How we handle your data and your decisions.
Workforce touches your ERP, CRM, and email - systems that hold the financial and personal data your business runs on. Here is exactly how that works: what's built into monomind's open-source engine, and what we configure per engagement on top of it.
Local-first, no cloud vector DB
Monomind stores memory and embeddings in local SQLite on infrastructure you control. We do not ship your data to a third-party vector database. Embeddings run locally.
Bring your own keys (BYOK)
All LLM calls are made with API keys you provide and own. We never proxy your prompts through a shared account. You can rotate or revoke keys at any time without our involvement.
Human-in-the-loop on what matters
Monomind's org runtime includes a real policy engine that governs what each agent role can do autonomously - tool allow/deny lists, file scope restrictions, and audit trails are part of the open-source engine. During an engagement, we configure that layer with the approval rules specific to your business - like pausing on an invoice over a threshold you set - so a reasoning model is never the thing that decides what's allowed.
Full audit trail
Every worker action is logged with inputs, outputs, timestamps, and the policy version in effect. When we build ERP, CRM, or email actions into a worker during an engagement, those actions are logged the same way, so you can replay any decision after the fact.
Connectors are bridges, not data stores
When we build connectors to your ERP, CRM, and email during an engagement, they are standard bridges that read and write through your existing APIs. They hold no data of their own and can be swapped without touching process logic. We never copy your system of record into ours.
Versioned, testable policies
The policy engine underneath every worker is versioned and auditable at the engine level. The specific rules - approval thresholds, vendor matching logic, exception routing - are configured per engagement, and you can see exactly which policy version governed any past action and roll back if needed.
Want the technical detail?
The architecture behind every safeguard above is documented in full on the How It Works page, and the engine itself is open-source and auditable on GitHub.
Questions about data residency, encryption, SSO, or sub-processors for a specific engagement? Ask us directly.