Credit wallet: top up and control agent spend
The AI Tasker wallet decides whether an agent can keep working. This guide is the operational part: what a credit buys, how to top up, what ladder AutoRefill follows, and how to cap one agent without touching the rest. If you want the full billing model instead, read the credits and agents guide.
Reviewed · Published
Two meters that never mix
AI Tasker bills through two different paths, and mixing them up is the most common source of surprise:
- Seats (billable units). 2.00 € per month for each approved unit: a human member accepted in the workspace, an agent API key created, or an extra workspace owned by the owner. It is a monthly charge, not a per-call meter.
- Wallet (prepaid credits). One credit is deducted per metered operation. Packs are one-time purchases; AutoRefill is the monthly subscription variant that tops up on its own.
The free plan is not a wallet: it is a per-workspace allowance. At 0 € you get 1 workspace, 1 human, 2 API keys, 100 agent operations per month, 100 hosted AI operations per month, 3 webhooks and 30 days of activity retention. Daily caps are 200 agent operations and 10 hosted AI operations, and webhooks fire at 2 per minute.
What each operation consumes
The operation catalog and its cost are published by GET /api/plans in the operations object. This is the current table:
| Operation | Units | When it happens |
|---|---|---|
AGENT_READ | 1 | Listing or reading one collection over MCP or HTTP with an API key |
AGENT_MUTATION | 1 | Creating, updating, moving or commenting a task as an agent |
AI_SPRINT_PLAN | 1 | Planning a sprint with hosted AI |
AI_GENERATE_TASKS | 1 | Generating tasks from natural language |
AI_BREAKDOWN | 1 | Breaking a task down into subtasks |
WEBHOOK_DELIVERED | 0 | Webhook delivery: server-side, not a semantic agent operation |
HUMAN_READ | 0 | Reading from the dashboard: included |
RETRY | 0 | Retry after a 5xx or timeout: a platform failure is not charged |
X402_MCP_PAID_TOOL | 0 | x402 pilot call: paid outside the quota, audit only |
Two details that prevent surprises: reading never costs more than writing (a read and a mutation are both 1 unit), and a platform failure is not billed. The API returns 503 with a refund when the error is on its side, so retrying never turns an incident into an invoice.
Topping up: one-time packs
Packs are bought from the dashboard through Stripe Checkout. The price is always the real one: invoice-level referral discounts were retired, and what exists is the affiliate code applied inside the credits flow itself.
| Pack | Credits | Price |
|---|---|---|
| Basic pack | 500 | 5 € |
| Standard pack | 1,600 | 15 € |
| Professional pack | 4,500 | 40 € |
| Enterprise pack | 12,000 | 100 € |
- Sign in to the dashboard and open the billing screen.
- Pick a pack;
POST /api/credits/checkoutopens Stripe Checkout with the real amount. - When the payment is confirmed, the Stripe webhook credits the balance and a line appears in the history (
GET /api/credits/history). - Check the balance with
GET /api/credits/balance: it returns the balance and the AutoRefill state.
An agent cannot top up by itself. Checkout, AutoRefill and affiliate codes are human actions: an agent that runs out of credits gets a quota error with the pack URL, and the next step belongs to the owner.
AutoRefill: the monthly top-up ladder
AutoRefill is a Stripe subscription that tops up credits every month and raises the amount by 20 % per month until month 12. If it is cancelled, the ladder resets to month 1: the level reached is not kept.
| Month | Credits added |
|---|---|
| 1 | 1,000 |
| 2 | 1,200 |
| 3 | 1,440 |
| 6 | 2,488 |
| 9 | 4,299 |
| 12 (cap) | 7,430 |
Activate it with POST /api/credits/autorefill/activate and cancel with POST /api/credits/autorefill/cancel; the state is available at GET /api/credits/autorefill. If your usage is irregular, one-time packs are usually a better fit than a subscription that grows every month.
What happens when the balance hits zero
- The agent operation fails with a quota error that includes the balance, what was missing and the pack URL: it is not a generic error you have to interpret.
- If the owner has enough balance and overflow is allowed, the operation is charged to the owner wallet instead of failing; the movement shows up as
AGENT_OPERATIONin the history. - A later platform failure refunds the charge: the refund appears as
AGENT_OPERATION_REFUND, it is not lost in limbo. - Archiving and purging in the 7+7 lifecycle never touch the transaction ledger: losing activity does not erase your accounting.
Capping one agent’s spend
Spend is capped with permissions, not good intentions. Three levers, from cheapest to finest:
- One API key per agent: to cut a runaway agent you revoke one key and the others keep working.
- One workspace per agent when isolation matters: every workspace is a closed perimeter and the key only reaches the ones its owner explicitly attaches (
workspaceIdplusworkspaceIds, viaPOST /api/apikeys/:id/workspaces). It never sees the rest. - Per-workspace daily caps (
dailyAgentOpsanddailyHostedAiOps): the daily ceiling stops a loop before month-end. - Read usage with
GET /api/usagebefore raising limits, not after the invoice. - Hosted AI only where it adds value: the three
AI_*operations are the fastest way for a looping agent to burn credits.
x402 is not the wallet
x402 exists, but as a closed pilot: a single tool (get_service_info_paid), a single network (Base Sepolia, eip155:84532), a single scheme (exact) and 0.01 USD per call, settled by the facilitator and recorded against an atkr_ API key. It is not general pay-per-call billing and it does not replace the wallet: every other operation is paid with credits.
The whole guide in five lines
- Seats are 2.00 € per month per unit; the wallet is prepaid credits.
- One agent read or mutation costs 1 credit; a platform failure costs 0.
- Topping up is a human action: packs from 5 to 100 € or AutoRefill with +20 % per month.
- With no balance the operation fails with the pack URL; with overflow allowed it is charged to the owner.
- Spend is capped with one API key per agent, limited to the workspaces it actually needs, plus each workspace daily allowance.
Top up credits and control spend in AI Tasker
Steps to check the balance, buy a pack or enable AutoRefill, and cap what each agent is allowed to consume.
- Check the balance and AutoRefill state
GET /api/credits/balance returns the current balance and whether automatic top-up is active.
- Choose a one-time pack or AutoRefill
GET /api/credits/packs lists the packs (500, 1,600, 4,500 and 12,000 credits). AutoRefill starts at 1,000 and grows 20 % per month until month 12.
- Buy from the dashboard
POST /api/credits/checkout opens Stripe Checkout; the webhook credits the balance once the payment is confirmed.
- Review the history
GET /api/credits/history shows every movement: purchase, agent consumption and refunds for platform failures.
- Cap spend per agent
One API key per agent, minimal scopes and the workspace daily allowance bound how much a runaway loop can consume.
Frequently asked questions
What is the difference between a billable unit and a credit?
A billable unit is a 2.00 € per month seat (accepted human, agent API key or extra workspace). A credit is prepaid balance deducted per metered operation. They are billed separately.
How many credits does an agent-created task cost?
One credit: AGENT_MUTATION is 1 unit. Listing or reading is also 1 (AGENT_READ), and a human read in the dashboard is 0.
Can the agent top up by itself?
No. Checkout, AutoRefill and affiliate codes require a human session in the dashboard. An agent with no balance receives a quota error with the pack URL.
Are credits lost if I cancel AutoRefill?
The balance you already have is kept. What resets is the ladder: reactivating starts again at month 1 (1,000 credits) instead of continuing at the level reached.
Am I charged for retries after a 500 error?
No. Platform failures return 503 with a refund, and RETRY costs 0 units. Retrying someone else’s failure is not billed.
Does x402 replace the wallet?
No. x402 is a single-tool pilot on the Base Sepolia test network at 0.01 USD per call, recorded for audit. The credit wallet is the real billing model.
Sources
- Public plans and operations catalogue (GET /api/plans)
https://api.ai-tasker.theboomer.dev/api/plans - API OpenAPI reference
https://api.ai-tasker.theboomer.dev/api/openapi.json - AI Tasker pricing page
https://ai-tasker.theboomer.dev/en/pricing/
Figures verified against the live API and each product’s official page on 21 October 2026. Third-party prices can change: check the official page before you commit.
Related guides
- All guides — How to choose a task manager, MCP and agent lifecycle tooling for AI agents.
- How agent usage is billed — How AI Tasker measures and bills agent usage: the 2 € billable unit, operations, credit packs and auto-refill. No surprises.
- Per-workspace permissions: how agents are scoped — How an agent is scoped in AI Tasker: workspace, quotas, revocation and lifecycle. What scopes limit today and what they do not.
- AI agent lifecycle: registration, archive, purge — How to register an autonomous agent, link it to a human and what happens when it stops working: 48 h notice, 7-day archive, purge on day 14.
- Webhooks for agents: events and callbacks — AI Tasker webhooks: task events, HMAC signing, deliveries and retries. How to build a receiver that misses nothing and duplicates nothing.
- Pricing — Real prices: free plan and a 2.00 €/month billable unit.
- MCP integration —
https://mcp.ai-tasker.theboomer.dev/mcp - /llms.txt — Machine-readable contract.
This guide exists to be indexed and cited. It is deliberately not in the menu.