Replit Agent: an agent that builds, tests and deploys apps

What does the agent do?

According to the official documentation, the agent sets up the project, creates applications, checks its work and fixes problems along the way. It can generate web and mobile apps, slides, data visualizations and documents within one project and one backend. It can also pull data from connected services such as BigQuery, Linear, Slack and Notion through the chat.

The version Replit documents is Agent 4, introduced on March 11, 2026. The announcement highlights parallel work, a task flow you can review and approve before changes merge, and a design canvas inside the project. The October 23, 2026 changelog adds a Monitoring Agent, which investigates production issues using your app’s logs and metrics and tries to fix them. You review and publish the fixes yourself, unless you turn on the experimental option to republish automatically.

What is plan mode and why does it matter?

In plan mode, the agent reads your project and guides you, but does not change your files. You ask it for a task list, which according to the documentation is broken into specific tasks with priorities and dependencies. You review it, choose Revise to adjust it or Cancel to discard it, and only then decide whether to build in the current session or in the background.

For a designer, this is the most valuable moment in the tool. It forces you to see the scope as tasks: which screens, which data, which states. If the plan leaves out error states or permissions, you see that before spending credits building them wrong. And the more detailed your requirements, the better the plan. One limit: according to the Starter plan documentation, plan mode requires Core.

How do checkpoints and background tasks work?

The agent creates a checkpoint automatically when it finishes each request. According to the version control documentation, it captures the complete project state, including AI context and connected databases, and lets you return to that point with one click. Checkpoints are stored in Git, show up in the Git pane and can be synced with GitHub. Still, if the project matters, connect a GitHub repository so you have a copy outside Replit.

Background tasks add a preview before applying. Each task runs in an isolated copy of the project and, according to the documentation, does not change your main version until you review and apply it. When it finishes you see its work log, test results and a live preview, and decide whether to apply or dismiss it. The parallel-work guide warns against batch-approving, because several unreviewed changes make a problem hard to trace. It is the pattern from previewing changes. How many tasks run at once depends on the plan: none on Starter, one on Core, up to ten on Pro and up to 64 on Enterprise.

How much does it cost?

According to Replit’s pricing page, checked on October 29, 2026, in USD:

The last column comes from the task documentation, not the pricing page. That page does not show the Starter plan: the documentation describes it as free, with daily Agent credits, one published app that goes offline after 30 days and building limited to Lite mode. The page also lists higher Pro tiers ($215, $425, $825 and $2,000 a month) without describing them.

Agent usage is charged by effort, not by message: a simple fix costs less than a complex build. Free Mode covers everyday work within an allowance that, on Core and Pro, refreshes every five hours and has weekly caps. Power and Max are paid modes that use monthly credits (Power balances speed and capability; Max is the most capable), and in them you can pick Auto so Replit chooses the model. Replit asks for confirmation before any paid action starts, and in Settings you can set usage alerts and budget limits.

Why does it matter to a designer?

Because it turns a prompt, or a Replit Design frame, into an app with data and publishing in one place. According to the Design documentation, both are entry points into the same project; building an app from a frame requires Core or Pro. It helps validate a flow with users, but the result looks more finished than it is. States you did not describe, business rules and accessibility rarely appear on their own. The pricing page itself warns that the agent uses language models and can make mistakes.

When should you NOT use it?

• When the product handles personal or payment data without a security review (see vibe coding security basics).
• When you need fine control over the infrastructure from the start.
• When the project already has a large, mature repository: an agent working on your own repository is usually better for small changes.
• When you will not approve the plan. Without review, the agent decides the scope for you.

To compare with other tools, see the vibe coding tools comparison.

Example

Illustrative example: a designer on the Core plan wants to test an app to track inventory for a café. She turns on plan mode and writes: “An app with a product list, stock in and stock out, and an alert when something drops below five units.” The agent proposes a task list. She chooses Revise and asks to add the error state when saving a stock-out. Then she builds in the background, reviews the task’s preview before applying it and, when a later iteration breaks the form, goes back to the previous checkpoint. Before sharing the app, she connects the project to GitHub to keep the history outside Replit.

Common mistakes

• Approving the plan without reading it.
• Applying several background tasks without reviewing each one.
• Leaving the project only inside Replit, with no copy on GitHub.
• Asking for a full app in a single prompt with no states or data defined.
• Publishing without reviewing permissions or stored data.
• Not setting a budget limit and discovering the spend at the end of the month.

Sources

• Replit: pricing page
• Replit Docs: Replit Agent
• Replit Docs: Plan vs. Build Mode
• Replit Docs: Task system
• Replit Docs: Build in parallel
• Replit Docs: Version control
• Replit Docs: Replit AI Billing
• Replit Docs: Agent modes
• Replit Docs: Starter Plan
• Replit Docs: What is Replit Design
• Replit Blog: Introducing Agent 4
• Replit Docs: changelog of October 23, 2026