Program Snapshot: Oct 16 – Nov 5, 2026
Hi team, we’re now targeting Nov 10 for the Pilot as we close out a few final items. The team has kept shipping, so check out all of the great work below.
Overall status
At risk. We’re down to three remaining workaround items for Artifact Central (AC) and IAM before the Pilot. More details in Callouts. We’re also validating GA scope and will share any plan updates by end of next week.
Highlights
GitLab’s E2E tests on Cloud Native GitLab now run on Caproni. Every
e2e:test-on-cng job in the monolith pipeline now runs on Caproni and blocks
merges, testing against the same CNG images and Helm charts we ship to
customers. When a job fails, you can rebuild the same environment on your
laptop with caproni up, using the config CI used. Each job runs about 1 to
1.5 minutes faster, and reliability held up: in our validation sample, 8 of 49
pipelines had a failing job on Caproni versus 13 of 49 on the old deployer.
We’re now retiring that deployer (~8,500 lines of Ruby).
Kudos to Harsha Muralidhar, Jay McCure and Manuel Schönlaub.
Boot a working, self-hosted GitLab from a single image with Project Crete. The prototype is complete, with settings preserved across updates, and the dev image docs show how it works. With scope now decided, including Community Edition support, the team moves into the Experimental phase in November.
Kudos to Grant and the Operate team for delivering this milestone early.
Theseus teams now have one standard way to build services, from laptop to production. TUBE, our build tool, is now integrated with Caproni, so the build you test locally is the same one our pipelines ship. That removes a common source of “works on my machine” bugs, and teams no longer maintain their own build setups.
Kudos to Alec and Robert Marshall.
Automated performance testing now runs in CI for both AC and Orbit. AC’s
first performance tests run on every MR and found CPU throttling at just 5
concurrent users, with p95 latency jumping from 9.5 ms to 79 ms. That led to a
production fix projected to give each pod about 2.5x the capacity. Orbit now
gets regression testing on every merge to main and every query-path MR, after
we cut run-to-run variation from 33% to 4% by running Caproni as a headless
perf environment in CI. The same playbook works for any service Caproni can
deploy.
Kudos to the Performance Enablement team.
Go and Rust services on Theseus can now use our standard rate limiting. Limits are set in code for now, so changes need a deploy, until config-file rules (ETA Dec 18) and a rule service that applies changes instantly (ETA Jan 29) remove that step.
Kudos to Sankalp and the Rate Limiting team.
Self-managed components using our Kubernetes-based install method can now all be configured the same way. Orbit and AC each describe what they need in a single file, and Fairway turns it into one deployment package for GitLab.com, Dedicated and Self-Managed, so there are no separate setups to maintain. Built-in checks catch configuration mistakes before anything is installed.
Kudos to Brittany Wilkerson, John McDonnell and Matthew J. Hoad.
There’s now one standard for every modular feature’s HTTP API. ADR 008 sets an OpenAPI 3.1 spec with a shared quality bar, whatever the runtime, enforced by a versioned lint ruleset and a CI component that blocks non-conforming MRs. New modules get a starter spec, and AC already passes.
Kudos to the API Platform team.
Callouts
What’s left for the Pilot. The remaining Pilot work is on the AC and IAM side. AC and IAM need a pre-shared key (PSK) set up during install to communicate securely. That requires a multi-stage deploy (deploy, read, deploy), which Instrumentor and Runway don’t support, so we’re finishing a workaround. Two of the three remaining items are targeting Nov 10. The third, Runway permissions on Dedicated, has fixes in review, and we’ll share a date once the team confirms. Separately, onboarding Orbit to the new LabKit config manager is underway.
The new Hangar dashboard is live, so we can now measure Hangar adoption on the way to GA. It shows how many engineers use Hangar each week, how many environments they start and how often setup succeeds. Next, it’ll add how quickly environments are ready, how teams share them and usage per team.
Shout out to Tiffany Rea for quickly getting these metrics up and running.
Next up
| Date | Landing |
|---|---|
| Oct 29 | Done — documented testing strategy for every new modular component; standardized build images; hardened, CI-compatible build image; automatic, self-serve monitoring across every component |
| Oct 31 | Done — single-machine install prototype, set up with one command (Crete) |
| Nov 10 | Theseus Pilot complete, with AC and Orbit |
| Nov 13 | Done, delivered early — stable, documented local dev environment that stays consistent with production; every Kubernetes-based self-managed component configured the same way |
| Nov 20 | Every service deploys from automatically generated configuration, fully self-serve; identity and secrets management handled centrally, with secure defaults built in |
| Nov 27 | A service can use multiple databases and add more as load grows |
| Nov 30 | A single, standard way for engineers to make changes to the monolith (about half already use it, and the older tool is being phased out) |
Check out our artifacts at go/theseus.