Dual deployment
Two postures. Pick per firm, not as a marketing toggle.
- Approved-model API egress. Jobs call approved models over the network you already allow, through mGateway policy.
- VPC / no-egress. Jobs stay inside the boundary you specify. No path out except what you put on the allow-list.
Identity on every request
Every job and every model or tool call resolves to the person who started it. Agents inherit that identity. They do not receive a service account with broader access than the starter.
Allow-list
The workspace may reach only what you approve. The sources we wire for a research job:
- Bloomberg
- FactSet
- Internal file share
- git
Models and tools sit on the same list. If it is not approved, the job cannot call it.
Audit trail
Who started the job, which model, which sources, what was retrieved, what was written, who reviewed, who published. Head of Tech can reconstruct a run. CCO can ask for the same record.
Memory scope
Workspace memory is not a firm-wide brain by default. Scope is one of:
- User (the default)
- Desk or pod
- An explicit named team
Share is a deliberate act. CCO can refuse a share the way they refuse a model. MNPI, restricted lists, and Chinese walls are the default posture: memory does not cross a wall because someone ran a job.
What we do not claim
We do not publish SOC 2 or other certifications on this page. If you need a current attestation pack, ask in a working session.
Jobs can start from Slack or from the AI chat the firm already allows, including assistants such as ChatGPT, Claude, and Codex. Those tools do not become the system of record. Identity, allow-list, and memory stay in this environment.