Scutum

What it is

Free for early adopters · Self-hosted · OpenAI-compatible 100+ models behind one OpenAI-compatible endpoint. Per-team budgets enforced before tokens burn. Every query auditable. Postgres as source of truth. Your data never leaves your infra. Per-team budgets enforced pre-request, not after the bill arrives. tiktoken-based prediction refuses calls that would breach budget — before they cost you anything. Built as a LiteLLM hook. Every request logged with team, project, user, model, cost, and latency. Postgres is the single source of truth. Ed25519 offline licensing — no phone-home, no vendor cloud. Air-gapped deployments supported. OpenAI, Anthropic, Google, Meta, Mistral, Cohere, xAI, DeepSeek, plus self-hosted via Ollama and vLLM. Routing in Cedar. Fallbacks, caching, A/B splits. Drop-in replacement for any OpenAI SDK call. Point your existing OpenAI SDK at Scutum. Get cost tracking, audit, routing, and budgets — without touching your application code. Same SDK. Same code. Different endpoint. Now every call is logged, budgeted, and routable.

What to verify before adopting

• Confirm current pricing, usage limits, supported regions, and contract terms on the official website.
• Test integrations, export paths, access controls, and failure recovery with low-risk data first.
• Review privacy, data retention, security, and compliance requirements for your market.
• Keep a migration or rollback plan before making the product part of a critical workflow.