Pricing & usage
Referencd uses a single internal token balance. Purchased packages credit that balance; work across drafting, knowledge-base preparation, and document processing draws it down using the rates below.
Token packages
Packages shown reflect what is configured for your deployment—typically seeded as Starter, Pro, and Ultimate tiers. Larger bundles improve the effective price per million internal tokens.
Starter - 250K Tokens
A lightweight plan for casual use — ideal for getting started or low-volume monthly usage.
$8 · 250,000 internal tokens
Effective rate ≈ $32 per 1M internal tokens
- Most popular
Pro - 1 Million Tokens
Built for regular, everyday use — covers a full month of moderate AI-assisted activity.
$29 · 1,000,000 internal tokens
Effective rate ≈ $29 per 1M internal tokens
Ultimate - 3 Million Tokens
For power users who rely on AI throughout their workflow — maximum capacity at the best per-token rate.
$78 · 3,000,000 internal tokens
Effective rate ≈ $26 per 1M internal tokens
How internal tokens map to work
Provider usage (tokens, pages, or tool calls) is converted into internal tokens using the multipliers configured for each model. Your balance is reduced by those internal amounts as you use the product.
| Capability | What you are measuring | Internal tokens |
|---|---|---|
Response drafting openai/gpt-5.4-mini | Draft and refinement work in the assistant, including input you provide and output returned. |
|
Knowledge base preparation openai/text-embedding-3-small | When material is prepared for search and drafting, usage is based on the amount of text processed. | 1.4 / 1K units of provider input |
Document conversion llamaindex/llama-parse-cost-effective | Per page of uploaded files that go through file-to-text conversion. Plain text and Markdown you paste are not charged per page. | 260 internal tokens per processed page |
Bringing documents into a workspace
For many uploaded files, conversion is billed per page at the document-conversion rate in the table. After conversion, preparing that text for search and drafting uses the knowledge-base preparation rate based on how much text is processed—so a long policy adds more usage than a short appendix.Together, these steps replace manual hunting through folders while keeping a clear link between processed material and the drafts your reviewers approve.
Figures above come from your live configuration and may change when plans or provider rates are updated. For contractual pricing or invoices, contact sales.