Pricing

Simple plans. Enterprise-ready.

Pricing is tailored to your model traffic and deployment. Tell us about your setup and we'll put together a plan.

Starter

The full AI-FW platform for small teams getting started.

Free

100,000,000 tokens per month, no credit card required.

  • Prompt & response guardrails
  • Semantic intent analysis (all 3 engines)
  • Model registry with per-model keys
  • Conditional routing rules + resilience
  • Completion cache (exact + semantic)
  • Risk profiles & auto-block
  • Metadata-only audit log
  • OIDC SSO & SCIM provisioning
  • Email support
Get started
Most popular

Enterprise

The full platform plus enterprise support for production fleets.

Custom

Custom token allowance, pricing is tailored to your deployment.

  • Everything in Starter, with higher token allowances
  • mTLS & Kerberos agent authentication
  • M365 Copilot admin & agent sync
  • Log export: pull API + syslog
  • Enterprise support: SLA, named support engineer
  • Dedicated deployment & onboarding
Engage sales

Validate it yourself with our Technical Plan

Don't take our word for it. Ask us for the detailed Technical Plan: a step-by-step guide to running your own proof of concept and a full comparison against any vendor you're evaluating, in your own environment, so you can validate that AI-FW truly delivers and is far ahead of the competition.

Frequently asked questions

AI-FW runs in your own environment, on-premises or in your cloud, so your traffic never needs to leave your network. The gateway is a single container service backed by PostgreSQL.

Yes. Any OpenAI-compatible or Anthropic-compatible backend, hosted providers like OpenAI, DeepSeek, Anthropic, Mistral, Groq, or OpenRouter, plus self-hosted engines like Ollama, vLLM, or LM Studio.

No. AI-FW is metadata-only: it logs models, identities, latency, and decisions with truncated previews. Raw prompt and response content is never persisted.

AI-FW is fail-closed by design, a scanner error or missing configuration blocks the request rather than forwarding it unfiltered. Resilience policies add retries, failover, and distribution per model.

No. Point any OpenAI-compatible or Anthropic-compatible client at the gateway, most integrations are a base-URL change.