GitLab Duo Agent Platform

GitLab Duo Agent Platform is GitLab's built-in agentic AI layer that runs planning, coding, code-review, CI/CD and security agents inside the GitLab DevSecOps platform. It is for engineering organisations already on GitLab that want AI agents governed by the same permissions, audit trail and deployment model, including self-managed instances and self-hosted models, instead of bolting on a separate assistant.

GitLab Duo Agent Platform is the agentic AI layer of GitLab's DevSecOps platform, generally available since January 15, 2026. Rather than a standalone coding assistant, it orchestrates AI agents across the software delivery lifecycle (planning, coding, review, CI/CD and security) using the project, pipeline and vulnerability context GitLab already holds. It ships foundational agents (Planner, Security Analyst, Data Analyst and CI Expert), Agentic Chat in the web UI and editor extensions, and foundational flows that automate multi-step work such as turning an issue into a merge request, migrating CI/CD pipelines, fixing failed pipelines and reviewing code. Teams publish custom agents and flows through an AI Catalog, connect external agents such as Anthropic's Claude Code and OpenAI's Codex CLI, and use MCP clients and servers to reach outside tools. Anthropic Claude Sonnet 4 is the documented default model; self-managed customers can instead run self-hosted models behind GitLab's AI Gateway. Usage is metered in GitLab Credits at $1 each, with Premium and Ultimate seats carrying 12 and 24 included credits per user per month under a promotional allowance, and agentic code reviews billed at a flat $0.25 each. GitLab 19.4, released September 17, 2026, added a /goal command in the Duo CLI (public beta), three open-weight models (GLM 5.3, Kimi K3 and MiniMax M3) that GitLab says deliver up to 4x more calls per credit, separate model selection for the Developer Flow, unified allow/ask governance for GitLab MCP server tools, and generally available per-user credit reporting with per-user caps. NatWest is a named customer. For organisations standardised on GitLab it is the most direct alternative to GitHub Copilot's agent features and to third-party coding agents.

Category

Developer Tools

Pricing

Subscription, Usage-based, Freemium

Target Market

CTOs, VPs of Engineering, Platform Engineering Teams, DevSecOps Leaders

Deployment

Cloud-first, Self-hosted, Hybrid

Customers

GitLab reports 50M+ registered users and about 50% of the Fortune 100 as customers (platform-wide, not Agent Platform specific)

• ✓
  Foundational agents

  Pre-built Planner, Security Analyst, Data Analyst and CI Expert agents work on issues, vulnerabilities, data and pipelines using GitLab's own project context.

• ✓
  Foundational flows

  Multi-step automations convert issues into merge requests, migrate CI/CD configuration, fix failed pipelines and review code without hand-offs between tools.

• ✓
  AI Catalog

  A central catalog where teams create, publish and share custom agents and flows, so reusable automation is governed rather than scattered across repos.

• ✓
  External agents and MCP tool governance

  Connects Claude Code and Codex CLI as external agents and exposes GitLab MCP server tools, with per-tool Always Allow or Always Ask modes.

• ✓
  Model choice including open-weight and self-hosted

  Administrators pick models per feature, including GLM 5.3, Kimi K3 and MiniMax M3, or run self-hosted models on self-managed instances via the AI Gateway.

• ✓
  GitLab Credits spend controls

  Usage-based credits with per-user caps, hard monthly subscription limits and per-event usage exports let finance teams forecast and cap agent spend.

• ✓
  /goal command in Duo CLI

  A public-beta command that delegates an open-ended objective to a governed local flow with verification steps, which developers can pause or redirect at any time.

• •
  Issue-to-merge-request automation

  A developer hands a well-scoped issue to the flow, which drafts the code change and opens a merge request for human review and approval.

• •
  Pipeline failure triage

  Platform teams let the CI Expert agent and pipeline-fix flow diagnose failed CI jobs and propose fixes, cutting time spent reading job logs.

• •
  Vulnerability remediation

  Security teams use the Security Analyst agent to review scanner findings, explain severity and suggest remediation inside the existing merge request workflow.

• •
  CI/CD pipeline migration

  Teams modernising their build systems use the migration flow to convert existing pipeline configuration into GitLab CI with far less manual rewriting.

• •
  Governed agent spend

  Engineering leaders set per-user credit caps and export per-event usage data to charge back agent consumption by team, namespace or project.

Agent Platform usage is metered in GitLab Credits at $1 each on top of seat subscriptions: Premium lists at $29 per user per month billed annually and includes 12 credits per user per month, while Ultimate is custom-priced with 24 credits per user per month; GitLab describes both allowances as promotional and subject to change. Agentic code review is a flat $0.25 per review regardless of merge request size. Free-tier GitLab.com groups can buy credits in monthly blocks, and admins can set per-user caps and hard monthly limits.

This page was written from 10 sources, 8 on domains other than about.gitlab.com.

1. 1.about.gitlab.com — gitlab duo agent platformvendor
2. 2.ir.gitlab.com — default
3. 3.docs.gitlab.com — gitlab 19 4 released
4. 4.docs.gitlab.com — duo agent platform
5. 5.about.gitlab.com — pricingvendor
6. 6.trust.gitlab.com — trust.gitlab.com
7. 7.infoq.com — gitlab flatrate view ai access
8. 8.stocktitan.net — git lab 19 4 brings new agentic automation at a lower 2xyng4
9. 9.news.ycombinator.com — item
10. 10.news.ycombinator.com — item