AI Brain operations are billed from your Vlozi credit balance. Credits are shared across all Vlozi services.
Two types of charges
1. Chat credit charge
Every owner copilot and customer chat message consumes credits based on the number of tokens processed.
Credits are charged after the response is delivered — you are only billed for what was actually generated.
The charge is not fixed per message. Longer messages, longer responses, and higher model tiers cost more. The system estimates the maximum cost before processing and checks your balance. If your balance is insufficient, you receive a 402 error and the message is not processed.
Rough cost guide (approximate):
| Message type | Typical credit range | Notes |
|---|---|---|
| Short question (< 50 words) | 0.02 – 0.05 cr | Haiku-tier fast response |
| Normal chat (50–200 words) | 0.1 – 0.3 cr | Sonnet-tier balanced response |
| Long analysis (200+ words) | 0.3 – 1.0 cr | Sonnet or Opus, large context |
| Knowledge base chat (customer) | 0.05 – 0.2 cr | Sonnet + retrieved knowledge chunks |
These are estimates. Actual cost depends on your input length, conversation history length, and the model selected by the router.
2. Knowledge embed charge
Uploading a document to the knowledge base embeds each chunk with a vector embedding model. You are charged per chunk created.
The upload process:
- Vlozi holds the estimated maximum credit cost before processing (based on document size)
- The document is chunked and embedded
- Only the actual number of chunks created is charged
- Any over-held amount is refunded automatically
Typical costs:
| Document | Chunks (estimate) | Credit cost (estimate) |
|---|---|---|
| Short FAQ (1 KB) | 2–4 chunks | 0.02 – 0.04 cr |
| Product doc (10 KB) | 15–25 chunks | 0.15 – 0.25 cr |
| Policy document (50 KB) | 70–100 chunks | 0.70 – 1.0 cr |
One chunk is approximately 500–1,000 characters of text.
Plan limits
In addition to credits, AI Brain enforces monthly interaction limits per plan:
| Plan | Monthly interactions | Models available | Notes |
|---|---|---|---|
| Free | 50 | Haiku-class | Very limited; for evaluation |
| Starter | 500 | Haiku + Sonnet | Small teams |
| Pro | 5,000 | Haiku + Sonnet + Opus | Growing businesses |
| Business | Unlimited | All models | Enterprise workloads |
An interaction is one message exchange (user message + AI response). When your monthly interaction limit is reached, new chat messages return a 402 until the next billing period begins.
Your current usage is visible at AI Brain → Settings → Usage.
When you run out
If your credit balance reaches zero mid-conversation:
- The current in-progress message is delivered (billing failure does not abort the response)
- The next message returns 402 — top up credits to continue
If your monthly interaction limit is reached:
- New messages return 402 until the billing period resets (first of the month, UTC)
- Interaction limits reset automatically — you don't need to do anything
Checking usage
In the dashboard → AI Brain → Settings → Usage, or via the API:
GET /brain/usage?period=2026-06{
"period": "2026-06",
"plan": "pro",
"limits": {
"max_interactions": 5000,
"used_interactions": 342,
"remaining": 4658
},
"tokens": {
"input": 856000,
"output": 68400,
"total": 924400
},
"cost_usd": 4.72
}Tips for managing costs
- Use the copilot for substantive tasks — casual test messages still consume credits
- Pass only necessary conversation history — very long history windows increase input tokens
- Keep knowledge documents concise — shorter, focused documents produce fewer chunks
- Customer chat is cheaper than owner chat — it doesn't use tools or multi-agent orchestration