Skip to content

🤖 Explain with AI

Private AI Gateway

Every model call made by an AI Studio pod goes through the 4Geeks Private AI Gateway. It gives the platform one place to authenticate, route, meter and audit model traffic, and it keeps the inference provider out of the customer’s view.

In summary

For: security, platform and procurement teams reviewing AI Studio.
Time: 5 minutes.
Result: a clear picture of how model traffic is routed, metered and protected.

For
Security and platform reviewers
Time
5 minutes
Level
Intermediate
Result
Model traffic you can audit

What the gateway does

Responsibility How it works
One interface The gateway exposes an OpenAI-compatible endpoint, so the planner, the sandbox and the agents use the same contract.
Model catalog The gateway publishes the models the workspace can use and the platform maps them to the 4Geeks Fast/Strong tiers.
Routing Each session runs the model the platform selects; the console never shows a vendor or a raw model id.
Metering Every request records input and output tokens, latency and cost for the usage ledger.
Budgeting The gateway enforces the Console service-account budget; AI Studio enforces the pod and wallet budgets.
Auditing Requests are attributable to a pod, a task and a session, so support can explain a cost.

Why it matters

One place to change models

The planner and the sandbox both read the same service-account key and the same model catalog. Changing the model is a platform decision, not a code change in every pod.

The provider is an implementation detail

The inference provider is not part of the customer experience. The timeline, the team card, the error messages and the model labels show 4Geeks Fast or 4Geeks Strong, never a provider name or a raw model id. This keeps the contract stable when the platform changes models.

Costs are explainable

A token is not a customer quota. The gateway records token usage so the platform can price compute and support a pod, while the customer pays for tasks and pod lead time. The usage ledger keeps provider cost and client price separate.

Budgets are enforced twice

A runaway pod is capped by the AI Studio budget and by the Console service-account budget behind the gateway. If one limit is reached, the pod pauses or asks for a top-up instead of continuing to spend.

Security model

  • Credentials stay server-side. The model key is a service-account key scoped to model requests. The sandbox never receives provider credentials.
  • The sandbox is isolated. Each pod runs in its own machine, clones the client repository, and talks to the platform through signed outbound calls.
  • Repository secrets live in the vault. The GitLab/GitHub token and tool credentials are read only by the orchestrator and injected into the sandbox when needed.
  • No provider names in customer-facing text. A branding layer redacts vendor names from events, errors and labels, including rows stored before the rule existed.
  • Customer data is not training data. The gateway is configured to use providers and endpoints that do not retain customer code for training. Ask your account team for the current data-processing terms.

What the gateway does not do

  • It is not a general-purpose LLM proxy for customers. It exists to serve AI Studio pods.
  • It does not deploy code, run migrations or merge pull requests. Those are pod and repository actions.
  • It does not replace your CI. The pod runs its own checks as evidence; your pipeline still runs on the merge request.

Best practices

  1. Treat the gateway as infrastructure. Do not put provider keys in a repository or in a task prompt.
  2. Use the platform model tiers. They are chosen for the task and kept consistent across pods.
  3. Watch the spend, not the token count. Tokens explain cost; tasks and pod lead hours explain the invoice.
  4. Export the timeline for audits. It links usage to tasks, approvals and deliverables.

Frequently asked questions

Which models does AI Studio use?

The platform runs the models exposed by its private gateway and maps them to 4Geeks Fast and 4Geeks Strong. The exact catalog can change without changing your workflow.

Can I bring my own model key?

AI Studio is a managed delivery service. The model key is managed by the platform; bringing your own key is not part of the standard product.

Do you store my prompts?

The gateway records usage metadata and the pod timeline records events. Provider retention is governed by the platform’s data-processing terms; ask your account team for the current details.

Next steps

Still have questions? Ask on Discord or explore the tutorials.