Usage & Billing

Every interaction with the Data Science Agent consumes AI tokens. Usage is tracked automatically and visible in your account settings.

How Tokens Work

Each conversation with the agent uses tokens in two directions:

• Input tokens — Your messages, system instructions, tool schemas, and tool results sent to the AI model
• Output tokens — The AI’s responses, SQL queries, Python code, and tool calls

A typical interaction (browse data + run a query) uses approximately 3,000-5,000 tokens total.

Pricing

Costs depend on which AI model you use. Gemini is available by default; OpenAI and Anthropic require your own API key (BYOK).

Example costs (Gemini 2.5 Flash):

• Exploring your datasets: ~$0.001
• Running a data query: ~$0.001-0.002
• Building a full model (browse + generate + preview): ~$0.002-0.005
• A full day of heavy usage (50 interactions): ~$0.05-0.10

Bring Your Own Key (BYOK)

To use OpenAI or Anthropic models, add your API key in Settings > AI Keys:

1. Navigate to Settings > AI Keys
2. Find the provider card (OpenAI or Anthropic)
3. Paste your API key
4. Click Test to validate the key works
5. Click Save Key

Once configured, the model selector dropdown in the agent header will show all available models for that provider. You can update or remove keys at any time.

When using BYOK models, you are billed directly by the provider — Vendo does not add any markup. Token usage is still tracked in Usage Metrics regardless of which model you use.

Exporting Data

Query results can be exported as CSV files:

1. Run a query in the agent
2. The results table will display in the right panel
3. Click the CSV button in the table header
4. The file downloads with the query title as the filename

Viewing Usage

Navigate to Settings > Usage Metrics to see:

Data Section

• Total storage across your BigQuery datasets
• Rows synced
• Per-dataset breakdown (source vs production data)

Compute Section

• Pipeline read/write operations
• Model compute costs
• Total compute usage

AI Agent Section

• Total Tokens — Input and output token counts
• Estimated Cost — Based on Gemini pricing
• Sessions — Number of chat interactions
• Tool Calls — Data queries and actions executed
• Usage by Model — Breakdown by AI model with token counts and costs

What Counts as an Interaction

Each time you send a message and the agent completes its response (including all tool calls), one interaction is logged. A single interaction may include multiple tool calls (browsing datasets, inspecting schemas, running queries) — these are all part of one interaction.

Limits

• Tool calls per interaction — The agent is limited to 15 steps per response
• Query results — Limited to 500 rows per query
• Query timeout — 30 seconds per query execution
• Table previews — Limited to 10 sample rows