LoginSign Up
← Back to all sparks
D

Depot

INFRA · APIS
Velocity6.3

Depot is expanding from faster builds into the whole CI stack — tests, source control, and its own metal.

ci-cdbuild-accelerationtest-analyticssource-controlinfrastructuredeveloper-tooling
◆Current state
Depot has spent the last month building outward from build acceleration. Test results went generally available with JUnit ingest, org-wide flaky and slow test analytics, and timing-based shard balancing. Underneath that, Depot Metal moved CI and Sandboxes onto bare-metal microVMs the company controls end to end, and Depot Code entered private beta as a diskless git server backed by blob storage. The smaller releases fill in the surrounding surface: Tailscale access to private networks, GitLab OIDC, Datadog CI Visibility, stacked pull requests, and macOS 26 runners on M4.
◆Where it's heading
Each layer Depot adds is one it previously rented — compute from cloud runners, source hosting from GitHub, test insight from nothing at all. Owning the storage and hypervisor tiers is what makes the performance claims possible, and owning test data is what turns a build accelerator into something that reports on the pipeline rather than just running it faster. The pattern suggests Depot is positioning as the full CI platform, with speed as the entry point rather than the product.
◆Prediction
Depot Code should move from private to open beta with tighter Depot CI integration, since a git server the company controls is what makes source-aware caching and test selection possible. Expect the test analytics to grow toward selecting which tests to run, not only how to split them.

◆Recent moves

  1. 2d ago

    macOS 26 is now the default macOS version for GitHub Actions runners

    depot-macos-latest now resolves to macOS 26 on M4 hardware with 8 CPUs and 24 GB of memory, with explicit labels left in place for pinning to 14, 15, or 26. Keeping current with Apple toolchains is table stakes for hosted macOS runners, and the disk accelerator carries over from the same platform work behind Depot Metal.

  2. 9d ago

    Depot CI now supports native GitHub stacked pull requests

    Depot CI now evaluates pull_request workflows against a stack's ultimate base branch and runs each layer with its own merge ref, exposing stack metadata for expressions. Stacked pull requests are a workflow that generic CI handles badly, so supporting them natively is a real differentiator inside the CI product Depot is assembling.

  3. 14d ago

    Test results are now generally available

    ⚡ SPARK

    Test results reaching general availability is the clearest step in Depot's move from accelerating builds to reporting on pipelines. It brings flaky-test detection, recurring failure history, and duration-based shard balancing into the same product that runs the jobs, which is a different business than making Docker builds faster.

  4. 22d ago

    AI analysis available for all Depot CI workflows and jobs

    Sherlock analysis moves out of the failure diagnosis card and onto every workflow and job page, so slow, passing, or in-progress runs can be handed to it too. A reach change rather than a capability one, but it reframes the AI from a failure triage tool into a general pipeline inspector.

  5. 28d ago

    Split CI test suites with historical timing data (beta)

    Test suites can be split across shards using durations from earlier JUnit reports, with Depot assigning each shard a time-balanced, non-overlapping subset. The beta that the general availability release two weeks later would fold in, and the first payoff from collecting test timing data at all.

  6. 1mo ago

    Depot CI now supports Tailscale

    CI jobs can join a customer's tailnet as ephemeral, tagged nodes, reaching private databases and internal APIs without public exposure or static IP allow lists. It closes a gap where Depot's runners and container builders had network access that CI jobs did not, which was a blocker for anyone testing against internal services.