x alpha releases. This milestone drives it to the
**developer-ready app platform** north star.
## Core Value
A third-party developer can publish an app via the signed/decentralized registry and a user
can install it on their node — every app manifest-driven, manifests shipped via the signed
registry (not OTA disk files), all rootless, secure, robust, and 100%-uptime-capable.
## Current State (brownfield baseline, 2026-07-29)
- Single-node production gate is **GREEN** (5/5 on .228, 2026-06-23) — that exit criterion is met.
- ~40 apps are manifest-based and Quadlet-migrated; all multi-container stacks use the
orchestrator stack pattern; the legacy per-app installer anti-pattern is deleted.
- Workstream B (registry-distributed manifests) phases 1+2 are code-complete; the signing
ceremony is done (release-root pinned in `anchor.rs`); the fleet flip is not yet authorized.
- Workstream C (marketplace) is design-only (`docs/marketplace-protocol.md`); no tooling or
trust UX built. Developer CLI suite (`archy app …`) does not exist yet.
- Phase-3 Quadlet default-flip is validated opt-in on .228/.198 but not default.
- Declared next exit criteria: the multinode pass (`docs/multinode-testing-plan.md`) and the
remaining workstreams.
## Requirements
### Validated
- ✓ Single-node lifecycle gate green 5× on .228 (install/UI/stop/start/restart/reinstall/
reboot-survive/daemon-restart-survive/uninstall) — 2026-06-23
- ✓ Manifest-driven app packaging for all ~40 apps incl. multi-container stacks (workstream A)
- ✓ Signed catalog + release-root signing ceremony (workstream B phases 1+2, code-complete)
### Active
See `.planning/REQUIREMENTS.md` — 20 v1 requirements across MNODE / LIFE / REG / SEC / DEV / MKT,
all mapped to phases in `.planning/ROADMAP.md`.
### Out of Scope
- Rootful containers, Docker, privileged containers — invariant (ADR-001/ADR-009)
- Per-app Rust installers / OS-level provisioning — the anti-pattern being deleted
- Centralized gatekept app store — decentralized Nostr marketplace instead (ADR-006)
- Web5 DWN spec compliance — deprioritized after TBD shutdown (ADR-011)
- Custom live voice-call protocol — deprioritized per user 2026-07-01; revisit later
- DHT/iroh distribution backbone (workstream D) — design-only, tracker-marked backlog; v2
## Context
- Repo: `core/` Rust workspace (no root Cargo.toml), `neode-ui/` Vue frontend, `apps/` manifests,
`tests/lifecycle/` + `tests/multinode/` gates, `docs/` authoritative plans.
- Authoritative narrative: `docs/PRODUCTION-MASTER-PLAN.md`; day-to-day open list:
`docs/UNIFIED-TASK-TRACKER.md`. Codebase map: `.planning/codebase/ARCHITECTURE.md` +
`.planning/codebase/CONCERNS.md`.
- Known debt informing this milestone (from CONCERNS.md): federation tombstone-write errors
swallowed; reconciler has no flap observability and no failed-unit self-healing; generated
AppArmor profiles are never applied; multinode test harness curl calls lack timeouts;
SPEC validation is narrower than ADR-009's mandates (non-root UID, pinned tags, seccomp).
- Fleet is live and OTA-updated; all destructive verification happens on designated test
nodes per the deploy roster — never uninvited on in-use nodes.
## Constraints
- **Security**: Rootless Podman only; manifest-declared secrets (0600, never logged);
mandatory container security defaults enforced at manifest level (ADR-009)
- **Data safety**: Migrations never destroy data — preserve `/var/lib/archipelago/<app>`,
secrets, credentials, ports, adoption container names; always a rollback path
- **Verification**: Real-node verification before any tag; lifecycle gate runs ON the node,
not via RPC; mesh changes need real-RF E2E tests; re-run the gate after orchestrator changes
- **Process**: Commit + push every unit of work (`git push gitea-ai main`); stage by explicit
path; deploy to the dev pair before any OTA; never commit secrets
- **Tech stack**: Rust (Tokio/Hyper, JSON-RPC 2.0) backend; Vue 3 + Pinia frontend;
Quadlet/systemd-user container units; Ed25519-signed release artifacts
## Key Decisions
<decisions>
All ten ADRs below are **locked** (Status: Accepted; ingest source `docs/adr/*.md`). They are
non-negotiable inputs to planning and cannot be overridden without a new ADR.
| ID | Decision | Scope |
|----|----------|-------|
| ADR-001 | Podman over Docker — rootless, daemonless, systemd-native; `archy-net` for inter-container DNS | Container runtime |
| ADR-002 | `did:key` (Ed25519) node identity — self-contained, offline-capable; gaps mitigated via federation trust lists | Identity |
| ADR-003 | Nostr relays (NIP-78, kind 30078) for node + app discovery — multi-relay query, 15-min cache, trust scoring, Tor-compatible | Discovery |
| ADR-004 | Tor hidden services for inter-node RPC/control plane — bulk data via registries, not Tor | Federation transport |
| ADR-005 | ChaCha20-Poly1305 + Argon2id (64MB, 3 iter) for backup encryption | Backups |
| ADR-006 | Nostr relays for marketplace discovery — DID-signed manifests, trust tiers (Verified/Community/Unverified), signature verification before install | Marketplace |
| ADR-007 | Bilateral DID federation trust via single-use invite codes; Trusted/Observer/Untrusted levels | Federation trust |
| ADR-008 | Dual keys from one master seed — Ed25519 canonical identity, secp256k1 for Nostr/Bitcoin/Lightning, linked via NIP-05 | Keys |
| ADR-009 | Manifest-level container security enforcement — readonly_root, no_new_privileges, non-root UID, drop-ALL caps, pinned tags, seccomp; overrides explicit + audited | Container security |
| ADR-011 | DWN deprioritized — keep custom `dwn_store.rs`, stop branding as Web5, invest in Nostr + Tor federation instead | Peer data sync |
(ADR-010 does not exist in the repo — numbering gap, noted in `.planning/INGEST-CONFLICTS.md`.)
</decisions>
Milestone-level decisions:
| Decision | Rationale | Outcome |
|----------|-----------|---------|
| Milestone version = 1.8.0-alpha | Decided 2026-07-08 per tracker | — Pending ship |
| Workstream D (DHT) deferred to v2 | Design-only, tracker-marked backlog; not needed for north-star metric | — Pending |
| App manifest canonical schema = `core/container/src/manifest.rs` | SPEC self-declares code wins over doc | ✓ Good |
| Phase-3 Quadlet flip gated on multinode gate reporting clean | Prior uncommitted-flip confusion; flip fresh as a 2-line change when gate is clean | — Pending |
---
*Last updated: 2026-07-29 after intel ingest (10 ADRs + 1 SPEC) + codebase mapping*