External architecture review flagged the SECRETS_ENCRYPTION_KEY env var on the platform as encryption-at-rest theater. The reviewer read only the platform repo and missed that the master key actually lives in AWS KMS at the control plane layer, with envelope encryption wrapping each tenant secret blob. Adds docs/architecture/secrets-key-custody.md as the canonical source of truth for the full chain: - Two-mode envelope (KMS_KEY_ARN vs static-key fallback) - Per-blob AES-256-GCM with KMS-wrapped DEKs - Where each key actually lives (KMS, CP env, tenant env) - Threat model per attacker capability - Rotation story (annual KMS CMK rotation, manual DEK rotation on incident) - Audit posture (SOC2 / ISO 27001 questionnaire bullets) Patches three downstream docs that previously stopped at the env-var level and link them to the new custody doc: - development/constraints-and-rules.md (Rule 11) - architecture/database-schema.md (workspace_secrets paragraph) - architecture/molecule-technical-doc.md (env-vars table) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
7.7 KiB
Database Schema
Postgres Tables
workspaces — Workspace Registry (Current State)
The mutable projection of structure_events. Represents the current state of all workspaces.
CREATE TABLE workspaces (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
name TEXT NOT NULL,
role TEXT,
tier INTEGER DEFAULT 1,
status TEXT DEFAULT 'provisioning',
source_bundle_id TEXT,
agent_card JSONB,
url TEXT,
parent_id UUID REFERENCES workspaces(id),
forwarded_to UUID REFERENCES workspaces(id),
last_heartbeat_at TIMESTAMPTZ,
last_error_rate FLOAT DEFAULT 0,
last_sample_error TEXT,
active_tasks INTEGER DEFAULT 0,
uptime_seconds INTEGER DEFAULT 0,
created_at TIMESTAMPTZ DEFAULT now(),
updated_at TIMESTAMPTZ DEFAULT now()
);
| Column | Purpose |
|---|---|
id |
Unique workspace identifier |
name |
Display name |
role |
The org chart role (e.g. "Marketing", "QA") |
tier |
1-4, determines deployment method |
status |
provisioning, online, degraded, offline, failed, or removed |
agent_card |
Full A2A Agent Card as JSONB |
url |
Current endpoint URL |
parent_id |
Parent workspace (defines hierarchy AND communication topology) |
source_bundle_id |
Original bundle ID this workspace was created from |
forwarded_to |
Redirect pointer when workspace is replaced, expanded, or moved (see Registry — Workspace Forwarding) |
last_heartbeat_at |
Timestamp of last heartbeat received |
last_error_rate |
Latest self-reported error rate (triggers degraded at >= 0.5) |
last_sample_error |
Latest sample error message (shown on canvas tooltip) |
active_tasks |
Number of tasks currently running (shown as busy indicator on canvas) |
uptime_seconds |
Seconds since container start |
current_task |
Human-readable description of what the agent is currently working on (set via heartbeat) |
agents — Agent Assignments
CREATE TABLE agents (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
workspace_id UUID REFERENCES workspaces(id),
model TEXT,
status TEXT DEFAULT 'active',
removed_at TIMESTAMPTZ,
removal_reason TEXT,
created_at TIMESTAMPTZ DEFAULT now()
);
Tracks which AI model is assigned to which workspace, and the history of assignments.
workspace_secrets — Encrypted Credentials
CREATE TABLE workspace_secrets (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
workspace_id UUID REFERENCES workspaces(id),
key TEXT NOT NULL,
encrypted_value BYTEA NOT NULL,
created_at TIMESTAMPTZ DEFAULT now(),
updated_at TIMESTAMPTZ DEFAULT now(),
UNIQUE(workspace_id, key)
);
Stores API keys, credentials, and other secrets needed by workspace agents. Values are encrypted with AES-256-GCM at the application layer. The encryption key comes from the tenant's SECRETS_ENCRYPTION_KEY environment variable, provisioned at tenant boot by the control plane (which holds the master key in AWS KMS — see secrets-key-custody.md). The key is never stored in the database.
The provisioner reads secrets from this table, decrypts them, and injects them as environment variables when spinning up workspace containers. Secrets are never included in bundles (see Constraints — Rule 5).
canvas_layouts — Node Layout
CREATE TABLE canvas_layouts (
workspace_id UUID REFERENCES workspaces(id) ON DELETE CASCADE,
x FLOAT NOT NULL DEFAULT 0,
y FLOAT NOT NULL DEFAULT 0,
collapsed BOOLEAN DEFAULT false,
PRIMARY KEY (workspace_id)
);
Stores the visual position and UI state of each workspace node on the canvas. One row per workspace. Updated via PATCH /workspaces/:id when the user drags a node.
canvas_viewport — Viewport State
CREATE TABLE canvas_viewport (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
x FLOAT NOT NULL DEFAULT 0,
y FLOAT NOT NULL DEFAULT 0,
zoom FLOAT NOT NULL DEFAULT 1,
saved_at TIMESTAMPTZ DEFAULT now()
);
Single row — upserted on viewport change. Stores the canvas pan and zoom position so the user returns to the same view. Separate from canvas_layouts to avoid bloating the per-node table.
structure_events — Immutable Event Log
CREATE TABLE structure_events (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
event_type TEXT NOT NULL,
workspace_id UUID,
agent_id UUID,
target_id UUID,
payload JSONB,
created_at TIMESTAMPTZ DEFAULT now()
);
Append-only. Never UPDATE or DELETE rows. See Event Log.
activity_logs — Operational Activity Log
CREATE TABLE activity_logs (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
workspace_id UUID NOT NULL REFERENCES workspaces(id) ON DELETE CASCADE,
activity_type TEXT NOT NULL,
source_id UUID,
target_id UUID,
method TEXT,
summary TEXT,
request_body JSONB,
response_body JSONB,
duration_ms INTEGER,
status TEXT DEFAULT 'ok',
error_detail TEXT,
created_at TIMESTAMPTZ DEFAULT now()
);
| Column | Purpose |
|---|---|
activity_type |
a2a_send, a2a_receive, task_update, agent_log, or error |
source_id |
Workspace that initiated the action (nullable for canvas-originated) |
target_id |
Workspace that received the action (for A2A communications) |
method |
A2A method name (e.g. message/send) or task action |
request_body |
Full request payload as JSONB (for A2A) |
response_body |
Full response payload as JSONB (for A2A) |
duration_ms |
Operation duration in milliseconds |
status |
ok, error, or timeout |
error_detail |
Error message when status is not ok |
Separate from structure_events by design: structure_events tracks lifecycle/structural changes (append-only, never deleted), while activity_logs tracks operational activity (A2A communications, task updates, agent logs) and has a 7-day retention policy with automatic cleanup.
Indexes: composite (workspace_id, activity_type, created_at DESC) for filtered queries, standalone created_at for retention cleanup.
Redis Keys
| Key Pattern | Value | TTL | Purpose |
|---|---|---|---|
ws:{id} |
"online" |
60s | Liveness detection |
ws:{id}:url |
"https://..." |
5min | URL cache for fast resolution |
events:broadcast |
pub/sub channel | -- | Push events to canvas/workspace WebSocket |
Redis Configuration
Keyspace notifications must be enabled for liveness detection:
notify-keyspace-events = KEA
This allows the platform to subscribe to key expiry events without polling.
Design Decisions
- Postgres is source of truth. Redis is ephemeral. If Redis is wiped, workspaces re-register on next heartbeat and state is restored. Nothing critical lives only in Redis.
org_idis omitted from MVP schema. Added later in the SaaS migration for multi-tenancy.wal_level=logicalis set from the start to enable future streaming of change events without a schema migration.
Related Docs
- Event Log — Event sourcing pattern
- Registry & Heartbeat — How Redis keys are managed
- Platform API — API that reads/writes these tables
- Provisioner — Workspace lifecycle states
- Workspace Tiers — What the
tiercolumn means - Communication Rules — How
parent_iddrives access control