One fixed price. Everything included.

Your monthly fee covers everything — AI inference, compute, storage, and networking. There are no per-token charges, no API markups, and no usage metering. Your AI assistants work as hard as they need to without running up a bill.

Yes — Premium and Pro tiers support BYOK (Bring Your Own Key) with any OpenAI-compatible provider, including OpenAI and OpenRouter. Through OpenRouter you can use Claude, Gemini, Grok, and dozens of other models. Your API calls go through our secure egress proxy, and you pay the provider directly. We never mark up API costs.

Your AI assistants keep working with brief added latency under fair-use thresholds. We never hard-block access to the AI or cut off your AI assistants mid-task. No overage charges, ever.

Pro pricing is a single monthly total based on how many AI assistants you need. Use the slider above to see your price. Per-AI assistant cost decreases as you add more — a single AI assistant is $1,097/mo, but at 10 AI assistants it works out to just $339 each.

Absolutely. Every customer gets sandboxed containers, outbound traffic filtering, and encrypted storage. Pro customers additionally get dedicated compute instances, a dedicated network subnet with network ACLs, and dedicated encryption keys.

Yes. Self-service export (JSON/ZIP) of all conversations, configs, and skills is available anytime through the customer portal. After cancellation, we retain your data for 30 days, then permanently delete it. For Pro customers, deletion includes destroying your dedicated encryption key; on Standard and Premium, your data is erased from encrypted storage.

The platform is designed for SOC 2 Type II, ISO 27001, and GDPR readiness from day one. Pro tier customers can review a Data Processing Agreement (DPA) during onboarding, plus compliance audit support. See our Enterprise page for full compliance details.

All service tiers receive six-layer security, encrypted messaging, outbound traffic filtering, and encrypted storage. The Pro tier adds an environment with dedicated compute resources where all of a customer's AI assistants can interact with each other and share a knowledge base common across the business. It includes a dedicated network subnet with network ACLs and dedicated encryption keys, and can optionally add dedicated, single-tenant inference capacity (rolling out; contact sales) so requests never queue behind other customers' work.

Every AI assistant has built-in web search powered by a private proxy that anonymizes your requests and strips scripts and tracking code from results. Your requests carry no customer identity. This is included at every tier at no additional cost.

Frameworks like Hermes Agent make iterative self-improvement the headline feature — the agent modifies its own instructions and tooling as it runs. We deliberately don't, because "self-improvement" is only as well-defined as whoever measures it: an unsupervised change that helps one task can quietly degrade another — and you find out in the middle of your workday. Your AI assistant absolutely gets better over time — it accumulates memory and knowledge as reviewable data you own. But changes to how it behaves are engineered by us, tested in a full staging environment, and rolled out with a rollback path before they ever touch your work. When self-improving frameworks pioneer a genuinely good idea, we adopt it in that controlled, tested form. Reliability is the feature.