Which AI platforms can run inside a bank's own environment?
A platform can run inside a bank's own environment in two ways, on-premise or in the bank's own cloud; a third option, vendor-hosting, keeps it outside. In the first two, client data can remain within the bank's perimeter, depending on the selected deployment, and the choice then shapes who runs the models and the application. Streetbeat for Finance deploys on-premise or in the client's cloud, and its platform orchestrates 4 model providers.
What should a bank be able to show about client data?
Where client data goes, who can read it and what happens to it inside the AI tool. The bank, not the tool, answers for that. Federal Reserve, FDIC and OCC guidance from June 2023 puts it plainly: using a third party "does not diminish or remove" a bank's responsibility for safe and sound operation, including the security of customer information. A deployment the bank can inspect gives its risk team something concrete to review.
Which deployment options exist?
There are three, and each moves a different part of the work to the bank.
| Option | Client data | Models | Operations |
|---|---|---|---|
| On-premise | Stays in the bank's own data centers | Ask whether each model runs locally or through a connection the bank approves | Ask who runs the infrastructure and who supports the application |
| Bank's own cloud | Stays in the bank's cloud account and region | Ask which models that cloud and the bank's contracts allow | Ask who holds access and keys |
| Vendor-hosted | Leaves the bank's environment | Chosen by the vendor | Ask how the bank audits it through contracts and reports |
Streetbeat's security page states the first two as a choice: deploy on-premise or in your cloud, and client data can remain within your perimeter, depending on the selected deployment. The vendor-hosted row is the one that cannot meet a rule of no external cloud processing. The Models column depends on the platform: Streetbeat for Finance works with 4 model providers, covered below.
What should a bank check beyond the model?
Check who runs the application, what it logs and what evidence the vendor can show. A model that answers well is only one part of an approval file.
| Check | Question to ask | Evidence to request |
|---|---|---|
| Application operator | Who deploys, patches and restarts it? | A named runbook and an escalation path |
| Logs and audit trail | Can every AI answer be traced to a user, an input and a source? | A sample log and the retention settings |
| Access | Who can read client data inside the system? | Role definitions and access reviews |
| Security evidence | Has an independent party examined the controls? | A SOC 2 Type II report |
On security evidence, Streetbeat shares its SOC 2 Type II report under NDA, and its security page lists 30+ patent filings, from agent security to tamper-evident audit logs.
Which model providers does the platform use, and where do they run?
Ask for the list per deployment, because the answer depends on the configuration. The 4 providers Streetbeat names on its security page are Vertex AI, OpenAI, Anthropic and Azure. Whether a given provider is reachable from inside a bank's perimeter is a decision for the bank's security team and the selected deployment, not a default.
Streetbeat's security page puts the training question in one sentence: "Streetbeat does not use client data to train shared Streetbeat models, subject to the selected deployment configuration and applicable third-party provider terms." Read both conditions, because the provider terms are a second contract to review.
How can a bank start with a governed pilot?
Keep it to one workflow, one deployment and a fixed length. Streetbeat's standard plan gives the pilot 3 months of live use inside the chosen deployment, then about 8 to 13 weeks to production.
| Phase | Length | What the bank gets |
|---|---|---|
| Live pilot | 3 months | Real usage inside the chosen deployment |
| Move to production | About 8 to 13 weeks | Production rollout of the validated workflow |
| Total | Roughly 5 to 6 months | From pilot to production |
One European bank shows what a pilot can reach: within three months, 85% of its 3,000 advisors were active and 99.64% of interactions ran without errors. That is one institution, and its hosting model is not stated.
What does the bank still decide?
The bank decides everything that makes the deployment its own: the deployment model, the providers it accepts, who gets access, what is logged and when a person reviews an answer. The institution remains responsible for configuration, oversight, suitability and final decisions. The 2023 interagency guidance covers a vendor relationship from planning to termination, and at no stage does it shift the bank's responsibility to the vendor. A vendor can supply controls the bank can verify; it cannot carry that responsibility for the bank.
About Streetbeat
Streetbeat helps organizations use AI to grow revenue, not costs. Streetbeat OS, the AI operating system for companies, is the foundation of Streetbeat for Finance, Streetbeat for Impact and Brain50 by Streetbeat.
Streetbeat for Finance
Streetbeat for Finance brings AI to banks, wealth and asset managers, insurers and real estate investment firms. It includes the AI Suite for Advisors, the Portfolio Engine, the AI Suite for Retail and the AI Suite for Operations, deployed on-premise or in the client's cloud and configured around the institution's workflows, data and controls. See Streetbeat for Finance.