Your keys. Your models. Zero token markup.
TensorGate is the single egress point for every model call your agents make. It holds your provider credentials centrally — your developers never see an API key — routes each request to the right model for the task, and meters every token. We never resell inference: you pay your providers directly, at cost.
The gateway between your agents and every model provider.
Agents multiply model spend fast, and without a single control point you can't see it, cap it, or reduce it. TensorGate is that control point — holding your keys and talking directly to EU provider endpoints or to models on your own hardware. In the default customer-operated deployment it runs on your infrastructure: your code and prompts never transit our servers, and what we receive is limited to licensing telemetry — sign-in (SSO), entitlement and configuration refresh, heartbeats, revocation acknowledgements, signed usage records, with usage records dimensioned by organisation, team, user, repository, project, model, provider, skill, task type, time — under your DPA. If you choose the vendor-hosted option, we operate the gateway for you in the EU and say so plainly.
Route, cap, and meter every call.
Routing & cost
Cost routing
Simple tasks go to cheaper models automatically; complex changes get frontier models. Routing rules are yours to tune.
Zero token markup
Your provider contracts, your keys, your prices. TensorGate never sits in your billing path and never resells inference.
Caching and dedup
Prompt and context caching across the whole organisation, so a hundred agents don't pay a hundred times for the same context.
Controls & budgets
Spend limits that degrade gracefully
Per-developer and per-team budgets alert first, then route to cheaper models — they don't kill an agent mid-task at 4pm.
Metered for audit
Every call is attributed to a developer, task, and repository, and exported as signed usage records your auditors can verify.
Keys & sovereignty
Built for European deployment constraints.
Keys developers never see
Credentials are vaulted centrally and injected at the gateway. No API keys on laptops, in dotfiles, or in CI logs.
Direct to EU endpoints
In customer-operated deployments, calls go straight from your gateway to EU-resident provider endpoints — no US intermediary, no reseller in that data path.
On your premises
TensorGate deploys beside your repositories, on-premises or in your VPC, and tolerates disconnection within the platform's offline grace window. Fully air-gapped operation is on the roadmap.
Your own models too
Route to open-weight models served from your own GPUs (any OpenAI-compatible endpoint) alongside commercial providers.
Put a gateway between your agents and your model bill.
Early access is free. Connect your own provider keys and see routing savings on real traffic.