Notion AI Credits: Custom Agent Cost Explained

Notion credits are a usage budget for Custom Agents and other credit-based capabilities. Monthly Notion credits cost $10 per 1,000 credits, are shared across a workspace, reset monthly, and do not roll over. A Custom Agent consumes credits according to how much it reads, how many steps and tools it uses, how often it runs, and which model it selects. If the balance runs out, credit-based agents pause.

The short answer

Notion Business and Enterprise include core AI features such as Notion Agent, AI Meeting Notes, and Enterprise Search. They do not include unlimited autonomous Custom Agent runs. Starting May 4, 2026, Custom Agents run on Notion credits purchased separately.

The correct budget formula is:

Monthly agent cost = sum of each workflow’s runs × observed credits per run × $0.01 per credit

Because 1,000 credits cost $10, one credit has a nominal monthly price of one cent. Actual enterprise arrangements may include annual or on-demand terms, so use the contracted rate when applicable.

What uses Notion credits?

Notion’s current documentation explicitly associates credits with:

• Custom Agent runs;

• Custom Agent Autofill;

• Claude agents hosted in Notion;

• Workers when their announced credit transition takes effect;

• other capabilities Notion may designate as credit-based.

Basic Autofill on Business and Enterprise does not use credits. Notion Agent, AI Meeting Notes, and Enterprise Search are described as included in Business and Enterprise.

Packaging can change. Admins should use the live Notion credits dashboard and current order form as the source of truth.

What determines credit consumption?

Content read

An agent that reads one short page generally needs fewer credits than an agent that searches several large databases, scans long documents, and loads connected-app context.

Reduce unnecessary reads with:

• narrow triggers;

• database filters;

• explicit source lists;

• bounded time ranges;

• compact authoritative pages;

• separate agents for unrelated jobs.

Steps and tool calls

Each additional search, classification, database update, notification, or external tool call adds work. A routing agent that assigns one owner is cheaper than a workflow that also researches background, scores urgency, writes a plan, updates multiple fields, and sends messages.

Frequency

Frequency is often the largest multiplier. A five-cent job running hourly for 30 days costs roughly $36. A 30-cent job running weekly costs roughly $1.30 per month.

Always calculate runs per month before debating the per-run range.

Model

Notion says advanced models use more credits and recommends Auto for most cases. Teams that pin a model for quality or consistency should measure the resulting consumption instead of assuming the same rate.

Retries and failure paths

A workflow that repeatedly encounters missing context, timeouts, or ambiguous instructions can consume credits without producing accepted output. Track failed and retried runs separately.

Official Notion cost ranges

Notion publishes indicative costs for common Custom Agent workflows.

The range is wide because “the same” workflow can read different amounts of data, take different paths, or use different models.

How many runs do 1,000 credits buy?

Notion’s published estimates translate a $10 pool into approximate run counts:

Treat the low end as the safer planning case until production evidence shows otherwise.

A practical credit forecasting model

Create one row per agent:

Then calculate three scenarios.

Expected

Expected runs × median credits per run

Use for the ordinary monthly plan.

Busy month

P95 runs × high-percentile credits per run

Use to determine the workspace reserve and alert threshold.

Failure month

Busy-month volume × retry multiplier

Use to test whether a broken trigger or source connector could exhaust the shared pool before an admin reacts.

Worked examples

Q&A agent in a Slack channel

Assumptions:

• 15 questions per workday;

• 22 workdays;

• 330 runs/month;

• official cost range $0.03–$0.11.

Estimated monthly cost: $9.90–$36.30.

The wide range makes a pilot valuable. If most questions can be answered from a compact verified handbook, the lower end may be realistic. If the agent searches many connected sources and resolves conflicting policy, cost can rise.

Weekly status agent for ten teams

Assumptions:

• 10 agents;

• one run per week;

• 4.3 weeks/month;

• 43 total runs;

• official cost range $0.08–$0.18.

Estimated monthly cost: $3.44–$7.74.

This looks inexpensive, but review time may dominate the total cost if the drafts are frequently corrected.

Daily brief for 50 managers

Assumptions:

• 50 people;

• 22 workdays;

• 1,100 runs;

• official range $0.10–$0.30.

Estimated monthly cost: $110–$330.

A smaller shared team brief or an event-triggered workflow may deliver similar value with fewer runs.

High-frequency routing agent

Assumptions:

• 100 requests per day;

• 30 days;

• 3,000 runs;

• official range $0.05–$0.15.

Estimated monthly cost: $150–$450.

Before scaling, test whether a rule-based filter can handle obvious cases and reserve the agent for ambiguous requests.

Why unused monthly credits matter

Monthly credits reset and unused credits do not roll over. That creates two opposing risks:

• underbuying: agents pause when the pool is exhausted;

• overbuying: unused monthly capacity expires.

A sensible rollout starts with a bounded amount, measures one or two billing cycles, and adjusts. Notion supports changing monthly credits and offers Auto Adjust, annual, or on-demand arrangements in some purchasing channels.

What happens when credits run out?

Notion says Custom Agents pause automatically until credits reset or an admin adds more. Admins receive notifications at 80% and 100% usage.

A paused agent is an operational event, not only a billing event. For each workflow, define:

• who is notified;

• whether incoming work queues safely;

• which manual fallback applies;

• whether adding credits requires approval;

• how skipped schedules are handled after restart.

Never let a credit pause silently drop customer requests, compliance reviews, or operational alerts.

How should admins allocate credits?

For one workspace, use agent-level limits where available and monitor each agent’s consumption.

For organizations with several workspaces, Notion supports allocation from the organization console. Credits may be pooled at the contract level, while organization admins place monthly limits on individual workspaces. When a workspace reaches its limit, credit-based work pauses there without necessarily stopping other workspaces.

A useful allocation policy separates:

1. production agents with measurable owners;

2. pilots with strict caps;

3. personal experiments;

4. incident reserve.

The four metrics that matter more than credits

Cost per accepted output

A cheap draft rejected by a reviewer has little value. Count outputs that actually pass the workflow’s acceptance rule.

Human review minutes

Agent cost can be lower than reviewer cost. Measure how long someone spends checking, correcting, and completing the result.

Failure and retry rate

Repeated runs may indicate weak source quality, bad instructions, or brittle integrations.

Cycle-time improvement

A workflow is economically useful when it reduces the elapsed time to route, decide, answer, or publish—not merely when it runs.

How to reduce Notion credit usage safely

1. Trigger on meaningful events. Avoid polling when a database or app can emit a specific change.

2. Filter before invoking the agent. Use deterministic rules for obvious exclusions.

3. Limit source scope. Give the agent the smallest authoritative context set.

4. Reduce duplicate work. Do not let several agents summarize the same event independently.

5. Separate retrieval from action. A lightweight lookup may not need a multi-step autonomous workflow.

6. Use structured inputs. Explicit fields reduce exploratory searches.

7. Set stop conditions. Ask the agent to abstain when required evidence is missing.

8. Review actual run traces. Find unnecessary reads, loops, and notifications.

9. Start with Auto model selection. Pin a more expensive model only when measured quality requires it.

10. Archive obsolete agents. Dormant schedules and forgotten triggers are a recurring cost risk.

Optimization must not weaken permission checks, citations, or human approval for consequential actions.

How do permissions affect credit planning?

Custom Agents have their own permissions. They can be granted access to pages, databases, and connected tools, and users of the agent may receive results derived from that access.

This makes access design part of cost and risk management:

• broader access can increase retrieval work;

• ambiguous access can cause failures and retries;

• an agent shared too widely can expose information through its response;

• read permission does not automatically justify write or action permission.

Review the agent’s access scope and the people allowed to use it before increasing its limit.

A rollout checklist

Before the pilot

• define one workflow and one owner;

• document its trigger and maximum frequency;

• restrict its source and action scope;

• define a correct output;

• set a credit cap;

• create a manual fallback;

• choose a review period.

During the pilot

• measure median and high-percentile credits per run;

• count accepted, rejected, failed, and retried outputs;

• inspect the most expensive traces;

• validate that permissions and citations remain correct;

• compare time saved with review time.

Before expansion

• forecast expected and busy-month usage;

• add alerts and incident ownership;

• decide how shared-pool contention is handled;

• document when to pause or disable the agent;

• review cost per accepted output at least monthly.

How does this affect a Notion-versus-Dokki decision?

Notion’s pricing unit combines per-member plans with a separate pool for autonomous work. That is appropriate to evaluate as two layers: human workspace access and agent execution.

Dokki is built around shared, addressable resources that people and authorized AI agents can work on together. A team comparing the platforms should ask:

• how many humans need full membership;

• how many agents run autonomously;

• whether the work must remain inspectable in the same document or table;

• whether external AI clients connect through scoped access;

• which unit best reflects value: seats, runs, shared resources, or completed workflows.

The answer depends on the operating model, not on the cheapest headline number.

Frequently asked questions

What is one Notion credit worth?

Monthly credits are priced at $10 per 1,000, which implies one cent per credit at the public monthly rate.

Do Notion credits roll over?

Monthly credits reset each month and unused monthly credits do not roll over.

Are credits included in Notion Business?

Business includes core AI capabilities, but Custom Agents use the separate credit system. Buyers should verify any trial or contracted credits in their workspace.

Can one agent consume the whole workspace pool?

Credits are shared, so an agent with high usage can affect other agents unless limits and monitoring contain it.

Can admins buy more credits mid-month?

Yes. Notion says increases become available immediately; billing treatment depends on the monthly service cycle and purchase arrangement.

Do Custom Agents stop when credits are exhausted?

Yes. They pause until credits reset or more are added.

How much should a team buy first?

Use a small capped pilot, observe one or two billing cycles, then size the pool from actual credits per accepted output and busy-month volume.

_Last verified: July 21, 2026._

Sources

• Notion: Understand pricing for Custom Agents

• Notion Agents

• Notion: Purchase Notion credits

• Notion: Allocate credits to each workspace

• Notion: Custom Agents sharing and permissions

• Notion pricing