arcadedb
CLAUDE.md1 starsChanged 5 months agoApache-2.0View on GitHub
Read this first. This file is the AI operating model's contract. It is loaded into every Claude Code session in this repo. Nested CLAUDE.md files add directory-specific rules (terraform/, helm/, control-plane/, docs/). The hooks in .claude/settings.json enforce the non-negotiables below deterministically — they are not suggestions. A production-grade home on AWS for ArcadeDB, the data-layer foundation of a multi-tenant Knowledge Base for an AI SaaS. One virtual DB per tenant; two jurisdictions (EU + US) with hard GDPR residency; built…
In this repository20 other files it gives its agents.
- control-plane/CLAUDE.mdCLAUDE.md
- docs/CLAUDE.mdCLAUDE.md
- helm/CLAUDE.mdCLAUDE.md
- terraform/CLAUDE.mdCLAUDE.md
- add-cellSkill
- cell-capacity-reportSkill
- deprovision-tenantSkill
- dr-drillSkill
- incident-triageSkill
- migrate-schemaSkill
- new-adrSkill
- provision-tenantSkill
- restore-tenantSkill
- retire-cellSkill
- review-terraform-planSkill
- rotate-secretsSkill
- security-baseline-checkSkill
- tenant-erasureSkill
- tenant-usage-reportSkill
- upgrade-arcadedbSkill
The file
Files
CLAUDE.md — ArcadeDB-on-AWS KB Platform (root project memory)
Read this first. This file is the AI operating model's contract. It is loaded into every Claude Code session in this repo. Nested
CLAUDE.mdfiles add directory-specific rules (terraform/, helm/, control-plane/, docs/). The hooks in.claude/settings.jsonenforce the non-negotiables below deterministically — they are not suggestions.
What this repo is
A production-grade home on AWS for ArcadeDB, the data-layer foundation of a multi-tenant Knowledge Base for an AI SaaS. One virtual DB per tenant; two jurisdictions (EU + US) with hard GDPR residency; built and operated with Claude Code for a clean hand-over to a cloud-ops team.
Current phase: CTO approval package (Phase D). The HLD + boilerplate IaC +
this AI kit ship for sign-off before any AWS spend. Nothing is applied to
AWS. Do not run terraform/tofu apply, create cloud resources, or run
cloud-mutating commands until the CTO approves and we enter Phase 0.
Authoritative design: docs/architecture.md (HLD) ·
docs/adr/ (one ADR per decision) · docs/assumptions.md.
The 7 prime directives — NEVER violate these
- Residency — EU tenant data + backups stay in EU regions; DR pairs stay in-jurisdiction; no EU↔US data path exists. (ADR-0007)
- Version floor — ArcadeDB ≥ 26.4.1 (closes the CVSS-9.0 cross-DB isolation CVE + Raft HA). Re-audit isolation after every upgrade. (ADR-0012)
- Quorum — every prod cell = 3 nodes, one per AZ, PDB
minAvailable: 2. Never drop below quorum. Non-prod may be single-node. (ADR-0010) - No public database — ArcadeDB ports (2480/2424/2434/5432/6379/7687) are never on a public subnet or public LB.
- Encrypt everything at the platform layer (EBS/S3/Secrets/snapshots via KMS) — the engine provides none. (F4)
- No click-ops — every resource is Terraform/Helm/GitOps; reproducible from clean state. Always plan before apply; prod apply needs manual approval. (ADR-0020)
- Sizing rule — pod memory limit ≥
maxPageRAM+ JVM heap + overhead, or the kernel OOM-kills a node and risks quorum.
Load-bearing ArcadeDB facts (why this isn't a generic "DB on K8s" repo)
- HA is leader-based Raft, GA from 26.4.1, min 3 nodes; replication is per-database; reads can fan to replicas, writes go to the leader.
- No per-DB resource quotas → the control plane caps cell capacity and the retrieval proxy enforces per-tenant runtime limits (ADR-0027).
- No native encryption, no native audit, root password is set-once → encrypt at AWS layer; build an app-layer audit trail; "rotate root" = new admin user.
- Backup is hot per-DB ZIP, excludes WAL, no incremental/PITR, no S3 target; restore requires the target DB to NOT exist. Supplement with EBS snapshots.
- Official Helm chart (StatefulSet), no Operator → we own upgrades;
/ready(HTTP 204) for probes;/prometheushas a MIME-type bug (workaround in the chart). - Single-leader write ceiling per cell; a DB can't be split across nodes → scale by adding cells, not by enlarging a Raft group.
Tech stack + pinned versions
| Thing | Pin | Where |
| ArcadeDB | ≥ 26.4.1 (pin a digest in prod) | helm/arcadedb/values.yaml, cell module |
| OpenTofu / Terraform | ≥ 1.10 (S3-native locking) | terraform/**/versions.tf |
| AWS provider | >= 5.80, < 6.0 | versions.tf |
| Helm / kubernetes / random providers | pinned ranges | versions.tf |
| Helm chart | 0.1.0 | helm/arcadedb/Chart.yaml |
| EKS | 1.31 | env tfvars |
| GitHub Actions | pinned (SHA in prod) | .github/workflows/ |
Repo map
docs/ HLD (architecture.md), adr/, assumptions.md, runbooks/ ← design + reasoning
terraform/ landing-zone/, modules/{network,eks,cell,backup-dr,observability}/, environments/
helm/arcadedb/ the cell chart (StatefulSet, PDB, probes, sizing, MIME workaround)
control-plane/ registry schema, router/retrieval interfaces, Step Functions ASL (stubs)
policy/conftest/ OPA residency + no-public-DB gates (+ unit tests)
.github/workflows/ CI policy gates
.claude/ settings.json hooks + skills/ (this AI operating model — handed to ops)Tiering rules (decide cell placement + config)
- Standard tenants → pooled cells (namespace in a shared EKS cluster),
txWalFlush0/1, placed by capacity caps (~150 DBs / 60% maxPageRAM / 70% disk). - Enterprise/regulated tenants → dedicated cells (optionally a dedicated
EKS cluster),
txWalFlush=2(fsync), per-tenant CMK option, 1h backups, Object Lock. - The boundary we trust for sensitive data is the dedicated cell. Pooled cells are a cost optimisation guarded by version floor + NetworkPolicy + isolation probe.
Conventions (enforced as CLAUDE.md rules + CI)
- Every decision gets an ADR stating its reasoning (use
docs/adr/0000-template.mdor thenew-adrskill). Every assumption gets adocs/assumptions.mdentry (rationale · impact-if-wrong · validation owner · linked ADR). - Match the surrounding style. Pin all versions. Tag every resource with
platform/geo/env/module/managed-by/residency-boundary. - Validate before you commit:
make validate(fmt, validate, tflint, conftest, helm lint, kubeconform). All offline; no AWS.
How to work in this repo (Claude)
- Plan before apply. For Terraform, propose a plan and run the
review-terraform-planskill before any apply (post-approval only). - Use the skills in
.claude/skills/for operational procedures — they carry the safety checks. Day-2 ops:provision-tenant,add-cell,upgrade-arcadedb,restore-tenant,dr-drill,rotate-secrets,cell-capacity-report,incident-triage,migrate-schema,tenant-usage-report,tenant-erasure. Build-time:review-terraform-plan,new-adr,security-baseline-check. - If something is ambiguous or you hit a real fork, ask — don't guess on residency, quorum, version, encryption, or public-exposure questions.
- Never lower the ArcadeDB image below 26.4.1, set
replicas < 3on a prod cell, remove a PDB, open a DB port to0.0.0.0/0, add an out-of-geo region literal, or put a secret in a diff. The hooks will block these; don't try to work around them — fix the underlying change.
Definition of done (for any change here)
make validateis green. 2. ADR added/updated if it's a decision.docs/assumptions.mdupdated if it rests on a new assumption.- The relevant nested
CLAUDE.md/ skill / runbook updated. 5. No prime-directive violation.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

