Multi-Tenancy for Agents

Workspace Isolation

Every agent is owned by a workspace and can only access that workspace’s data.

How isolation is enforced:

1. Agent creation — workspace_id is extracted from the JWT at creation time and stored immutably
2. Tool calls — All tool calls include workspace_id in the request context. OpenSearch queries are automatically scoped to tenant-suffixed indexes
3. Execution logs — Execution records are stored with workspace_id and filtered in all API responses
4. Agent Runner — Fetches agents per-workspace; each execution runs in a workspace-scoped context

// Tool calls always include workspace context
async function callSemanticSearch(query: string, context: AgentContext) {
  const index = `${DEFAULT_INDEX}-${context.workspaceId}`;
  return semanticSearchClient.search({ query, index });
}

RBAC for Agents

Role	Create Agents	View All Agents	Edit Others	Delete
Admin	✅	✅	✅	✅
Analyst	✅	Own only	Own only	Own only
Viewer	❌	Own workspace	❌	❌

Role-based tool access

Role	Available Tools
Viewer	None (cannot create agents)
Analyst	semantic_search, opensearch_query, get_scan_results, analyze_logs
Admin	All tools including future write/external categories

Rate Limiting

Rate limits are enforced per workspace at the Backend API layer:

Tier	Agents	Executions/day	Tools/execution
Free	2	50	5
Pro	20	500	20
Enterprise	Unlimited	Unlimited	Unlimited

When a rate limit is hit:

• Execution is queued (not dropped) if within 2x the daily limit
• Agent transitions to paused state with rate_limited reason
• Discord notification sent to workspace admin
• Agent auto-resumes at midnight UTC

Cross-Workspace Visibility

System agents (type 1) are defined at the platform level but execute in per-workspace contexts. When a System Agent runs for workspace abc, it only sees abc’s data.

Admins can view execution logs across all workspaces via an internal admin endpoint (not exposed in the public API).

Data Retention

Data	Retention
Agent execution logs	90 days
Agent configs	Permanent (until deleted)
Alert history	30 days

Retention policies are enforced by a nightly cleanup job on the Backend.

Audit Trail

All agent CRUD operations are logged to the Logrite audit trail:

{
  "timestamp": "2026-03-02T08:15:00Z",
  "action": "agent.create",
  "actor": "user:[email protected]",
  "workspace": "acme-corp",
  "details": {
    "agentId": 42,
    "agentName": "Payment Error Monitor",
    "toolsAllowed": ["semantic_search", "opensearch_query"]
  }
}

See Audit Logs for how to query the audit trail.