Documentation Index
Fetch the complete documentation index at: https://learn.getodin.ai/llms.txt
Use this file to discover all available pages before exploring further.
Odin AI usage is billed in credits, a unified currency that abstracts the underlying costs of running AI-powered workflows. Credits cover two types of activity: flat-rate Platform Action Credits for discrete operations like uploading documents, sending messages, and invoking tools; and variable LLM Token Credits based on how much text is processed and generated by the language model. Together, these two components make up the total cost of any operation in Odin AI.
This page breaks down how each credit type is calculated, walks through real examples, and explains what factors drive usage up or down.
Core Concepts
What Are Credits?
Credits are the unit of measurement Odin AI uses to track and bill for platform usage. Rather than billing separately for every underlying resource (compute, API calls, LLM usage), credits provide a single, unified currency that abstracts those costs into a predictable format.
Credits are consumed in two ways:
- Platform Action Credits are flat-rate charges tied to specific user actions — uploading a document, sending a chat message, or invoking a tool. These are fixed and predictable regardless of content size or model used.
- LLM Token Credits are variable charges based on actual language model usage. Because LLM costs depend on how much text goes in and comes out of the model, these credits fluctuate based on document size, response length, and model selection.
Your total credit consumption for any operation is always the sum of both types.
What Are Tokens?
Tokens are the unit that Large Language Models use to process text. Before any text is read or generated by an LLM, it is broken down into tokens — small fragments that roughly correspond to words or parts of words.
As a general rule of thumb:
- 1 token ≈ 4 characters, or roughly ¾ of a word in English
- A 100,000-word document is approximately 133,000 tokens
- A short sentence like “Explain this document” is around 4–5 tokens
LLM providers charge separately for input tokens (the text sent to the model — your prompt, retrieved document chunks, system instructions) and output tokens (the text the model generates in response). Input and output rates differ, and both contribute to your LLM Token Credit consumption.
In Odin AI, token usage grows when large documents are retrieved into context, many knowledge base chunks are included in a prompt, or the model produces long or structured responses.
Credit Consumption Table
A. Knowledge Base (KB) Ingestion
Uploading a document to a Knowledge Base triggers a multi-stage ingestion pipeline. Each stage has its own cost structure, and some stages incur additional LLM Token Credits depending on your project configuration.
Ingestion Pipeline
| Stage | Description | Cost |
|---|
| 1. Document Upload | File is received and word count is calculated | 1 credit per 10,000 words |
| 2. LLM Extraction (optional) | An LLM parses and extracts content from the document | Input + output tokens × model rate |
| 3. Platform Chunking | Tokenization, splitting, and metadata assembly | Free (platform) |
| 4. Chunk Enrichment (optional) | An LLM generates a context prefix for each chunk | Input + output tokens × model rate |
| Item | Credits |
|---|
| 100,000 words | 10 credits |
| Item | Credits |
|---|
| 100,000 words (upload) | 10 credits |
| LLM Extraction tokens | Varies by model |
| Platform Chunking | Free |
| Chunk Enrichment tokens | ~10 credits (see Section B) |
| Estimated Total | ~20+ credits |
The more optional stages you enable, the higher the per-document ingestion cost. Word-based upload cost is always fixed and predictable.
B. Document Chunking
Chunking is the step between document extraction and embedding. It takes the cleaned text from an uploaded document and breaks it into smaller pieces (chunks) that are indexed in the vector store and later retrieved for Chat or Agent queries.
Chunking costs fall into two categories:
- Platform Chunking: Deterministic operations (tokenization, splitting, metadata assembly) that run locally on the server and are not billed as LLM usage.
- Chunk Enrichment (LLM Token Credits): Billed only when Chunk Enrichment is enabled for the project in Knowledge Base Settings. The LLM generates a short context prefix for each chunk; both input and output tokens are charged.
Example: Document Chunking Costs
The following example assumes a moderately sized document with Chunk Enrichment enabled:
- Document size: 100,000 words (~133,000 input tokens, at 4 chars/token)
- Chunk settings:
chunk_size=64 tokens, chunk_overlap=10 tokens → ~2,300 chunks
- Chunk Enrichment: Enabled with
gpt-4o-mini
- Output text (joined chunks incl. enrichment prefixes): ~140,000 output tokens
- Model pricing (example): input
$0.15 per 1M tokens, output $0.60 per 1M tokens
| Component | Tokens | Cost |
|---|
| Input Tokens (document) | 133,000 | $0.01995 |
| Output Tokens (chunks) | 140,000 | $0.084 |
| Total LLM Cost | 273,000 | $0.104 |
$0.104 (total LLM cost) / $0.01 (cost per credit) equals to 10.4 credits, displayed as ~10 credits.
If the same document were uploaded without Chunk Enrichment enabled, chunking would consume 0 LLM Token Credits.
C. Chat / Agent Interaction
Fixed platform credits
| Action | Credits |
|---|
| User sends a chat message | 1 credit |
| Tool call invoked | 1 per call |
-
Input: $3 per 1M tokens
-
Output: $15 per 1M tokens
D. Workflow Executions
| Action | Credits |
|---|
| Workflow execution | 1 per execution |
The cost is 1 credit per execution, regardless of the number of steps involved.
Full Chat Example
Scenario
User asks: “Explain the attached document.”
Platform credits
| Component | Credits |
|---|
| Question asked | 1 |
| Tool calls (document retrieval) | 2 |
| Subtotal | 3 |
| Type | Tokens | Cost |
|---|
| Input | ~53,634 | ~$0.161 |
| Output | ~900 | ~$0.0135 |
| Total | — | ~$0.1745 |
| Component | Credits |
|---|
| Fixed platform | 3 |
| LLM usage | 17 |
| Total | 20 |
What Drives LLM Credit Usage?
LLM cost increases when:
▶ Large documents are retrieved into context
▶ Many KB chunks are injected into the prompt
▶ Responses are long or structured
▶ Multiple tool calls are triggered
▶ Higher-cost models are selected
Important Notes for Customers
✓ Word-based ingestion cost is predictable.
✓ Chat costs vary significantly depending on document size and the number of tokens used.
✓ Model pricing is configurable in Super Admin.
✓ LLM credits are consumption-based and cannot be flat-rated.
✓ Final credit total = Platform Credits + LLM Credits.
