Skip to content

Meet DeepintShield at Bengaluru Tech Summit 2026, 17 to 19 November, BIEC Hall 1, booth H1A106. Book 20 minutes

Start For Free

Give every team safe LLM access

Put 29 built-in model providers behind one OpenAI-compatible API, with Virtual Keys, model lists, budgets and guardrails per team.

If you need to work out how to give every team safe LLM access, start with one gateway instead of one provider key per project. DeepintShield puts 29 built-in model providers behind one OpenAI-compatible API and gives each team its own Virtual Keys, approved models, budgets, guardrails and logs.

The challenge

Once a few teams ship AI features, a platform team inherits a mess of keys, providers and rules:

  • Shared provider keys. The same key sits in many repositories, so revoking it breaks everything.
  • No model policy. Teams pick any model, including ones the organisation has not approved.
  • Duplicated controls. Each team builds its own filters, retries and logging, all slightly different.
  • No attribution. One bill arrives, and nobody can say which team or product spent what.
  • Slow offboarding. Removing one team’s access means rotating a key that others share.
  • Uneven resilience. Some apps retry and fail over; others simply break when a provider is down.

The approach

Run a central platform with self-service inside clear limits. Connect providers once. Give each application its own credential that carries its permissions. Apply common guardrails by default and let teams add their own. Use your identity provider for console access. Make usage visible by team so budgets and reviews are based on data. Teams keep their existing SDKs and change only the base URL and key. A new team then follows one path: request a Virtual Key, start with the default model list and guardrail baseline, ask for extra models through review, and watch its own usage. Decide early which settings stay central and which teams may change: model lists and guardrail baselines usually stay central, while budgets within a team’s limit can be delegated.

How to give every team safe LLM access with DeepintShield

  1. Connect providers once. Add provider keys to the gateway, with weights for load balancing and deployment mappings for Azure and Bedrock. Teams call provider/model through https://app.deepintshield.com/v1 from their existing OpenAI, Anthropic, Google GenAI, Bedrock, LangChain, LiteLLM or PydanticAI clients. Unknown models are rejected. See the integration quick start and the integrations directory.
  2. Model teams and keys. Create Customers and Teams (Team plan and above). Each Virtual Key belongs to one team or one customer and carries its allowed providers and models, budgets, rate limits and MCP tools. Add required headers, such as a cost-centre header, so requests without them are rejected with HTTP 400 missing_required_headers before reaching a provider. See the Virtual Keys docs.
  3. Apply shared policy and access. Bind guardrail policies to Virtual Keys so every team starts with the same baseline. Console users sign in with SSO over OIDC, with JIT provisioning, SCIM sync with Microsoft Entra ID, MFA and role-based access: Admin, Viewer or custom roles across 14 resources.

What you can show afterwards

  • Usage, cost and latency by Virtual Key, team and customer, in the logs and as Prometheus labels.
  • Which team can call which models and tools, from the key configuration.
  • Guardrail findings per team, with the policy that applied.
  • Every configuration change in the hash-chained audit log.
  • Which applications still call providers directly, by comparing provider usage with gateway traffic.

Before opening the gateway to every team, test one key against each rule: a request without the required header, a call to an unapproved model, a burst over the rate limit and a revoked key. Each should be refused before it reaches a provider.

There is no token markup and no logging surcharge, so the gateway bill does not grow with provider spend. For the design, read about Virtual API keys and workspace access boundaries.

Start For Free