Control whatcrosses.

Telesto checks credentials, company meaning and model policy before an AI request leaves your environment.

Start a private pilot See the boundary
Private deployment CPU-only inference OpenAI-style API
Live boundary Gateway online
Agent request01 / semantic

“The renewal floor stays at 14.2 percent even if procurement pushes.”

Meaning score0.959
Provider requestBlocked
0.735

Fixed synthetic case, scored by the local model

Telesto / incidentsLast 24 hours
Blocked

Pricing meaning

Semantic policy · Procurement · score 0.959

Credential patternRedacted
Renewal floorBlocked
Routine schedulingAllowed

One governed boundary.

Applications, coding agents and supported browser flows use the same governed route. Deterministic rules catch known identifiers. A compact local model catches company-specific meaning. Approved requests continue to an allowed provider.

Follow one request Check the evidence

One request. Four controls.

Authenticate

Confirm the agent key and its model rules.

Inspect

Find credentials, personal data and policy matches.

Understand

Score company-specific meaning inside the controlled environment.

Route

Forward only an approved request to an allowed model.

Small runtime. Hard evidence.

Measured Telesto facts, not category promises.

64passing admin and DLP tests at the launch baseline
4.39mparameters in the frozen semantic encoder
0raw administrator correction texts retained
CPUonly for local semantic classification
3provider families configured through one gateway: OpenAI, Anthropic and Google. Each launch environment is verified with its own credentials before support is claimed.

The browser extension currently adds local pattern protection. Company-specific meaning is handled by the learned gateway and shared dashboard when requests use Telesto.

Choose your boundary.

Use the hosted gateway for a fast start or deploy the same control plane inside your own environment.

£12 / user / month

Business plan from £12 per user

Core plan from £6 per user

Regulated deployments from £20,000 a year

Private deployment and CPU-only semantic inference

Start a private pilot

Direct answers.

Where does Telesto run?

Telesto can run inside the customer's own environment. The gateway, classifier, policy state and audit database stay under that deployment's control.

Is classification local?

Yes. Deterministic scanning and the compact semantic encoder run locally on CPU before an approved request is sent to the selected model provider.

What does the learning loop store?

The correction text is discarded after processing. The current design retains allowed derived adapter state, a hash and version metadata for auditability.

Which providers are supported?

The gateway is configured for OpenAI, Anthropic and Google model families. Each launch environment is verified with its own credentials before support is claimed.

Can one agent use every model?

Only when its virtual key permits that access. Keys can carry model rules, budgets and an auditable identity for the calling application or agent.

Put the boundary first.

Tell us where AI traffic starts today. We will map a private pilot around your environment and approved providers.