Automation
GitLab 19.6 Doubles Credit Efficiency, Adds Security Agents
The platform now automates vulnerability remediation at scale and lets teams schedule recurring agent tasks while cutting per-credit costs in half.
Key takeaways
- GitLab Credits now perform up to twice as much work per credit through reduced model requests on the Duo Agent Platform.
- Security agents can remediate vulnerabilities across multiple projects automatically, with new exploitability analysis for SAST findings.
- Scheduled automations run recurring agent tasks on cadences from 15 minutes to three months without manual initiation.
- GitLab for Slack integration lets teams create issues and run agent flows directly from Slack threads using their existing permissions.
- New governance controls provide audit trails for all agent activity and let administrators set tool-level permissions for MCP servers.
GitLab has released version 19.6 with significant improvements to its AI agent platform, including a cost efficiency gain that makes each GitLab Credit perform up to twice as much work. The update also introduces automated security remediation capabilities, scheduled agent tasks, and native Slack integration, according to details first reported by VMblog.
The efficiency improvements stem from reducing the number of model requests required for each task on the GitLab Duo Agent Platform. Organizations can redirect the freed-up credits toward new features or other recently introduced services including GitLab Secrets Manager, Dedicated Hosted Runners, and the upcoming GitLab Orbit. The new pricing takes effect November 19, 2026 for new customers, with existing customers transitioning December 18 or opting in earlier.
Why it matters
Security backlogs grow faster than teams can address them, and AI agent costs can spiral as usage expands. GitLab's dual approach—cutting per-task costs while adding automation that works through vulnerability queues without human initiation—addresses both problems simultaneously. The combination lets organizations scale agent use across the software lifecycle without proportional budget increases.
Agentic Security Remediation at Scale
GitLab 19.6 extends agentic remediation so security managers can activate agent flows across multiple projects at once. Teams can now run these flows on non-default branches to fix vulnerabilities before code merges into production.
A new vulnerability context flow, currently in beta for Static Application Security Testing findings, evaluates whether detected flaws are actually exploitable in practice. This allows teams to prioritize based on real-world risk rather than severity scores alone. A mean time to remediation chart tracks whether fixes keep pace with new discoveries.
For exposed GitLab personal access tokens, the platform now offers a "rotate and fix" capability in beta. When secret detection identifies an active token, the system revokes it, generates a replacement with identical permissions, stores the new token in GitLab Secrets Manager, and opens a merge request to remove the hardcoded credential from source code—all in a single automated sequence.
Scheduled Automations and Slack Integration
Scheduled automations, now in beta, run GitLab Duo Agent Platform flows on a recurring cadence ranging from every 15 minutes to every three months. Teams can automate issue triage, continuous integration health checks, and other recurring tasks that previously required manual initiation.
Each automation runs as a dedicated service account with scoped permissions. Audit events separate agent activity from human actions and record every change, providing the governance controls organizations need for automated workflows.
GitLab for Slack, now generally available on GitLab.com for Premium and Ultimate customers, lets teams mention @GitLab in threads or direct messages to create issues, merge requests, and agent flows. Each request runs with the requester's GitLab permissions and creates an agent session that draws from GitLab Credits. Support for self-managed GitLab instances is planned for next month.
Additional Updates
Version 19.6 also includes Model Context Protocol governance controls in beta, allowing group owners to set tool permissions to Always Allow, Always Ask, or Always Deny, with audit events for every tool call. A rebuilt merge request view, now generally available as Rapid Diffs, loads the first file approximately three times faster. Agent sessions can now pause and resume, and teams can define time windows for auto-merge operations.
These details were first reported by VMblog.