AI credits and usage

AI Assistant uses token-based billing. Every message consumes credits depending on tokens.

Where to find it

/manager/settings/ai-assistant → Credits and Usage tabs.

→ Detailed credit purchase: AI credits

How consumption is calculated

Claude (Anthropic) is token-based:

Input tokens: what you send (question + context + system prompt)
Output tokens: AI's response

Per-million-token pricing (input and output separately). Platform adds small markup.

Typical consumption

Simple question ("Today's revenue?"): ~5-20 HUF
Complex analysis ("Compare 2 quarters"): ~50-200 HUF
Content generation (service description): ~20-50 HUF
Autonomous action (tool-uses): ~50-300 HUF

Credits tab

What you see:

Current balance (HUF)
Packages / custom top-up (purchase)
Transaction history
- PURCHASE
- USAGE (message token usage)
- MANUAL_ADJUSTMENT (admin)
- REFUND

Usage tab

Detailed stats:

Lifetime usage (all-time HUF)
Last 30 days (consumption)
Avg per day
Avg per message
Top users (who consumes most, role-broken)

Activity Logs tab

AI action list with token costs:

Message text (short)
Tool-use count
Input + output tokens
HUF value
Confirmation status

Detailed drill-down: which message is expensive, why.

Low balance

Email warning when below threshold (default ~1000 HUF)
Banner in Bookinda UI
0 HUF balance → AI chat blocked (must buy credits)

Cost optimization

Saving methods

Concise custom instructions: no 500-word system prompt, it's in every request
Tight questions: "Today's revenue?" beats "Could you please tell me what our revenue was today?"
More frequent conversation reset: multi-turn context sends with every message (pricier)
Simpler model (Haiku) if Sonnet isn't needed

Pricier modes

Autonomous actions (more tool-use rounds)
Long context (lots of data being read)
Content generation (many output tokens)

When to use it

Scenario 1, AI intro Buy 5,000 HUF. Use for a month. Usage tab shows 3,000 HUF/mo consumption. Plan package strategy.

Scenario 2, AI spike One day spent 5,000 HUF on AI. Activity Logs → check. Maybe an autonomous marketing campaign was pricier than expected.

Scenario 3, low balance warning Email "200 HUF left." Buy 10,000 HUF immediately, don't let AI stop.

Scenario 4, team consumption analysis Usage → top users. Reception consumes 80%, lots of customer lookups. Custom instruction: "Reception uses Bookinda's built-in search for SQL-like queries, not AI."

Scenario 5, cost reduction Monthly usage too high. Tighten custom instructions, narrow context. Cost down 30%.

Tips

5,000 HUF is a good trial purchase, don't start with 50,000.
Autonomous actions are priciest, start cost reduction there.
Usage tab deserves weekly review, no month-end surprise.
"Low balance" threshold around 1,000-2,000 HUF, AI consumption is dynamic.
Custom instructions length is a real cost factor: in every message.
Multi-turn context is expensive, sometimes new conversation is cheaper.

AI credits and usage stats