Data Inventory

For enterprise security, compliance, and data-governance teams evaluating Dandori.


Summary

# Category Storage PII risk Sensitivity
1 Workspace metadata (projects, teams, agents) DB None Low
2 User accounts + API keys DB (hashed) Low (names, emails) Medium
3 Context Hub content (5 layers) DB High potential Varies by layer
4 Skill Library content DB Medium Medium
5 Task data DB Low Medium
6 Agent run records (prompts + outputs) DB + object storage Highest Match highest input
7 Approval records DB (append-only) Low Restricted
8 Audit log events DB (append-only) Low Restricted
9 Cost / billing records DB None Low
10 Lifecycle hook execution records DB (append-only) Variable Medium
11 Tool metadata + usage (MCP + others) DB None Low
12 Quality gate + evaluation results DB Low Low
13 Sub-agent run traces (rollup of run records) DB Low Match parent
14 Imported context (Confluence, Drive, GitHub) DB Inherits source Inherits source
15 Integration credentials Sealed secrets None Secret
16 Session tokens DB (hashed) Low Restricted

Two categories deserve the strictest review: Context Hub content (#3) and agent run records (#6) — both can contain arbitrary text.


What Dandori does NOT store

Category Where it lives instead
AI provider API keys Host env vars of the runtime — Dandori never sees them
Model weights Provider-side only
Source code repositories GitHub / local filesystem
Customer production data Production DBs — Dandori only sees what engineers paste into context
Tool call arguments/results Runtime invokes tools directly — Dandori logs metadata only

Key controls

  • Append-only tables for audit log, approvals, hook executions (UPDATE/DELETE blocked at DB level)
  • Optional hash chain on audit log for tamper-evidence
  • Version control on all configurable entities (context, skills, agent templates, hooks, tool descriptions) with diff + rollback
  • PII tags per context layer; agents need explicit permission for PII-tagged layers
  • Secret scanner warns before committing context matching known secret patterns
  • Run records immutable after creation; export requires admin + produces audit event
  • Retention configurable per project (default: runs 365 days, audit 365 hot + 7 years cold)

Data residency

Dandori is self-hosted. All data stays within the customer’s infrastructure. No phone-home, no Dandori-operated SaaS endpoints.


Review questions

  1. Which data categories are blocked by your policy?
  2. What retention windows apply? (Default 365 days for runs)
  3. Does run prompt/output need encryption at rest?
  4. Which ecosystem integrations are allowed in the pilot?
  5. Who owns each RBAC role (Admin, Compliance, Team leads)?
  6. Single-tenant or multi-tenant deployment?

For full per-category detail (storage, retention, access, controls), see the module detail pages or request the extended data inventory document.


Outer Harness → Back to the concept behind Dandori

hello@dandori.io — Start a pilot conversation