Coding agents 10x’d your PR volume—now your self-hosted GitLab monorepo is drowning. Wake combines deterministic polyglot & Nix dependency graphs with an autonomous CI engineering agent running inside your own VPC and internal LiteLLM.
$ wake agent --gitlab-url https://git.internal --project core/monorepo --open-mr Gathering live X-Ray telemetry from self-hosted GitLab Runner traces + JUnit... ⚡ WAKE X-RAY: 9 jobs · Wall clock 2m 59s · Nix cache hit rate 28% · Lint runs LAST → Turn 1: get_xray_summary() + inspect packages/api/tests/test_integration.py → Turn 4: edit_file() — promote function-scoped DB fixture to module scope + fix CI cache keys → Turn 7: run_command("pytest ...") + verify wake plan child pipeline ✅ VERIFIED: 19.66s → 0.15s (127.7x faster) · 6/6 tests passing · 0 deleted/skipped tests ✅ Opened GitLab MR !3: perf(ci): scope DB fixture, remove test sleeps, fix DAG & cache keys
Engineering leaders never trust probabilistic ML to guess which tests to skip. Wake never guesses—it uses deterministic dependency graphs on MRs and deploys tool-calling agents to permanently fix slow tests, caches, and Nix rebuild cascades.
Parses GitLab Runner trace section timestamps, JUnit timings, child pipelines, and Nix
build logs into an air-gap-safe HTML flamegraph. Spots late linters, missing needs:
DAGs, cache poisoning, and low Nix binary-cache hit rates in 60 seconds.
Scans Python, TypeScript, Rust, Go, and Nix flakes / callPackage
graphs. Emits a native GitLab trigger:include:artifact child pipeline per MR:
UI edits skip unrelated backend integration tests; full suites always run on develop/main.
Runs a multi-turn tool-calling loop against your internal LiteLLM or Anthropic endpoint inside an isolated sandbox. Profiles slow tests, fixes root causes, passes a mechanical anti-cheating soundness gate, and opens proof-backed GitLab MRs.
Real wall-clock telemetry from our polyglot monorepo benchmark running on self-hosted GitLab Runners—zero mocks.
| Scenario | Jobs Executed | Wall-Clock | Runner Compute | Result |
|---|---|---|---|---|
| Baseline Cold Run (No Wake) | 9 / 9 | 2m 59s | 4m 32s | Failed at minute 2:49 on formatting |
| Baseline Warm Cache (No Wake) | 9 / 9 | 1m 23s | ~2m 40s | Every UI edit rebuilds backend & Nix |
| Wake Fast-Fail Stage 0 | 1 / 9 | 21s | 21s | 4.0x faster failure feedback |
| Wake Parallel Dynamic Child Pipeline (UI MR) | 3 / 9 (6 skipped) | 30s (19s child) | 38.3s | 7.1x less runner compute |
| Wake Autonomous Agent (MR !3 — 8 turns) | 6 / 6 tests verified | 19.66s → 0.15s | Isolated sandbox | 127.7x faster test suite |
Generated directly from live GitLab pipelines with zero external scripts or CDNs—ready to open inside restricted VPCs.
Full critical-path waterfall, runner prep overhead breakdown, slow JUnit tests, and 7 prioritized findings.
Live X-Ray against GitLab’s own CLI repo (including !reference YAML tags), spotting late linters and missing caches.
We’ll hand you a single offline Docker command (wake xray) to profile your
slowest pipeline inside your own VPC—zero data egress, real speedup numbers in 15 minutes.