# What does an always-on agent actually cost?

Count subscriptions, inference, computers, channels and failed retries separately. Included usage is an allowance, not a guarantee of zero marginal cost.

- Citation: https://botclaw.tech/items/obs-topic-cost
- Revision: 2
- Published: 2026-10-01T16:24:26.193Z
- Reviewed: 2026-10-01T15:50:00.000Z
- Updated: 2026-10-01T16:24:32.310Z
- Section: topic
- Underlying source date: Unknown; do not infer a release date
- Review state: current; 0 pending source changes
- Applicable version: Not pinned; check current documentation

Start with one completed task as the unit of comparison. A cheaper model call is not cheaper work if it produces repeated failures or needs extensive review.

## Build the cost ledger

For each task record plan fees attributable to use, inference usage, infrastructure time, paid external actions, retries and human review. Keep observed charges separate from estimates and vendor examples.


## Do not assume open means free

A self-managed runtime can avoid a hosted subscription while still requiring model inference, a machine and maintenance. For a hosted agent, check included work, overage and any separately billed services before scheduling.


## A bounded trial

1. Use one non-sensitive read-only task with a time and usage ceiling.
2. Disable optional paid actions and review the actual usage record.
3. Estimate a month from successful tasks and failed retries, not just one demo.

## Evidence and limits

- official_documentation
- Source-backed facts plus BotClaw editorial interpretation. No product account, inference or agent performance test was run.

## Sources

- https://x.ai/bot
- https://manus.im/zh-cn/blog/introducing-manus-2-0
- https://help.openai.com/en/articles/20001530-getting-started-with-your-dot