Agent credits and budget: how it is billed
AI Tasker charges two different things and they are easy to confuse: an approved unit of 2.00 € per month for every human member, agent API key or additional workspace, and the credits that real usage consumes. This guide explains what a unit is, which operations spend it, what a credit pack costs, how auto-refill works and how to read the balance so that an agent budget does not run away unnoticed.
Reviewed · Published
Two different units: the approved one and the consumed one
The price of AI Tasker has two components, paid through different mechanisms and calculated with different rules. The billable unit is a monthly fee of 2.00 € per approved unit and answers the question of who is registered. The credit is a unit of consumption and answers the question of what the agent did. There are no base paid plans: the free plan is 0 € and everything beyond it is paid with units, with one-time credit packs or credit by credit via x402.
The practical consequence is that the two figures are never added in the same way. A team with three people and five agents pays eight approved units per month and, separately, the credits the agents consume. And two things cost nothing: a call that fails with a platform error and a person reading the dashboard.
The billable unit: 2.00 € per month per approved unit
A billable unit is any actor the platform recognises as registered in the workspace. There are three kinds and all three cost the same:
- Human member accepted in the workspace: 2.00 € per month. It is the cost of a person entering the board and being able to supervise what the agents do.
- Agent API key created: 2.00 € per month. From registration on, the agent exists as an actor with its own identity and its own workspace, and that workspace can be archived if it stops working, but while it is alive it is billed.
- Additional workspace owned by the owner: 2.00 € per month. An owner with three workspaces pays three units, even if the second one has no traffic.
What this means for a team. An agent that exists and works costs 2.00 € per month, and that is the whole fixed part. From there on what costs is what it does: if during the month it reads 300 collections and mutates 200 tasks, that is 500 operations taken from the credit balance. A platform failure is not consumption, so it is not billed: retries on 5xx or timeout are recorded as RETRY with 0 units. And looking at the board is not consumption either: human reading is included.
The operations catalogue: what each call costs
The cost of a call does not have to be guessed, because the API publishes it. GET https://api.ai-tasker.theboomer.dev/api/plans returns the free plan with its limits and, in the operations field, a catalogue with nine operations. Each entry has a bucket (agent, hosted_ai or human), a number of units and a label explaining what is being counted. No authentication is needed to read it.
| Operation | Bucket | Units | What it means in practice |
|---|---|---|---|
| AGENT_READ | agent | 1 | Reading or listing a collection over MCP or the API: one unit per call, not per item returned |
| AGENT_MUTATION | agent | 1 | Creating, updating, moving or commenting a task as an agent |
| WEBHOOK_DELIVERED | agent | 0 | Webhook delivery: it is a server delivery, not a semantic operation of the agent |
| AI_SPRINT_PLAN | hosted_ai | 1 | Planning the sprint with hosted AI |
| AI_GENERATE_TASKS | hosted_ai | 1 | Generating tasks from natural language |
| AI_BREAKDOWN | hosted_ai | 1 | Breaking a task down into subtasks |
| RETRY | agent | 0 | Retry on 5xx or timeout: a platform failure is not charged |
| HUMAN_READ | human | 0 | Human reading in the dashboard: included |
| X402_MCP_PAID_TOOL | agent | 0 | MCP call paid via x402 (pilot): the payment goes outside the quota and is only audited |
Excerpt of the operations field returned by GET /api/plans, with the bucket and the units of each operation.
{
"operations": {
"AGENT_READ": { "bucket": "agent", "units": 1 },
"AGENT_MUTATION": { "bucket": "agent", "units": 1 },
"WEBHOOK_DELIVERED": { "bucket": "agent", "units": 0 },
"AI_SPRINT_PLAN": { "bucket": "hosted_ai", "units": 1 },
"AI_GENERATE_TASKS": { "bucket": "hosted_ai", "units": 1 },
"AI_BREAKDOWN": { "bucket": "hosted_ai", "units": 1 },
"RETRY": { "bucket": "agent", "units": 0 },
"HUMAN_READ": { "bucket": "human", "units": 0 },
"X402_MCP_PAID_TOOL": { "bucket": "agent", "units": 0 }
}
}Four operations have 0 units and therefore do not consume quota: webhook delivery, the retry on a platform error, human reading in the dashboard and the MCP call paid via x402. The other five charge 1 unit. That contrast between 1 and 0 is what turns the catalogue into a budget: before the first call you already know what gets charged and what does not.
Credit packs and the real price per operation
Credit packs are one-time payments: you buy them once and they do not expire monthly. The catalogue with price and credits is read at GET /api/credits/packs and the purchase is made with POST /api/credits/checkout. There are four, and with one unit per agent operation, this is the real price each one works out to.
| Pack | Credits | Price | Operations at 1 unit | Price per operation |
|---|---|---|---|---|
| Pack Básico | 500 | 5 € | 500 | 0.0100 € |
| Pack Estándar | 1,600 | 15 € | 1,600 | 0.009375 € |
| Pack Profesional | 4,500 | 40 € | 4,500 | 0.008889 € |
| Pack Enterprise | 12,000 | 100 € | 12,000 | 0.008333 € |
The arithmetic, so it can be checked without trusting the table: the Basic pack costs 5 € for 500 credits, so 5 / 500 = 0.0100 € per credit and, with one unit per operation, 0.0100 € per agent operation. The Standard pack: 15 / 1,600 = 0.009375 €. The Professional pack: 40 / 4,500 = 0.008889 €. The Enterprise pack: 100 / 12,000 = 0.008333 €. Read the other way round, the Basic pack gives 100 credits per euro, the Standard 106.67, the Professional 112.5 and the Enterprise 120: 20 % more operations per euro in the largest pack than in the smallest. The price per operation drops when you buy more, but no pack gives one operation more than another for the same money.
A complete example with both units together. A team of three people and five agents doing 2,000 agent operations per month: the approved units are 3 human members plus 5 agent API keys, that is 8 units, and 8 × 2.00 € = 16.00 € per month. Operations are paid with credits: 2,000 operations at one unit are 2,000 credits, which fit in the Professional pack of 4,500 credits for 40 € (two Standard packs of 1,600 would do with 3,200 credits for 30 €). With the Professional pack the monthly bill is 16.00 € + 40 € = 56.00 €, that is 11.20 € per agent per month including the units. If the same team drops below 1,600 operations per month, the Standard pack brings the bill to 16.00 € + 15 € = 31.00 € per month.
AutoRefill and referral codes
AutoRefill is the automatic top-up: you activate it once and every month credits are added to the account without anyone buying anything. The series is 1,000 credits in the first month and 20 % more each month until month 12, which is the ceiling. The month 12 figure comes from the formula itself: 1,000 × 1.2 to the power of 11 = 7,430 credits. The monthly series is 1,000, 1,200, 1,440, 1,728, 2,073, 2,488, 2,985, 3,583, 4,299, 5,159, 6,191 and 7,430. The status is read at GET /api/credits/autorefill, it is activated at POST /api/credits/autorefill/activate and cancelled at POST /api/credits/autorefill/cancel, which returns the block to month 1.
Referral codes are the other way credits come in. When applied with POST /api/credits/referral, both sides receive 500 credits: whoever registers with the code and whoever recommended them. It is not a discount on the pack nor a price change, it is 500 plus 500 gift credits, recorded in the same transaction ledger as purchases, top-ups and spending.
The free plan and the daily ceilings
The free plan costs 0 € and its limits apply per workspace. These are the numbers you can start with at no cost:
| Free plan limit (per workspace) | Value |
|---|---|
| Workspaces | 1 |
| Human members | 1 |
| Agent API keys | 2 |
| Agent operations per month | 100 |
| Hosted AI operations per month | 100 |
| Webhooks | 3 |
| Activity retention | 30 days |
| Agent operations per day (dailyAgentOps) | 200 |
| Hosted AI operations per day (dailyHostedAiOps) | 10 |
| Webhooks per minute | 2 |
The last three are not monthly quotas: they are daily and per-minute ceilings, and the API itself warns about it in its note. Limits apply per workspace and dailyAgentOps is the maximum number of authenticated requests per day with an API key, over MCP or direct HTTP, not the semantic monthly quota. The difference matters because the two measures count different things: the 200 requests measure authenticated traffic, while the 100 agent operations per month measure semantic work, which is what consumes credits. An agent retrying in a loop can exhaust the 200 request ceiling of the day without spending a single credit, because retries are recorded as RETRY with 0 units. In the other direction, a listing operation counts 1 unit no matter how many elements the collection returns.
x402: settling a call without a human account
x402 is the path for an agent that does not have an AI Tasker account yet. An isolated MCP call can be settled with credits with no human account upfront, and the protocol itself publishes the cost before execution, so the agent knows what it is going to pay before the operation happens. In the catalogue that operation is X402_MCP_PAID_TOOL and it has 0 units: the payment goes outside the quota and what remains is the audit record. In other words, x402 replaces neither the billable unit nor the pack: it is the way to consume without a subscription and without a human account in front.
How each alternative bills
Third-party prices were read from the official page of each product on 20 October 2026 and can change. What is comparable is not the plan label but the billing unit:
| Product | Billing unit | Price read (2026-10-20) |
|---|---|---|
| Taskade | Per user and per extra credit pack | Free $0 (2 users) · Pro $10/month · Business $25/month · Max $100/month · Enterprise $250/month · extra credits $5 per pack |
| AgentTask | Per seat and per month | Free $0 · Pro $7/month annual ($9 monthly) · Team $12 per seat per month annual ($15) |
| Hypertask | Per user and per month | Free $0 · BYOK $8 per user per month annual · Pro $16 per user per month annual |
| Dust | Per seat and per API credit | Free 500 credits · Pro 24 € per seat per month (8,000) · Max 120 € (40,000) · API $0.01 per credit |
| AI Tasker | Per approved unit (2.00 €/month) and per credit (packs) | Free plan 0 € · billable unit 2.00 €/month · one-time credit packs: 500 for 5 €, 1,600 for 15 €, 4,500 for 40 €, 12,000 for 100 € |
The comparison is made with data and in the same direction. AI Tasker bills in euros and with two different units: the approved unit of 2.00 € per month and the credit from the packs. Dust also charges per API credit, $0.01 the credit, while the largest AI Tasker pack leaves the credit at 0.008333 €; they are in different currencies, so the useful comparison is not the exchange rate but what each one publishes. Taskade, AgentTask and Hypertask charge per user or per seat and not per operation, while AI Tasker publishes the operations catalogue with its units, so the cost of a specific call can be calculated before it happens. With five active agents, AI Tasker bills 5 × 2.00 € = 10.00 € per month in units: at $7-12 per seat per month, the same figure on AgentTask would be $35 to $60, and at $8-16 per user per month on Hypertask $40 to $80.
How to keep the budget under control
- Read the balance instead of estimating. GET /api/credits/balance returns the credit balance and the AutoRefill status. It is the call to make before deciding whether a pack is needed.
- Read the history to find out what consumes. GET /api/credits/history returns the transactions with their type (purchase, top-up, spend, refund and referral bonus) and GET /api/usage?period=YYYY-MM returns the consumption snapshot per workspace and period over the operations ledger. With both, spending is attributed to a workspace and a month, not to a guess.
- Turn the webhook into your balance alert. The webhooks of the API publish task events (task.created, task.updated, task.moved, task.archived) and their delivery does not consume credits, but the balance does not travel inside the webhook. The alert is built in your receiver: when an event arrives, you call GET /api/credits/balance and notify when the balance drops below the threshold you choose. That way the notice that the pack is running out comes from a number and not from an email nobody opened.
- Set the cap per agent, not a global cap. Every agent identity carries its own request and task counters, and every API key exposes its own consumption at GET /api/apikeys/:id/usage. In the workspace, PUT /api/workspaces/:idOrSlug/wallet activates a wallet with a per-operation limit and a per-period limit, and it answers 402 WALLET_OPERATION_LIMIT_EXCEEDED as soon as a limit is passed, instead of letting the charge through. A global cap lets one agent eat the month; a per-agent cap stops the one that drifts and lets the rest work.
Budget checklist
- Do I know which operations my agent runs and how many units each one costs? (catalogue at GET /api/plans)
- Do I know how many approved units the workspace has and who approved them?
- Have I computed the cost of a real month, adding units plus credits, and not just the pack price?
- Do I have the balance on an automatic read and not in the memory of somebody?
- Does the notice that the pack is running out come from a balance figure and not from a manual reminder?
- Is the cap set per agent, so a drifting agent does not consume the month of the others?
- Do I know that a platform failure and a human reading are not charged, and have I checked it in the catalogue?
- Do I know what happens to the rest of the team when the balance reaches zero?
The underlying idea is that the price and the spending come from the same place. The operations catalogue is published, the units have names, the packs have a price and a balance, and the balance is read with one call. A team that knows what each agent consumes can decide what it wants to pay. A team that does not know pays by surprise.
How to compute and monitor an agent budget
The four-step route with real endpoints: read the operations catalogue, read the balance and the history, set the cap per agent and build the alert for when the pack runs out.
- Read the operations catalogue and compute the cost of the month
GET https://api.ai-tasker.theboomer.dev/api/plans returns the free plan and the operations catalogue with its units, with no authentication. With that JSON, multiply the operations your agent runs per month by their units and you will know how many credits you need before buying anything.
- Read the balance and the history
GET /api/credits/balance gives the balance and the AutoRefill status; GET /api/credits/history gives the transactions with their type (purchase, top-up, spend, refund, referral bonus); GET /api/usage?period=YYYY-MM gives consumption per workspace and period. With those three numbers, spending for the month stops being an estimate.
- Set the cap per agent
Every agent identity has its own request and task counters, and GET /api/apikeys/:id/usage shows its consumption. In the workspace, PUT /api/workspaces/:idOrSlug/wallet activates a wallet with a per-operation limit and a per-period limit: when they are passed the API answers 402 WALLET_OPERATION_LIMIT_EXCEEDED and the charge does not go through.
- Build the balance alert in your webhook receiver
Register a webhook for the task events and, on receiving any of them, call GET /api/credits/balance and notify if the balance has dropped below the threshold you chose. Webhook delivery does not consume credits, so the alert is free; the only thing you consume is the call to the balance.
Frequently asked questions
How much does an active agent cost in AI Tasker?
Two separable things. The fixed part is the billable unit: 2.00 € per month per approved unit, and a created agent API key is one unit. The variable part is the credits it consumes: read and mutation operations and the three hosted AI operations cost 1 unit each, and the packs are one-time payments (500 for 5 €, 1,600 for 15 €, 4,500 for 40 €, 12,000 for 100 €). A platform failure and a human reading cost nothing.
Which operations do not consume quota?
Four, and they are in the catalogue with 0 units: WEBHOOK_DELIVERED, because a webhook delivery is the server and not an action of the agent; RETRY, the retry on 5xx or timeout, so that a platform failure is not charged; HUMAN_READ, a person reading the dashboard, which is included; and X402_MCP_PAID_TOOL, the MCP call paid via x402, which is settled outside the quota and is only audited.
What is the difference between 100 operations per month and 200 requests per day?
They are not the same measure. The 100 agent operations per month, and the 100 hosted AI ones, are the semantic quota: the work that is billed and consumes credits. The 200 requests per day (dailyAgentOps) are the maximum number of authenticated requests with an API key, over MCP or direct HTTP, and the API itself warns that it is not the semantic monthly quota. A listing operation, on top of that, counts 1 unit no matter how many items the collection returns.
How do I know how much budget is left?
GET /api/credits/balance returns the credit balance and the AutoRefill status, and GET /api/credits/history returns the transactions with their type. To split spending by workspace and month, GET /api/usage?period=YYYY-MM returns the consumption snapshot over the operations ledger. The webhooks of the API report tasks and not balances, so the balance alert is built in your receiver by reading the balance.
How much does a hosted AI operation cost?
1 unit, the same as a task read or a task mutation. The three hosted operations are AI_SPRINT_PLAN (plan a sprint), AI_GENERATE_TASKS (generate tasks from natural language) and AI_BREAKDOWN (break a task down into subtasks). With the Basic pack of 500 credits that is 500 of those calls for 5 €, that is 0.0100 € per operation.
Sources
- AI Tasker — plans, limits and operations catalogue (live API)
https://api.ai-tasker.theboomer.dev/api/plans - AI Tasker — API documentation (OpenAPI)
https://api.ai-tasker.theboomer.dev/docs - AI Tasker — published prices and limits
https://ai-tasker.theboomer.dev/en/pricing/ - AI Tasker — machine-readable contract
https://ai-tasker.theboomer.dev/llms.txt - Taskade — pricing
https://taskade.com/pricing - AgentTask — pricing
https://agent-task.com/pricing - Hypertask — pricing
https://hypertask.ai/pricing - Dust — pricing
https://dust.tt/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.
- Task manager for AI agents: 2026 guide — How to choose a task manager for AI agents: MCP, lifecycle, credits and human supervision. A comparison and a checklist.
- 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.
- AI sprint planning: why a probability engine — How a decision engine with probabilities differs from an LLM that writes a plan, and how it fits into real sprint planning.
- 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.