Pay AI providers directly. No per-message markup.

Most platforms in this category sell “messages.” A message is whatever they say it is — sometimes an inbound visitor turn, sometimes a model call, sometimes a token bucket. As your volume grows the per-message price doesn’t fall as fast as the underlying token price you’d be paying yourself. You’re not buying compute. You’re buying access through a meter the platform controls.

This is fine when you’re tiny. A few hundred messages a month is a rounding error. The math gets uncomfortable somewhere around the point where your bot starts working — when your widget is converting, your channels are connected, your help center is fully ingested, and your conversation count starts to climb. Suddenly the meter is the dominant line in your support stack. You’re not paying for the platform anymore. You’re paying for somebody else’s margin on tokens you could have bought directly.

There’s a second thing nobody mentions: the meter rarely reflects what the underlying model actually costs. A long reasoning response on a frontier model consumes real tokens. A short factual reply from a small model is essentially free. Per-message pricing flattens that distinction and charges you the same per “message” either way, so the platform makes more on the cheap conversations and you subsidize the expensive ones for them. With BYOK, the cost you see is the cost you actually generated.

NebulaHex offers two ways to pay for the AI side of your bots. Both ship the same product. The difference is where the AI compute bill comes from.

Neither shape is universally better. The default plans are simpler and the right call for most teams starting out. BYOK becomes the better economics once your volume crosses a threshold or your team has reasons to want direct provider billing — cost transparency, compliance, model choice, an existing provider relationship.

Across Anthropic and OpenAI, the BYOK lineup covers the speed-versus-quality range most teams care about. Your workspace picks one default model, and you can change it any time without losing conversation history, knowledge sources, or customizations — the only thing that changes is which model produces the next reply.

Speed and Quality on the cards reflect general positioning within each provider’s lineup, not a head-to-head ranking. The right model is the one that fits your bots’ job. Most teams settle on Haiku or GPT-4o Mini for high-volume workspaces, Sonnet or GPT-4o for general purpose, and Opus or GPT-4.1 for workspaces where quality is the constraint.

The mechanics are short and reversible. Three steps once you’re on any paid plan.

Your key is encrypted before it’s stored and never returned to the dashboard in plaintext after you save it. If you need to rotate it, paste a new key and re-validate — the old one is replaced atomically. If you want to leave BYOK, toggle it off in workspace settings. You stay on your current plan; the AI billing flips back to managed. There’s no migration step, no data movement, no waiting period.

From your customers’ side, the change is invisible — the widget looks the same, the inbox looks the same, the conversation history looks the same. But the underlying economics are now yours.

Most BYOK customers running modest volume see a small monthly AI bill at the provider, paid directly to Anthropic or OpenAI. The platform fee from us stays flat regardless. Both providers let you set spending caps in your account, and we recommend turning that on the moment you connect a key.

BYOK is a feature on every paid plan, not its own tier. On Growth, Pro, Business, or Enterprise, you can flip BYOK on at the workspace level and your bots immediately start signing requests with your provider key. Flip it off, and your bots go back to managed AI using NebulaHex’s models against your plan’s monthly message allowance.

The difference between “managed” and “BYOK” isn’t your subscription — it’s where the AI cost shows up. Managed: in your NebulaHex bill. BYOK: in your Anthropic or OpenAI bill. The platform fee stays the same regardless.

If your team needs Claude Opus on every reply, BYOK lets you do that on a $99 plan instead of a $999 plan. If your message volume is predictable and the managed allowance covers you, leave BYOK off. Both choices are available on every paid plan.

Your API key is the credential that controls your spend with Anthropic or OpenAI. We treat it accordingly. When you save a key, it’s encrypted at rest before it touches our database. The key is never returned to the dashboard in plaintext after you save it — we show you a placeholder confirming one is on file, and you can replace it but you can’t read it back. The validation step on save uses the key for a single low-cost test request and discards the in-memory copy immediately.

When any bot in your workspace generates a reply, the key is decrypted in the request path, used to sign the outbound API call, and dropped — it isn’t logged, isn’t echoed in audit trails, and isn’t included in any export. If a key is ever compromised, you rotate it the same way you’d rotate any cloud credential: revoke the old one in your provider’s dashboard, issue a new one, and paste the new one into NebulaHex. The compromised key stops being able to bill you the moment your provider revokes it.

Workspace permissions matter here. Only owners and admins can change the workspace AI configuration. Editors can update knowledge and bot behavior but can’t see or change which key is in use. Viewers have read-only access. Your data is never used to train AI models on our side or your provider’s side — see our privacy policy and terms for the full posture.

If you’re already a NebulaHex customer, the path is short: open Workspace Settings → AI Configuration, pick your provider, paste your key, validate, choose your default model, save. The next reply any bot in your workspace sends is on your account.