AI writing and translation: usage and pricing
The AI features come with the editor license. You never pay us per request. Your AI provider bills the tokens your users consume, at that providerâs own rates. This page shows where the cost comes from, how to estimate it, and the controls that keep it predictable.
Who pays for what
| Cost | Who bills it | How it is charged |
|---|---|---|
| Editor license with the AI Toolkit | Richscripts | Once per license, from $129 (perpetual, one year of updates included). Same AI features in every tier. |
| AI model usage | Your AI provider (OpenAI, Anthropic, Azure OpenAI or any OpenAI-compatible service) | Per token, at the providerâs price for the model you choose. Billed to your account, or to each tenantâs account if you use BYOK. |
| Self-hosted model (for example Ollama) | Nobody, per request | No per-token fee. You pay for the hardware that runs the model. |
| Demo resolver | Nobody | Runs in the browser with fixed, predictable output. Use it for evaluation and UI tests, not for real writing help. |
Requests go from the editor to your server endpoint, and from there to the provider. They do not pass through Richscripts. See security and procurement for the full data flow.
What uses tokens
Ask AI has seven one-click modes: proofread, rewrite, shorten, expand, summarize, translate and justify. AI Chat handles free-form requests. Every request sends the selected text (or the whole document, if the writer picks document scope) plus a short instruction, and gets the result back. The provider counts both directions:
- Input tokens: the instruction and the text the writer selected. Choosing âwhole documentâ instead of a selection is the biggest single cost multiplier.
- Output tokens: the returned text. Proofread, rewrite and translate return roughly as much text as they receive. Expand returns more. Shorten and summarize return less.
- Chat history: in AI Chat, earlier turns of the conversation are part of the context, so long chats cost more per message than short ones.
Translation: output is about as long as the input, but token counts per word vary by language. Languages that use non-Latin scripts often need more tokens for the same meaning. Check real counts in the cost ledger (below) before budgeting a multilingual rollout.
Estimate your monthly cost
The arithmetic is the same for every provider:
monthly cost â requests per month
Ă (average input tokens Ă input price per token
+ average output tokens Ă output price per token)Count requests. Active writers Ă AI requests per writer per day Ă working days. Measure a pilot group; donât guess.
Measure tokens. Turn on the cost ledger during the pilot. Each AI call is recorded with its input, output and total tokens, so you get real averages for your content and languages.
Apply your providerâs price. Look up the current per-token price for your model on the providerâs pricing page. Prices change often and differ a lot between models, so this page doesnât quote any.
Add a margin, then cap it. Set the per-request output limit and the rate limit (below), so a runaway client canât exceed your estimate.
Worked example with made-up prices. 50 writers Ă 10 requests a day Ă 21 days = 10,500 requests. If an average request is 600 input tokens and 400 output tokens, and your model costs $1 per million input tokens and $4 per million output tokens, the month costs about 10,500 Ă (600 Ă $0.000001 + 400 Ă $0.000004) = $23.10. Replace every number with your own measurements and your providerâs current prices.
Controls that keep cost predictable
Output limit per request
The server-side resolvers send MaxTokens with every request (default 1024), so the provider stops generating at that point. Lower it if your writers mostly work on short selections.
Rate limit
The ASP.NET Core integration rate-limits the AI endpoints per IP address, by default 30 calls per minute. You can change the limit or use a Redis-backed limiter. Production operations
Per-call cost ledger
Implement IRichTextBoxAiCostSink to receive one AiUsageRecord per AI call: provider, model, mode, input, output and total tokens, latency and key id. Send these records to your own billing or reporting system.
Bring your own key (BYOK)
In a multi-tenant app, each request can use the tenantâs own provider key, so the provider bills that tenant directly, not you. BYOK key vault
Choose the model
Set the model in the resolver options. Smaller models usually cost much less and are often good enough for proofreading. You can use a different resolver per editor or per tenant.
Turn AI off where it isnât needed
Set aiToolkitAutoToolbarButtons: false and add the AI button only to the editors and user roles that need it. AI Toolkit docs
Running with no per-token cost
If data must stay on your network, or you want a fixed cost, point the OpenAI-compatible resolver at a model you host yourself, for example with Ollama. Set BaseUrl to your server, and nothing leaves your infrastructure. Quality and speed depend on the model and the hardware you run it on, so test it with your own documents. Providers and streaming has the configuration.
Questions
Do I pay Richscripts per AI request?
No. The AI Toolkit is part of every editor license. Requests go from your server to the AI provider you configure, and the provider bills you (or your tenant) for the tokens used. Richscripts does not proxy, meter or resell AI traffic.
Is there an AI add-on or credit pack to buy?
No. All tiers, from the $129 domain license up, include Ask AI, AI Chat and AI Review. There is no premium AI tier and no credits to buy.
Can I try the AI features without an API key?
Yes. With no resolver configured the toolkit uses a built-in demo resolver that runs in the browser with predictable output and no network calls. It shows the interface, not real model quality.
Which providers can I use?
OpenAI, Anthropic and Azure OpenAI have built-in resolvers. Any OpenAI-compatible endpoint, such as a self-hosted Ollama server, works through the BaseUrl setting. You can also write your own resolver for any other service.
Does translation cost more than proofreading?
Not by itself. Every mode is billed by tokens in and tokens out. Translation and expand produce about as much text as they read, or more. Shorten and summarize produce less. Long documents cost more than short selections whichever mode you use.
Can each customer of my SaaS pay for their own AI usage?
Yes. With the BYOK key vault, each request uses the tenant's own provider key, so the provider bills that tenant directly. The cost ledger records which key was used for every call.
See it with your own content
Download the trial, connect your provider key on your server, and turn on the cost ledger for a week.