pnpm install --frozen-lockfile failed on a stale lockfile; pnpm 10 also refused
to run sharp's build script (ERR_PNPM_IGNORED_BUILDS). Use --no-frozen-lockfile
and add pnpm.onlyBuiltDependencies (sharp, esbuild, @tailwindcss/oxide).
Co-authored-by: Hanzo Dev <[email protected]>
The api crashed on import after a half-finished model refactor, and the qdrant
healthcheck used a binary not in the image.
- compose.yml: qdrant ships no wget/curl, so its healthcheck always failed and
blocked the api (depends_on service_healthy). Dropped the broken healthcheck and
gate the api on qdrant service_started (qdrant itself serves fine on :6333).
- db/__init__.py: reconcile to the real models — drop removed Node + Cluster
(now CRDs) and MPCWalletModel; export the actual set (User, OrgCluster,
SmartWallet, Subscription, TeamMember, etc.).
- api/mpc + api/keys: MPCWalletModel -> SmartWallet (the renamed model).
API now boots healthy: GET /health 200 {"status":"healthy","version":"2.0.0"},
/docs serving, 42 node/chain/fleet endpoints live (fleets deploy/scale, presets,
pod health/logs). Full local stack up: postgres, redis, clickhouse, qdrant,
meilisearch, api.
Note (follow-up, not blocking boot): api/nodes/db_nodes.py + db/seed_nodes.py
still import removed Node/Cluster models — dead code to remove or port to the
fleet/CRD path; the web frontend build fails on a stale pnpm lockfile.
Co-authored-by: Hanzo Dev <[email protected]>
These two repos share root commit 8999f3c3 ("Initial commit") but
diverged long ago — hanzoai/bootnode evolved the recent stack
(api-go, brand package, etc.) while bootnode-org/main retained the
mainline 2026 history (175 commits we don't have).
Brand-leak fix: this work belongs in github.com/bootnode/bootnode,
not github.com/hanzoai/bootnode. Per the no-cross-brand-leak policy,
merge both histories together and push the result back to
bootnode/bootnode as the canonical home.
After this lands:
- bootnode/bootnode is the canonical repo
- hanzoai/bootnode gets archived with a redirect notice
- @bootnode-io/brand npm package's `repository.url` already points at
bootnode/bootnode (set in the earlier package.json scaffold)
# Conflicts:
# .github/workflows/ci.yml
# api/bootnode/api/__init__.py
# api/bootnode/api/billing/__init__.py
# api/bootnode/config.py
# api/bootnode/core/billing/__init__.py
# api/bootnode/db/models.py
# api/bootnode/main.py
# api/tests/test_billing.py
# web/app/dashboard/infrastructure/page.tsx
# web/app/dashboard/organization/billing/page.tsx
# web/app/dashboard/organization/page.tsx
# web/app/dashboard/page.tsx
# web/app/layout.tsx
# web/components/navbar.tsx
# web/components/unified-navigation.tsx
# web/lib/brand.ts
# web/public/logo/zoo-icon.svg
# web/public/logo/zoo-logo-white.svg
# web/public/logo/zoo-logo.svg
# web/tsconfig.tsbuildinfo
Adds a Go subsystem alongside the existing Python FastAPI service so
bootnode can fold into `superbase` per HIP-0106. The Python service in
`api/bootnode/` stays untouched.
What landed:
- `schema/bootnode.zap` — public typed surface (chains, RPC, tokens,
NFTs, transfers, smart wallets, ERC-4337, gas, webhooks, fleets,
nodes, usage). No `.capnp` anywhere — ZAP only, per brand policy.
- `zap/gen/` — hand-rolled Go bindings that match what `zapc generate`
emits, so `go build` works without zapc installed locally.
- `api/mount.go` — the HIP-0106 contract:
func Mount(r chi.Router, deps Deps) error
mounts every FastAPI-parity route group through `api/handlers/`.
- `api/deps.go` — cross-subsystem ZAP client interfaces (IAM, KMS,
Base, O11y, Commerce). No business logic, no hard imports — keeps
the Bootnode module independent of sibling subsystems until they
ship their typed clients.
- `api/handlers/` — 149 chi routes mounted (Python has 150). Each
handler body is a 501 stub with a `TODO(bootnode-port)` pointing at
the Python file + symbol to port.
- `api/middleware/` — IAM-backed auth, OTEL-friendly telemetry,
per-org rate limit (stub).
- `api/store/` — per-tenant SQLite access via `deps.Base.Open(orgID,
"bootnode")` per HIP-0302.
- `api/web3/` — chain registry + RPC pool surface (production version
prefers luxfi/* over upstream go-ethereum).
- `cmd/bootnode/main.go` — standalone shim; same `api.Mount` superbase
calls in production.
- `docs/PYTHON_TO_GO_PORT.md` — endpoint-by-endpoint port map.
Build: `go build ./...` and `go vet ./...` clean. Standalone shim
smokes 200 on /health, /ready, /v1/chains; 401 on auth-required
endpoints with no JWT; 501 on every TODO route.
Out of scope (not in this commit):
- Handler bodies (Phase 2 — porting in RPC → tokens → NFTs → ...
priority order).
- superbase wiring (lands when the superbase repo opens its main).
- Python service removal (stays until handlers reach prod parity).
Required for hanzoai/.github/.github/workflows/docker-build.yml@main —
without it the workflow_call dies as startup_failure with no jobs
dispatched. Caller permissions are a CEILING.
Replaces all ghcr.io/hanzoai/<svc>:latest pins with the latest published
semver tag for each service. Per CLAUDE.md auto-bump policy: mutable
branch tags (:latest, :main, :dev) are deprecated for cluster pins —
only immutable semver permitted.
Bulk update across services with published v* tags. Services without a
published semver remain on :latest until their release pipeline cuts a
v* tag.
generateMetadata() already used the host header, but RootLayout itself
read only process.env.BRAND with a hard-coded "bootnode" fallback. That
made the body theme/SSR brand drift from the metadata brand on per-host
deploys. Make RootLayout async and reuse the host detection so both
paths agree.
GHCR 403s on first-push package creation for ghcr.io/hanzoai/bootnode-web.
Push web build to the existing bootnode package with tag-suffix=-web,
api with tag-suffix=-api, so they coexist without colliding tags.
Deployment image becomes:
ghcr.io/hanzoai/bootnode:vX.Y.Z-web
ghcr.io/hanzoai/bootnode:sha-XXXXXXX-web
Update K8s deployments via kubectl set image after CI succeeds.
The zoo brand config in web/lib/brand.ts references /logo/zoo-icon.svg,
/logo/zoo-logo.svg, /logo/zoo-logo-white.svg but only the references
existed — the files were missing, producing 404s on web3.zoo.ngo and
zoo.cloud. Added the canonical Zoo CMYK Venn mark from
zoo.industries/public/zoo-logo.svg.
- web/lib/brand.ts: detect zoo.cloud as zoo brand (alongside zoo.ngo, zoo.network)
- infra/k8s/ingress.yaml: add zoo.cloud + www.zoo.cloud as additional hosts on
bootnode-zoo-ingress (TLS SAN via cert-manager, rules → bootnode-web-zoo:3001)
Makes zoo.cloud serve the same Bootnode platform as web3.zoo.ngo with Zoo brand.
Migrate every ingress resource from nginx annotations to
traefik.ingress.kubernetes.io for hanzoai/ingress controller.
Update network policies to match hanzo-ingress pod selector.
Update operator config to use className: hanzo.
NEXT_PUBLIC_BRAND gets baked at build time by Next.js, always resolving
to the Dockerfile default ("bootnode"). Use headers() + Host-based
detection in generateMetadata() so cloud.lux.network shows "Lux Cloud"
at request time, not "Bootnode".
Root cause: ghcr-secret was expired/missing, causing ImagePullBackOff
on all web portal pods. Now auto-refreshes using GITHUB_TOKEN on each
deploy. Also cleans up stale pending pods from previous failed rollouts.
Instead of storing DIGITALOCEAN_ACCESS_TOKEN and cluster names as
GitHub Secrets, authenticate to KMS at deploy time and fetch them
from /deploy path. Falls back to GitHub Secrets if KMS is unavailable.
Matches the pattern used by hanzo/app and other Hanzo services:
only KMS_CLIENT_ID + KMS_CLIENT_SECRET stored in GitHub.
The build-and-push and deploy jobs were duplicated in ci.yml with missing
REGISTRY env vars, causing Docker login failures. docker-release.yml is
the canonical deploy pipeline (triggers on CI workflow_run completion).
- Add Explorer product page with branded block explorer deployment, contract
verification, Etherscan-compatible API, and multi-chain support
- Add MPC product page with deep Lux MPC integration: CGGMP21 (ECDSA),
FROST (EdDSA), LSS resharing, 15+ chain support, threshold signing API
- Update landing page with 6-product grid (Node, Data, Wallets, Explorer,
MPC, Rollups) and 8 API cards including Explorer API and MPC Signing
- Update navbar and command menu with Explorer and MPC products
- Add Lux fleet management dashboard, observability, and services pages
- Add AI chat component and unified navigation with org switcher
- Add Kubernetes operator (Rust) for multi-chain node lifecycle management
- Add Lux Helm charts with genesis configs for devnet/testnet/mainnet
- Add docker-release.yml CI/CD pipeline for automated DOKS deployment
- Add monitoring, network policies, web deployments, and ingress updates
- Refactor API: fleet management, billing tracker, observability endpoints
- Update brand config, API client, and infrastructure manifests
The K8s DATABASE_URL env var uses postgresql:// which makes
create_async_engine try to load psycopg2 (sync driver). Add
_ensure_async_dsn() to convert postgresql:// and postgres:// to
postgresql+asyncpg:// at engine creation time. asyncpg is already
in dependencies.
Add BOOTNODE_GHCR_PAT secret-based login to push API and web images
to the canonical ghcr.io/bootnode/bootnode registry in addition to
ghcr.io/hanzoai/bootnode. Also add provenance=false/sbom=false to web
build to avoid attestation timeouts on linux/amd64.
capnp.AsyncIoStream.create_server() must be called from within an active
KJ event loop context. Wrap the entire serve lifecycle in a background
asyncio Task that runs under `async with capnp.kj_loop()`, fixing the
startup error: "The KJ event-loop is not running on this thread".
The server now signals readiness via `_running=True` after bind and
the caller polls for up to 1 s before raising if startup fails.
zeekay
