# Prefect Cloud — card for AI agents

Workflow automation and orchestration platform for scheduling and monitoring data pipelines.

Maker: Prefect · category: Integrations and automation

## Price (official page)
- Free tier: Hobby plan: 2 users, 5 deployments, 500 min serverless, 7 days retention
- Free tier terms: free plan · amount: 2 users (read from the wording above and the pricing page; empty = not stated)
- Paid from $100/month
- Pricing page: https://www.prefect.io/pricing (checked 2026-10-09)
- Note: Added automatically: its SDK was in 8 of the public agent repos we sampled when it was added. Prices read from the official pricing page.

## Connect
- PyPI: `pip install prefect`
- Repository: https://github.com/PrefectHQ/prefect

## For AI agents on Prefect's own site (AgentGid check 2026-11-13: 4 of 6)
- llms.txt: https://www.prefect.io/llms.txt
- Docs pages as Markdown: 9% of the pages llms.txt lists
- OpenAPI description: https://api.prefect.cloud/api/openapi.json
- Official MCP server: no · official SDK: yes · status page: no
- robots.txt lets AI assistants (ChatGPT-User, Claude-User) read the site: yes

## A year of price history (Internet Archive copies of the pricing page, read by rules)
- Months read: 1 (2025-11 to 2025-11) of https://www.prefect.io/pricing
- Lowest $/month figure on the page: $100 in 2025-11, $100 in 2025-11
- 2025-12: free plan shown (archived copy: https://web.archive.org/web/20251207073150/https://www.prefect.io/pricing)
- A figure is any "$X/month" on the page (plans, seats and add-ons alike); open the archived copy to see which plan it is.

## Price history in AgentGid's catalog (since 2026-10-13)
- Changes recorded: 0 (a vendor price change or an editor's correction; the evidence archive keeps the pricing page of each date)

## Other APIs in this category
- Trigger.dev: https://agentgid.com/apis/trigger-dev/agent.md
- Temporal Cloud: https://agentgid.com/apis/temporal/agent.md
- Inngest: https://agentgid.com/apis/inngest/agent.md
- Arcade: https://agentgid.com/apis/arcade/agent.md
- GitHub: https://agentgid.com/apis/github/agent.md

## Links
- AgentGid page: https://agentgid.com/apis/prefect/


## Prefect Cloud: run and watch scheduled Python jobs

Prefect is a free, open-source Python library that turns ordinary Python functions into jobs that run on a schedule, retry when they fail and keep a history of every run. Prefect Cloud is the hosted dashboard that schedules and watches those jobs; you can also run Prefect's own server yourself. It suits data jobs, such as copying and cleaning data every night.

### Where to use it
- Nightly data copy and cleanup: Pull data from an outside service every night, clean it and load it into your database, with each run logged and visible.
- Process many records at once: Run the same step for each customer, file or order in parallel, and see which ones succeeded.
- Retry calls to a flaky service: When an outside service fails now and then, retry only that step a few times with a pause, instead of rerunning the whole job.
- Slow work behind a web form: Hand long jobs from your web app, such as processing an upload or sending emails, to background workers and show their status.
- Wait for a person's approval: Stop a job until someone checks the result and approves it, then carry on.
- Weekly report to your team: Build a summary of last week's sign-ups or sales every Monday and post it to your team chat. (idea, not from the official pages)

### How to start
- Install Prefect in a virtual environment (Python 3.11 or newer).

```
pip install -U prefect
```

- Log in to Prefect Cloud from the terminal through the browser, or paste an API key.

```
prefect cloud login
```

- Or skip the account and start a local Prefect server with its dashboard at localhost:4200.

```
prefect server start
```


### License
The Prefect library and server are open source under Apache 2.0, free for paid products. Prefect Cloud is a hosted service under Prefect's terms, and you keep the rights to your data.

### Good to know
- Prefect Cloud mostly does not run your code. Your own machine, server or worker runs it, unless you use Prefect's paid-for managed compute.
- The free Hobby plan keeps run history for 7 days and allows 5 deployments and 2 users.
- Every log line counts toward a per-minute limit on Prefect Cloud, so very chatty jobs can be slowed down.
- Prefect sends anonymous usage data, such as its version and first-run events, unless you turn it off.
- A Prefect Cloud access key is shown only once when you create it; save it right away.

Checked 2026-11-10 against: https://www.prefect.io, https://www.prefect.io/pricing, https://www.prefect.io/legal, https://www.prefect.io/legal/terms, https://www.prefect.io/legal/data-processing-addendum, https://www.prefect.io/docs/llms.txt, https://www.prefect.io/docs/v3/get-started/index.md, https://www.prefect.io/docs/v3/get-started/install.md, https://www.prefect.io/docs/v3/get-started/quickstart.md, https://www.prefect.io/docs/v3/concepts/rate-limits.md, https://www.prefect.io/docs/v3/concepts/states.md, https://www.prefect.io/docs/v3/concepts/server.md, https://www.prefect.io/docs/v3/concepts/telemetry.md, https://www.prefect.io/docs/v3/how-to-guides/workflows/retries.md, https://www.prefect.io/docs/v3/how-to-guides/cloud/connect-to-cloud.md, https://www.prefect.io/docs/v3/how-to-guides/cloud/manage-users/api-keys.md, https://www.prefect.io/docs/v3/how-to-guides/ai/use-prefect-mcp-server.md, https://www.prefect.io/docs/v3/how-to-guides/deployment_infra/managed.md, https://www.prefect.io/docs/v3/how-to-guides/deployments/create-schedules.md, https://www.prefect.io/docs/v3/advanced/background-tasks.md, https://www.prefect.io/docs/v3/examples/simple-web-scraper.md, https://github.com/PrefectHQ/prefect/blob/main/README.md, https://github.com/PrefectHQ/prefect/blob/main/LICENSE, https://pypi.org/pypi/prefect/3.9.2/json

### Mistakes to avoid
- Don't: Running flows with no server set up and relying on the temporary server Prefect starts on its own. → Do: Run prefect server start (or log in to Prefect Cloud) and keep it running while you work. (In our run with prefect 3.9.2 the temporary server stopped when the script ended and printed database errors on the way out, and its history was gone.)
- Don't: Running prefect cloud login in CI or a container without flags. → Do: Pass both --key and --workspace, or set PREFECT_API_KEY and PREFECT_API_URL. (The interactive prompts are not available there and the command fails.)
- Don't: Logging every row or item of a big job. → Do: Log summaries, lower the log level, or batch work into fewer runs. (Each log line counts against the per-minute limit and can trigger 429 answers.)
- Don't: Upgrading prefect in the app but not on the self-hosted server. → Do: Keep the same prefect version on both. (New clients can send fields old servers do not know, which ends in 422 errors.)
- Don't: Leaving usage telemetry on for a client project by default. → Do: Set PREFECT_SERVER_ANALYTICS_ENABLED=false or DO_NOT_TRACK=1 when the person wants nothing sent. (Prefect sends anonymous milestone events and server heartbeats unless turned off.)

### Questions
- Is Prefect free? The library and the self-hosted server are free and open source. Prefect Cloud has a free Hobby plan; Starter costs $100 a month.
- Do I need my own server? Not to start. Prefect Managed work pools run flows on Prefect's machines with no worker or cloud account; the free plan includes 500 minutes a month.
- How long is run history kept? 7 days on Hobby and Starter, 14 days on Team, 30 days on Pro, custom on Enterprise.
- Does Prefect train AI models on my data? No, not without your written consent, according to its terms.
- Can my AI assistant look at my runs? Yes, through the Prefect MCP server, which is read-only and still in beta.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
