Integrations: AI model providers

Argent with Azure OpenAI and Foundry

Make Argent the way your Azure workloads reach Azure OpenAI and Foundry models. Argent resolves each deployment to the model it really runs, swaps banned ones to an approved class, redacts, budgets and signs every call.

Azure OpenAI logo
01 / INTEGRATIONS: AI MODEL PROVIDERS

What Argent does with Azure OpenAI and Foundry

Argent governs both Azure route styles and judges each call by the model that will really run, not by the name a deployment was given.

  • Both routes: the classic route, which names a deployment in the path, and the v1 route, which names the model in the body. Both are tested end to end.
  • Deployment names resolved: Azure lets you name a deployment anything. Argent maps each deployment to the model it runs, first from your mapping on the endpoint, then, for endpoints that sign in with Entra, from the Azure management API. A deployment named gpt-4o that really runs another model is judged as that model. Under enforce, a deployment no source maps is refused.
  • Banned deployment, approved model: a Model Swap rule sends a banned deployment to an approved model class, and the caller does not change. In watch mode the call passes and is logged.
  • No API key at all: Argent can sign in to Azure with an Entra client credential, so no API key sits anywhere in the upstream setup.
  • Data and money: sensitive data such as US Social Security numbers redacted on Azure traffic, and agent budgets enforced.
02 / INTEGRATIONS: AI MODEL PROVIDERS

How it connects

One-time setup in your tenant, then two settings per workload.

  1. Add the Foundry resource as an endpoint on the AI Providers page, with its key or an Entra client credential. The secret goes to your vault; the agent never holds it.
  2. Map deployments to models for every deployment a rule names or a swap can send traffic to.
  3. Create a model class holding the approved model, and a swap rule to it.
  4. Register each workload as an agent with a human owner. It gets an Argent key.
  5. Point the workload at Argent. An Azure Function App moved behind Argent with two app settings, AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_API_KEY, and no code deploy. In production the Argent key comes from a Key Vault reference.
  6. Watch, then enforce once the record shows what would have been refused.
03 / INTEGRATIONS: AI MODEL PROVIDERS

Close the side door with Azure Policy

Argent decides, swaps, redacts and records. Azure holds the force.

Two Azure Policy settings and one private endpoint make Argent the only way in: public network access off, key (local) authentication off, and a private endpoint for the AI Services account in the network that holds the Argent gateway. With all three in place, a workload that tries to skip Argent is refused by Azure itself. Argent ships sample policy assignments for both. See the Azure coverage page and Microsoft Entra ID for directory sync.

04 / INTEGRATIONS: AI MODEL PROVIDERS

What you configure

Everything below is set once in the Argent console and applies to every agent that calls Azure OpenAI through the gateway.

SettingWhereWhat it does
Approved modelsPolicy > Model and endpoint accessDefault deny: a call is allowed only when the model is on the approved list. The list is platform AND tenant AND agent, and each layer can only narrow the one above.
Access rulesSame pageAllow or deny by provider, route type, model, hosted tools, function tools and output type, for the tenant, a department, a person or one agent. The narrowest rule wins, and a 7-day impact preview shows what a draft would change.
Model SwapModel Swap pickerRuns an approved model in place of the one the agent asked for, including when a budget runs low, with no change to agent code.
Model classesPolicy > Model classesA name such as fast stands for an ordered list of approved models, with failover on connection errors, 5xx, 429, timeouts and retired models.
Sensitive dataDLPPer entity type: redact and record, record only, or block the request.
Budgets and capsBudgetsBudgets for the company, org, department, person and agent. Each request is priced and held against the budget before it leaves, and an emergency cap stops a runaway agent now.
ModePer controlOff, watch or enforce. Watch records what would have been refused and changes nothing on the wire.
Deployment mapThe endpoint on AI ProvidersWhich model each deployment runs, so rules judge the real model.

Start in watch, read what would have been refused, then switch on enforce. See Watch Before You Enforce.

05 / INTEGRATIONS: AI MODEL PROVIDERS

One request, step by step

What happens between your agent and Azure OpenAI, in order.

  1. The agent sends its request to its Argent route, carrying its own Argent key and nothing from Azure OpenAI.
  2. Who is calling. Argent identifies the agent from its key and checks it may use this endpoint. The endpoint comes from the route the gateway matched, never from a header the caller sets.
  3. Which model. Argent reads the model from the deployment in the path or from the body, then resolves the deployment to the model it runs, and checks it against the approved list and access rules before anything leaves.
  4. Swap or class. If a Model Swap rule or a model class applies, an approved model runs instead, and Argent translates the request and reply between provider formats where needed.
  5. Sensitive data. The request is scanned by a reader that knows the Azure OpenAI format, and secrets and personal data are redacted, recorded or blocked under your policy. If the scan cannot run, the request is refused, never passed as clean.
  6. Budget. The request is priced and the amount is held against the agent's budget, so parallel calls cannot overshoot it.
  7. Credential. The gateway attaches the Foundry key from your vault, or a token it minted with the Entra client credential.
  8. Reply and record. Token usage is read from the reply, the cost is settled, and the signed record is written.
06 / INTEGRATIONS: AI MODEL PROVIDERS

What the record shows

Every Azure OpenAI call leaves one signed record, allowed or refused.

  • One row per call: the agent and the person who owns it, the tenant, the model, the decision and the policy that made it, latency, and the cost, token and cache figures.
  • Tamper-evident: each row is hash-chained to the one before and signed with an ed25519 key held in Vault Transit, so an edited or missing row shows.
  • Real cost: cost comes from the usage block in Azure OpenAI's own reply, priced from your price book in USD per million tokens.
  • No unpriced spend: an agent under a budget that calls a model with no price is refused, so a new model cannot slip past a cap.

Conversations can be opened and replayed from the record. See Session Replay and One-Click Audit Evidence.

07 / INTEGRATIONS: AI MODEL PROVIDERS

Related capabilities and goals

More AI model providers

Every provider below is governed through Argent.

  • OpenAI
  • Anthropic
  • Google Gemini
  • Amazon Bedrock
  • Mistral
  • DeepSeek
  • OpenRouter
  • GitHub Copilot
  • Kiro
  • Ollama
  • vLLM
  • NVIDIA NIM
  • LM Studio
  • Hugging Face TGI