Pricing
To create and use a knowledge base, T tokens are needed. A T token is the pricing unit common for all types of expenses in Tovie Data Agent.
Upon registration, you’ll receive 1,000 T tokens for free. When the tokens run out, purchase a token package to continue utilising the knowledge base. You can check and top up your token balance on the Account page.
The up-to-date pricing information is listed on this page only. Knowledge bases cannot provide precise information on their usage costs.
T token packages
| Cost | Package size, in T tokens |
|---|---|
| $5 | 666.67 |
| $10 | 1333.33 |
| $40 | 5333.33 |
Corporate clients benefit from tiered pricing: larger packages offer greater savings. If interested, contact us at contact@tovie.ai.
T token usage
You can see how T tokens are spent in the Statistics section of the knowledge base project.
Requests to LLM
A knowledge base sends requests of several types to an LLM:
- User requests: to rephrase a user’s question to the knowledge base and to generate a response. Questions in the test chat, in channels, and via API are considered here.
- Agent requests: to retrieve chunks if the chunk retrieval method using LLM is selected in the settings.
- Quality evaluation: to generate test sets and perform evaluations.
- Chunking: to split text to chunks, if chunking using LLM is selected, and for obtaining image descriptions.
Request costs in T tokens depend on:
- The language model selected.
- The length of the request (including context) and response.
| Model | Cost per 1,000 tokens for the model, in T tokens | |
|---|---|---|
| Input ↑ | Output ↓ | |
| DeepSeek-V3 | 0.27 | 1.1 |
| GPT-4o | 2.5 | 10 |
| GPT-4o mini | 0.15 | 0.6 |
| Qwen2.5-32B-Instruct-fp8-dynamic | 0.625 | 0.625 |
| Qwen/Qwen2.5-72B-Instruct-AWQ | 0.875 | 0.875 |
Data storage
The cost of storing source documents is 0.756 T tokens per GB per day (23 T tokens per month).
If your T tokens run out, the files are stored for another 30 days and then deleted.
Vectorisation
Vectorisation is the conversion of text into numerical vectors. It applies to both your data at the indexing stage and to user queries.
The cost depends on the vectoriser model selected:
text-embedding-3-large: 0.13 T tokens for 1,000 model tokens.intfloat/multilingual-e5-large: 0.1 T tokens for 1,000 model tokens.