Security

Know what stays local, what leaves and what is still disabled.

Security starts with a readable data path. The tables below describe the current builds, including the places where a connector or chosen model provider can receive data.

Current product boundaries

2 products, 2 explicit paths.

Global investment research

Global Markets Research

In development

Local record: Research cases, documents, evidence, calculations, conflicts and settings.
External paths: Selected market sources and a user-chosen model provider when that provider is used.
Write boundary: No brokerage connection and no trade execution path.
Release boundary: R8 continuity complete. There is no supported public release yet.

Multi-channel commerce

Open Commerce Workbench

In development

Local record: Source snapshots, mappings, ledgers, profit calculations, exceptions and audit records.
External paths: Read-only marketplace requests and a user-chosen model provider when that provider is used.
Write boundary: Remote marketplace writes remain disabled.
Release boundary: P24 technical release candidate. A signed and notarized public desktop release is not verified.

Shared controls

The rules beneath every current product

Local working record

Cases, ledgers, evidence, settings and audit records are designed to live on the operator's machine rather than in a Phoenix Tree AI customer-data cloud.

Narrow connections

A connector receives the least permission needed for its current job. Read access comes before any separately reviewed write path.

Evidence retained

Source records and deterministic transformations stay attached to the result so a person can inspect the path without trusting the model's wording.

External writes gated

A remote action needs its own permission, approval and audit design. The products do not expose a remote write tool today.

Questions

Short answers to the important ones

Does local-first mean nothing leaves the machine?

No. A marketplace connector must contact that marketplace, a market-data path must contact its source and an external model provider receives the request sent to it. Local-first describes who holds the working record. Each external path still needs its own terms and permission review.

Does Phoenix Tree AI train models on customer data?

Phoenix Tree AI does not train models. When a user chooses an external model provider, that provider's own retention and training terms apply.

Can the commerce agent change a marketplace account?

No. The current commerce workbench uses read-only paths, and remote marketplace writes remain disabled.

Can the finance product place a trade?

No. The current portfolio screens are research views, not a brokerage connection or an execution path.

Who can access a system during custom work?

Only access explicitly agreed for the scoped build should be granted. Phoenix Tree AI is operated by one person and has no support team or shared support login.

Current limits

No certification claims.

Phoenix Tree AI is operated by one person. There is no SOC 2 report, penetration test on file or security team on call. Product code is intended to become public at release, but no current product has been published yet.

If procurement requires a certification, Phoenix Tree AI is not the right supplier today. If the first requirement is a precise account of the current data and permission path, ask directly.

Ask a security question Read the build method