Frequently asked questions
Moflet AI is the agent platform for regulated environments: native and external (A2A) agents, MCP tools, and cited answers over your knowledge. Identity, authorization and audit from day one, on cloud, VPC or air-gap.
What is Moflet AI?
Moflet AI is the agent platform for regulated environments. It builds native agents, imports external agents over A2A and uses MCP tools, with cited answers over the tenant's knowledge. Identity, authorization and audit are there from day one, on cloud, VPC or air-gap.
How is it different from a chatbot?
Chat is the control plane, not the product. The runtime is yours: agents, tools and evidence live on the same host. Every run can be explained (allow, deny, tool call, signed export), and the workspace includes data, documents, workflows, semantics and dashboards that agents operate.
Where does it deploy?
The same product runs in three perimeters: multi-tenant (logical isolation, many organizations), dedicated (the same chart, one tenant) and air-gap (no egress, with models and identity on your network). The API, the PDP and the audit model do not change.
How is a tenant isolated?
Every row carries tenant_id. There are two layers: a PDP consulted at every boundary (before invoking an agent, before every tool and before reading a credential) and RLS in the database. When the contract needs its own perimeter, the same chart is installed dedicated.
Do secrets reach the model?
No. Secrets are encrypted with a per-tenant KEK, are write-only, and are injected at the transport. They do not enter the prompt, the log or the trace. The model reasons; the vault does not leak.
What evidence does a run leave?
An append-only log: no update and no delete. Every event has an actor. The export can be signed and, optionally, hash-chained. With the semantic layer, every query also leaves a minute tied to the run.
What does the semantic layer do?
It is a governed catalog of terms, metrics and axioms with temporal validity. semantic_ask signs a plan and computes the number deterministically: the language model does not make it up.
How do native agents, A2A and MCP fit together?
Native agents are built on the owned runtime. External agents are imported by reference (A2A) and live on the same host. MCP discovers servers, connects over OAuth and stores credentials in the vault; the tools become available to the agent after the connection is tested.