Spend controls for personal assistants

# Spend controls for personal assistants

> Each user sets which merchants their assistant can use, how much, and when to ask.

- Human URL: https://conto.finance/solutions/personal-assistants
- Raw Markdown: https://conto.finance/solutions/personal-assistants.md
- Terminal view: https://conto.finance/ai/solutions/personal-assistants

## Proof Points

- **Best for:** Assistants in iMessage, WhatsApp, and Telegram
- **Controls:** Merchant rules, spend caps, approval thresholds, per user
- **Outcome:** Users delegate spending because they set the boundary

## Workflow

### 1. Your user sets the rules

Merchants, spend limits, and the amount that needs their approval first.

### 2. Conto checks every attempt

Merchant, amount, and category, weighed against that user's rules in the payment path.

### 3. It clears, asks, or stops

Routine spending settles. Anything over the threshold goes back to them in one tap.

## Controls

### Spend limits, per purchase and rolling

A ceiling on any single purchase, plus daily, weekly, and monthly budgets. A runaway loop or a misread instruction hits the cap, not the account balance.

### Merchant and category rules

The user decides where the assistant can spend. Groceries and rides can be routine while gift cards, transfers, and unknown merchants stay off the table entirely.

### Approval thresholds, and where the question lands

The user sets the amount that needs their approval. Above it, a signed one-time approve-or-decline link goes to Telegram, WhatsApp, Slack, email, or your own webhook, landing in a thread they already have open.

### Guardrails the assistant cannot talk its way past

Velocity limits, repeat-charge detection, and hard stops sit under the user-facing dials. They are enforced in the payment path, so a prompt injection that reaches the model still meets the same wall.

## Outcomes

- Users hand an assistant real spending power, because they set the limits themselves instead of trusting yours.
- You ship granular spend controls without building a policy engine, a per-user limits ledger, or an approval system.
- A compromised prompt cannot become an unbounded charge, because the rules are checked at the payment rather than in the model.
- Support and trust teams can show which purchase happened, which rule allowed it, and who approved it.

## Related Industries

- [commerce](https://conto.finance/industries/commerce)
- [travel-events](https://conto.finance/industries/travel-events)
- [delivery](https://conto.finance/industries/delivery)
- [hospitality](https://conto.finance/industries/hospitality)