SAOS Governance — public documentation for the Sovereign Agentic Operating System governance model: agent policy, cost gates, approval workflow, and audit architecture.
Active. SAOS v1 governance model operational. Audit log architecture deployed.
policy/— Agent policy documentation (what agents can and cannot do autonomously)approval-workflow/— Approval workflow spec (what requires founder approval before execution)cost-gates/— Cost gate policies (per-run and per-day spend limits)audit/— Audit trail architecture + cryptographic logging speccontext-packs/— Documentation for the 11 context packs that govern agent behavioroutput-contracts/— 14 output contract definitions (what each agent is permitted to output)
SAOS (Sovereign Agentic Operating System) operates under a four-layer governance model:
| Layer | Mechanism | Implementation |
|---|---|---|
| L1 Infrastructure | Policy Enforcer kill switch + cost gates | policy-enforcer.ts |
| L2 Policy Enforcement | PHI detection + output contracts | output-contract-validator.ts |
| L3 Cryptographic Audit | AALLogger — hash-chained cognitive traces | aal-logger.ts |
| L4 Human Override | Founder approval queue for high-stakes actions | /ops/decisions dashboard |
No agent can take an external action (send email, publish content, make financial transaction) without passing through an approval gate. The approval-handler enforces this at the database level with TOCTOU-safe optimistic concurrency.
| Action type | Agent | Approval required |
|---|---|---|
content_publish |
content-agent | YES — founder reviews before public publication |
external_send |
investor-agent, grant-agent | YES — founder reviews before external email |
financial_action |
revenue-agent | YES — any financial transaction |
regulatory_submission |
protocol-agent | YES — compliance-critical |
data_export |
data-agent | YES — data governance |
Governance documentation contributions welcome from AI safety researchers and governance experts. See CONTRIBUTING.md.
CC-BY-4.0
EvidenceOS Engineering — engineering@evidenceos.com
clinical-safety-governance— Clinical AI safety layerceaih-scientific-council— Scientific oversight