Work the day shift.
Ship the night shift.

Those tools assist you while you code. Nightgauge runs the entire issue-to-PR pipeline on its own — six structured stages, 94 verified phases, deterministic gates after every phase. You start with a GitHub issue; you end with a merged PR. The IDE editing experience isn't the product. The pipeline is.

Yes. The core — VS Code extension, SDK, portable skills, Claude plugin, and the Go orchestrator — is Apache-2.0 licensed at github.com/nightgauge/nightgauge. Everything runs locally with model and forge credentials you control: no account required, cloud connectivity off by default, telemetry opt-in. Signed release binaries and the Homebrew cask ship with the first public release; until then you can build from the reviewed source tree. An optional proprietary platform (team analytics, licensing, billing) integrates through public contracts.

A VS Code extension paired with a deterministic Go orchestrator and the Claude Agent SDK. Skills are portable across Claude, Copilot, Codex, Cursor, and Gemini — write once, run anywhere. Pipeline state lives in JSON handoff files between stages, never accumulated conversation history.

The whole reason the product exists: six guardrail layers sit alongside the pipeline. Cascade circuit breaker pauses on clustered failures with no auto-resume. 21 deterministic phase gates verify work before context handoff. Prompt-injection firewall blocks dangerous patterns at input and output. Audit trail captures every stage, retry, and fix.

Your code stays in your repos. Nightgauge runs locally in your VS Code extension and connects to your model provider directly with your own keys — and because the core is open source, you can verify that yourself. Cloud connectivity is off by default and telemetry is opt-in. No code is ever sent to our servers for training or analytics.

Two costs: AI tokens (paid to Anthropic at API rates — typically $0.20–$1.00 per issue depending on complexity, with the dashboard showing per-issue spend) and the Nightgauge core itself, which is free and open source (Apache-2.0) — you run it locally with your own keys. The optional hosted platform is in early access and not yet publicly priced. Performance modes set the cost/quality envelope — from Efficiency (≈ half cost) to Frontier — and cost-aware routing automatically prefers cheaper models when your own run history shows comparable results. Token budgets are configurable at workspace, per-stage, and per-pipeline-run levels with hard caps.

It's actually built for them. The Brownfield Dashboard scores readiness across 8 axes, scaffolds characterization tests before refactoring starts, and emits a phased modernization plan with confidence levels. Hybrid recommendations (refactor backend / rewrite frontend in parallel) are common output.

Three layers catch that. Phase gates verify each phase's promised artifacts before moving on. The RALPH self-healing loop diagnoses and retries failures within bounded retries. If something falls outside the auto-fix envelope, the pipeline escalates to a developer with the stage-exit diagnostic (last bash, stderr tail, classification) — not a vague stack trace.

GitLab support (CE, EE, and self-hosted) is in private preview behind the same forge abstraction that runs GitHub today. Slack, Discord, and Mattermost chat-ops are in beta — notifications work today, with Mattermost's inbound slash commands the furthest along. The roadmap section on the homepage tracks shipping vs preview status.

Nightgauge was created by Mark McCorkle and is built by Edibu, LLC — small, focused, engineering-led. The product itself is built using Nightgauge (dogfooding) — every dashboard you see has shipped real issues through itself.