Replit Agent is the AI agent inside Replit's browser-based development platform that turns plain-language requests into working apps, then hosts and publishes them on Replit's own infrastructure. Replit changed its agent, its modes and its billing several times between September 2025 and November 2026, so many descriptions found online describe an older product. This page sets out what Replit Agent is, what the Core, Pro and Enterprise plans include, how Free, Power and Max modes and effort-based billing work, which deployment types exist, what admins control and how the agent has developed since Agent 3. It is written for builders, engineering managers and the people who approve tools for a team. Every fact comes from Replit's own website, documentation, changelog and blog. Last checked: 2026-11-15.
This is an independent guide from NHI Mgmt Group, based on Replit's public documentation as of November 2026. It is not sponsored or reviewed by Replit. Product features change, so confirm settings against Replit's current documentation.
Key takeaways
- Agent, workspace and hosting in one place. Replit Agent sets up projects, builds the app, tests it in a browser, fixes problems and publishes it to Replit deployments, without a local toolchain.
- Two self-serve paid plans plus Enterprise. As of 15 November 2026 the pricing page lists Core at $20 per month ($18 billed annually), Pro at $100 per month ($90 billed annually) and custom Enterprise pricing.
- Free Mode plus paid modes. Core and Pro include Free Mode within an allowance that resets every five hours with weekly limits; Power and Max are paid modes billed by effort.
- Agent 4 is the current generation. Replit's changelog dates Agent 3 to 12 September 2025 and Agent 4 to 13 March 2026, with Monitoring Agent and Cloud Agents added in October 2026.
- Admin controls are Enterprise-centred. SSO, SCIM, audit logging, role-based access, private deployment policy, deployment geography and bring-your-own-cloud are documented for Enterprise workspaces.
What is Replit Agent?
Replit's documentation describes Agent as a tool that turns plain-language ideas into apps, designs, slides and more. Unlike a chat assistant that suggests code, it takes action: it sets up the project, builds the application, checks its work and fixes problems. The core workflow, in Replit's words, is to "describe what you want in everyday language".
The agent runs inside Replit's cloud workspace, so the editor, runtime, database, secrets store and publishing tool all live on the same platform. That is the main difference between Replit Agent and an AI feature added to a desktop editor: the output is a running, hosted app rather than a diff on a laptop. Replit lists the output types as web apps, mobile apps, data dashboards and AI-powered tools; visual designs and prototypes on the Design Canvas; slides and videos; data visualisations; and files such as CSV, PDF, PowerPoint and Markdown. Several outputs can share one backend in a single project.
Around the main agent Replit now offers further agents. Monitoring Agent, added on 23 October 2026, investigates production issues from app logs and metrics, then attempts and validates fixes. Replit's blog announced Replit Cloud Agents on 27 October 2026 for bringing ideas into a company codebase.
Key facts at a glance
As of 15 November 2026, from Replit's own pages.
| Field | Value | Source page |
|---|---|---|
| Vendor | Replit | Replit website |
| Product type | AI app-building agent inside a cloud development and hosting platform | Agent docs |
| Current generation | Agent 4 (13 March 2026) | Changelog |
| Agent modes | Free, Auto, Power, Max; Plan mode toggle; Enterprise uses a single Auto mode | Agent docs |
| Self-serve plans | Core $20 per month ($18 annual); Pro $100 per month ($90 annual) | Pricing page |
| Free plan | Starter, with limited daily Agent credits and monthly cloud credits | Billing docs |
| Billing model | Monthly credits, effort-based pricing for paid work, top-ups, usage limits and budgets | Billing and AI billing docs |
| Deployment types | Autoscale, Static, Reserved VM, Scheduled | Publishing docs |
| Deployment geography options | North America, South America, Europe (EU), Asia, Australia, or all | Privacy and deployment settings |
| Hosting | "Primarily hosted in the United States" (privacy policy, Section 10) | Privacy policy |
| Certification shown | AICPA SOC 2 Type II | Enterprise page |
Plans and pricing
The table follows replit.com/pricing and the billing documentation as of 15 November 2026. Prices are in US dollars, exclude applicable taxes, and the pricing page states that yearly billing gives "up to 10% off".
| Plan | Price | Included usage | Notable inclusions |
|---|---|---|---|
| Starter | Free | Limited daily Agent credits up to a monthly cap, plus monthly cloud credits | Daily cap on Agent use |
| Core | $20 per month, or $18 per month billed annually | Up to 30 hours of chat and up to 60 projects on Free Mode; $20 toward more powerful models | Unlimited workspaces, AI integrations, up to 5 collaborators, 1 active background task |
| Pro | $100 per month, or $90 per month billed annually, with higher credit tiers available | More Free Mode usage; $100 credits toward the most powerful models; credits roll over for two months | Up to 15 builders with no per-seat fees, 10 parallel Agent tasks, database rollback up to 28 days, access controls on published apps, premium support |
| Enterprise | Custom | Contract | SSO, SCIM and audit logs, advanced privacy controls, centralised security management, single-tenant and private networking options, dedicated premium support |
How Agent work is charged: Replit's AI billing page states that "paid modes and confirmed paid work use effort-based pricing based on request complexity". Agent creates one checkpoint per request, so a complex request bundled into one checkpoint can cost more than a simple fix. Replit asks for confirmation before paid work starts, and a request may offer an "Allow once" option, after which Agent returns to Free. Plan mode can incur billable reasoning on the paid path even when it changes no code.
Third-party services: some features call services such as Claude, ChatGPT and Nano Banana or Gemini, billed at the provider's public API rate.
Overage and spending controls: extra usage is charged only after monthly credits are spent. On Core and Pro, charges accumulate until an invoice threshold that starts at $50 for new accounts and can rise to $250 with account standing, and are otherwise invoiced at the end of the monthly cycle. A usage limit caps spending, and a budget can suspend an app when it reaches the cap. Since 23 October 2026, eligible top-up customers can keep published apps running at zero credits up to an overage limit. Usage appears in Settings > Usage, with data taking up to 30 minutes to show.
Models available
Replit documents Agent by mode rather than by a fixed list of model names. As of 15 November 2026 the Agent and AI billing pages describe the modes below; a per-mode model list is not stated by Replit on those pages.
| Mode | Description (per Replit docs) | Billing |
|---|---|---|
| Free | No-cost everyday work within the plan allowance; Core and Pro limits reset every five hours with weekly limits | Included |
| Auto | Replit selects the model for paid work in Power or Max; Enterprise workspaces use a single Auto mode instead of Free Mode | Effort-based |
| Power | Paid mid-tier mode using powerful models | Effort-based |
| Max | Paid top-tier mode using frontier models for demanding work | Effort-based |
- Intelligent Model Routing: announced on Replit's blog on 26 August 2026.
- Enterprise model governance: the privacy settings documentation links to an "Agent modes and models" page covering approved models and access to higher effort levels for workspaces.
- Behaviour: the pricing page states that Agent is powered by large language models, that its behaviour is probabilistic and that it may occasionally make mistakes.
Features and modes
| Feature | What it does (per Replit docs) | Added |
|---|---|---|
| Plan mode | Agent brainstorms, asks questions and proposes an ordered task list before changing code or data; you accept or revise | Agent 4 |
| Checkpoints and rollback | Agent creates checkpoints as it works so you can roll back to any previous state | Core feature |
| App Testing | Agent opens the app in a real browser, finds problems and fixes them in a check-and-fix loop | Agent 3 |
| Long autonomous runs | Up to 200 minutes of independent work with Live Monitoring | Agent 3 |
| Agents and Automations | Build Slack bots, Telegram bots and scheduled workflows; external triggers need an Autoscale or Scheduled deployment | Agent 3 (beta) |
| Design Canvas | Explore mockups and prototypes before committing to code | Agent 4 |
| Parallel tasks | Agent works on several tasks at once (1 active background task on Core, 10 on Pro) | Agent 4 |
| Team collaboration | Teammates join a project and build at the same time | Agent 4 |
| Connectors | Query BigQuery, Linear, Slack, Notion and other connected services from chat; Replit-managed and third-party integrations | Agent 3 and 4 |
| Skills | Agent suggests reusable workspace skills that you approve before saving | 23 October 2026 |
| Monitoring Agent | Investigates production issues from logs and metrics, then attempts and validates fixes | 23 October 2026 |
Supported IDEs and platforms
- Browser workspace: Replit Agent runs in Replit's cloud workspace, so there is no IDE plugin to install; the editor, runtime and agent sit on Replit's platform.
- Desktop app: Replit's enterprise page refers to choosing which folders agents can access in the desktop app.
- Publishing: four deployment types. Autoscale adjusts resources to usage; Static hosts sites that do not change with user input at low cost; Reserved VM runs continuously with predictable pricing and performance; Scheduled runs the app at chosen times. Custom domains are supported.
- Data: Replit advises against relying on a published app's filesystem and points to its Storage and Database offerings. Pro includes database rollback up to 28 days.
- Source control: admins can require a Git remote and restrict allowed Git remote targets to GitHub, GitLab or Bitbucket.
- Data platforms: Replit's blog announced a Databricks integration with native Lakebase support on 10 September 2026.
Enterprise and admin controls
| Control | Documented setting or feature | Where |
|---|---|---|
| Identity | SSO (with "SSO enforced" setting), SCIM provisioning, role-based access, people and groups | Enterprise page |
| Audit | Searchable audit logs of security-relevant activity | Enterprise page |
| Publishing policy | Private deployments: Off, Require, or Require for non-admins; public publishing exceptions reviewed under Manage | Settings > Advanced |
| Access to apps | Password-protected deployments: Enabled, Disabled, or Disabled for non-admins | Settings > Advanced |
| Development previews | Require private development URLs (applies to existing apps, and uses SSO when SAML is enabled) | Settings > Advanced |
| Data exfiltration routes | Ban source code export (ZIP); ban attachment uploads in chat and file imports | Settings > Advanced |
| Location | Deployment geography: all, North America, South America, Europe (EU), Asia or Australia | Deployment settings |
| Source control | Require Git remote, private remotes, allowed Git remote targets | Workspace settings |
| Vulnerability gate | Security Center dependency scans, Auto-Protect, Block publishing at severity | Security settings |
| Integrations | Control which groups can use connected systems | Enterprise page |
| Infrastructure | Bring Your Own Cloud: run the data plane in your company's cloud tenant (announced 16 October 2026) | Enterprise page, blog |
Replit's documentation notes limits that matter for policy: the private deployment and password settings do not change apps already published, the export ban does not stop code leaving by other routes, deployment geography controls where projects publish rather than where every connected service stores data, and "a clean scan does not guarantee that an application is secure". Connector API keys, app secrets and database credentials created by or for the agent are non-human identities, so give each an owner and a rotation plan. See AI Coding Tools and Secrets Sprawl.
Data handling and privacy
- Machine learning use: Section 6 of Replit's privacy policy lists a legitimate interest in using personal data "to improve the accuracy of our machine learning technologies such as code generation".
- Service providers: Section 4 states that third-party providers powering Agent and voice dictation, including machine learning companies, process data on Replit's behalf.
- Hosting: Section 10 states that the services are primarily hosted in the United States and may be hosted in other jurisdictions; Enterprise deployment geography and Bring Your Own Cloud change where apps and the data plane run.
- Retention: Section 8 states that data is deleted or de-identified once no longer needed in identifiable form, unless law requires longer retention; no fixed period is given.
- Enterprise statement: the enterprise page states "Your knowledge stays yours"; organisations should confirm training and retention terms in their Enterprise agreement and data processing terms.
- Assurance: the enterprise page shows AICPA SOC 2 Type II.
Release history
Dated milestones from Replit's changelog and blog. Earlier agent versions are not covered by the pages checked for this guide, so the history starts with Agent 3.
| Date | Release or event | Headline changes |
|---|---|---|
| 12 September 2025 | Agent 3 | App Testing; up to 200 minutes of autonomy with Live Monitoring; agent and automation generation; connectors |
| 13 March 2026 | Agent 4 | Design Canvas, Plan mode, parallel tasks, team collaboration, tool connections |
| 16 August 2026 | Enterprise governance | "Govern Replit at scale" governance tools |
| 18 August 2026 | Free Mode | Free Mode introduced |
| 26 August 2026 | Intelligent Model Routing | Model routing announced |
| 10 September 2026 | Databricks | Databricks integration with native Lakebase support |
| 16 October 2026 | Bring Your Own Cloud | Data plane in the customer's cloud tenant |
| 23 October 2026 | Monitoring Agent and skills | Production monitoring agent, workspace skills, chat sharing, apps kept running at zero credits |
| 26 October 2026 | Replit for Business | Business offering introduced |
| 27 October 2026 | Replit Cloud Agents | Bringing ideas into the company codebase |
| 12 November 2026 | Agent for business tasks | Agent can buy for your business and help you reach customers |
Limits and known constraints
- Free Mode windows: on Core and Pro, Free Mode limits reset every five hours and have weekly limits; Starter has a daily cap.
- Effort-based cost: the cost of a paid request depends on its complexity, and one checkpoint can cover a lot of work, so set usage limits before heavy use.
- Background tasks: 1 active background task on Core and 10 on Pro.
- Collaborators: up to 5 on Core and up to 15 builders on Pro.
- Published app access controls: available on Pro and Enterprise.
- Policy settings are forward-looking: private and password publishing policies apply to newly published apps only.
- Pricing churn: record the date and billing period whenever you quote a Replit price.
Security guidance
This page describes what Replit Agent is and costs. Hardening is covered separately in the Replit Security Guide, which goes through identity and access, admin roles, API tokens, development and production separation, checkpoints and approvals, publishing controls and logging. The short version for a buyer: decide on Enterprise before company-wide use so SSO, SCIM, audit logs and the private deployment policy are in place; set usage limits and budgets; use Plan mode and checkpoints for changes that touch data; and treat apps the agent builds as code that still needs review before it handles real users. The wider discipline is in the Vibe Coding Security Guide and the AI IDE and Coding Agent Security Guide.
Practitioner checklist
- Record the date, plan and billing period whenever you quote a Replit price, and recheck replit.com/pricing before renewal.
- Set a usage limit and app budgets before enabling Power or Max, since paid work is billed by effort.
- Choose Enterprise for organisational use and connect SSO and SCIM before inviting builders.
- Set Private deployments to Require or Require for non-admins, and turn on Require private development URLs.
- Pick a deployment geography that matches your data residency obligations, and check connected services separately.
- Turn on Auto-Protect and Block publishing at severity, while keeping human review of agent-built apps.
- Inventory connector keys, app secrets and database credentials created for agent-built apps, with an owner and rotation date for each.
- Read the Replit changelog monthly and update internal guidance when modes, limits or defaults change.
Standards and references
- Replit: Pricing
- Replit Docs: Replit Agent
- Replit Docs: AI billing
- Replit Docs: Billing
- Replit Docs: About publishing
- Replit Docs: Privacy and deployment settings
- Replit Docs: Changelog, Agent 4
- Replit: Enterprise
Related NHI Mgmt Group resources: Replit Security Guide · Vibe Coding Security Guide · AI IDE and Coding Agent Security Guide · AI Coding Tools and Secrets Sprawl