An agentic AI platform from GitLab Inc., built directly into GitLab itself. It became generally available on January 15, 2026. Work such as code review, CI/CD pipeline fixes, and vulnerability analysis is handled by “agents” that each have a defined role. The platform also supports “flows” that chain multiple agents together to automate multi-step work, custom agents that encode an organization’s own rules, and external tool integration through the Model Context Protocol (MCP). A key distinction is billing: instead of per-user AI seats, usage is drawn from “GitLab Credits” pooled across the entire organization.
Key Features
- Specialized agents by role: Foundational Agents for each stage of development ─ release planning, code security, and more ─ are available by default. On top of that, you can define custom agents tailored to team-specific review standards, compliance checks, and deployment automation, as well as external agents that call third-party AI model providers from issues and merge requests
- Flows that chain multiple agents: Multi-step work such as security scanning, code review, test generation, and documentation can be automated as flows that coordinate several agents. GitLab provides foundational flows, and custom flows can also be created
- Same agents from the UI, IDEs, and CLI: GitLab Duo Agentic Chat is available from the GitLab web UI, from multiple IDEs, and from the GitLab Duo CLI. You can instruct the same agents no matter where you are working
- Two-way MCP integration: You can extend agent capabilities by connecting external MCP servers, and GitLab itself can act as an MCP server so that external AI tools can reference GitLab data
- Credit-based billing with no AI seats: With a Premium or Ultimate subscription, every member with a GitLab account can use the agent features. There is no need to purchase additional AI seats per user, and usage is drawn from credits pooled across the organization
- AI governance for enterprises: Policy-based usage controls and support for self-hosted models are provided, so organizations can manage what each agent is allowed to do
Pricing
| Plan | Monthly price (per user) | Included credits for Agent Platform |
|---|---|---|
| Free | $0 | Not eligible for included credits |
| Premium | $29 (billed annually) | $12 per user per month (limited-time promotion, refreshed monthly) |
| Ultimate | $99 (billed annually) | $24 per user per month (limited-time promotion, refreshed monthly) |
| Additional credits | $1 per credit on demand | Volume discounts available with annual commitments (contact sales) |
Usage beyond the included credits is billed as committed monthly credits or on-demand credits. Both synchronous agent usage (such as chat) and asynchronous usage (background execution) draw down credits.
Pricing is current as of August 2026. Please check the official site for the latest information.
Pros & Cons
✅ Pros
- Everything stays inside existing GitLab workflows ─ issues, merge requests, CI/CD ─ so there is no need to carry context over to another tool
- No AI-specific seats to purchase; every Premium or Ultimate member can start using it right away
- Credits are pooled across the organization, so there is no need to allocate usage per person in advance
- Custom agents and flows let you encode your own review standards and release procedures directly into automation
- MCP support allows extension in both directions: pulling in external tools and being connected to from outside
- Policy controls and self-hosted model support make it easier to adopt in organizations with strict governance requirements
⚠️ Cons
- Because billing is usage-based, heavy agent use makes monthly costs harder to predict
- The included credits ($12 for Premium, $24 for Ultimate) are a limited-time promotion, not a permanent free allowance
- The Free plan is not eligible for included credits, so a Premium or higher subscription is effectively required
- Organizations that do not use GitLab as their development platform will see little benefit
- Designing custom agents and flows takes effort to first organize your existing development process
- Agent output cannot be trusted as-is; human review is still required
Comparison with Similar Services
| Item | GitLab Duo Agent Platform | GitHub Copilot | Amazon Q Developer | Cursor |
|---|---|---|---|---|
| Provider | GitLab Inc. | GitHub (Microsoft) | Amazon Web Services | Anysphere |
| Main integration | GitLab (issues / MRs / CI/CD) | GitHub (issues / PRs / Actions) | AWS environments and IDEs | Dedicated editor (VS Code based) |
| Strength | Cross-lifecycle automation and governance | Code completion and agents on PRs | Development and operations support including AWS resources | Code generation and editing inside the editor |
| Billing model | Usage-based credits (no AI seats) | Per-user subscription | Per-user subscription | Per-user subscription |
| Self-hosted models | Supported | Limited | Limited | Not supported |
Who Is It For
- Teams already using GitLab as their development platform who want to reduce rework in reviews and pipeline fixes
- Companies that want to roll AI out across the whole organization rather than a subset of members, but are concerned about per-seat costs
- Engineering leaders who want their own review standards and compliance requirements reflected in agent behavior
- Regulated organizations that need policy-based control over generative AI usage, or want to self-host models
- Platform engineers who want to connect internal tools to AI agents through MCP
Summary
GitLab Duo Agent Platform takes the approach of embedding AI agents into the development platform itself rather than offering them as a separate tool. Its ability to cover the path from issue to merge request to CI/CD in one place, together with credit-based billing that spreads across the organization without buying AI seats, makes it a realistic option for teams centered on GitLab. On the other hand, the difficulty of estimating usage-based costs and the fact that the included credits are a promotion are worth confirming before adoption. A safe path is to start small with the foundational agents and flows, measure actual credit consumption, and then widen the scope.