Argy AI

Argy AI is Argy’s multi-provider LLM governance layer (the LLM Gateway).

It sits between your users/agents and your providers to enforce quotas, security filters, auditability, and routing without changing your client integrations.

Supported providers

Argy AI supports multiple providers (for example: OpenAI, Anthropic, Mistral, xAI, Google, Azure OpenAI). Provider choice can be routed and policy-driven.

Routing and fallback

Argy AI can route requests based on:

• task type (chat, code, OCR, embeddings),
• fallback chains (provider failover),
• sovereignty preferences (forcing specific providers),
• cost-aware choices.

Security filters

Argy AI provides built-in filters such as:

• PII redaction
• secret detection (API keys, passwords, tokens)
• prompt-injection defense
• forbidden topics filtering
• output masking/blocking

Quotas, budgets, and dashboards

You can define token allocations and quotas:

• per plan,
• per organization/team (shared budgets),
• with alert thresholds (e.g., warning/critical).

The billing/usage dashboard can expose real-time consumption, blocked requests, and cost signals (e.g. cost per 1K tokens).

Auditability

Every request can be traced (user, model, tenant), with correlation IDs, retention (minimum 90 days), and CSV exports. Optional AES-256 encryption can be used for request/response content.

OpenAI-compatible API

Argy AI exposes OpenAI-compatible endpoints:

• POST /v1/chat/completions
• POST /v1/embeddings
• POST /v1/agent/steps
• GET /v1/models (dynamic discovery)

Use GET /v1/models to discover available model ids at runtime. Avoid hardcoding model names in clients.

Studio note: the Argy AI action

In Module Studio, one of the available actions is Argy AI: a module-defined AI step (custom prompts + tools) that can orchestrate sub-agents. This is how enterprises build their own governed AI agents inside modules.

Next steps

• LLM Gateway overview: LLM Gateway
• Build modules with Studio: Building Modules
• Governance controls: Policies & Guardrails