Local-first deployment intelligence
LaunchGuard turns a local checkout or public GitHub repository into a deterministic audit, a reviewable execution plan, and an approval-gated deployment pull request.
$ launchguard audit ./project
→ acquiring repository evidence
✓ project detected React / Vite
✓ manifests parsed package.json · lockfile
✓ trivy complete db: recorded
✓ osv-scanner complete 0 critical
◆ local runtime absent verify track unavailable
soft block · review required
Evidence before action
Classify frameworks, manifests, services, and build evidence without executing repository code.
Run trusted scanners and merge their output into stable, versioned security records.
Generate a content-addressed execution plan with commands, resources, and network policy.
Require approval, generate provider configuration, and record every override and degradation.
Capability-aware by design
Detect, scan, plan, generate configuration, and open a pull request. Missing optional tools become typed degradations—not dead ends.
When a container runtime is available, LaunchGuard can build, test, isolate, and health-check the project before its configuration leaves your machine.
Publication gating
LaunchGuard separates hard blocks, explicit overrides, and clear decisions so users can understand the difference between an unsafe repository and an unverified one.
Read the security model ↗Publication is refused. The block cannot be overridden.
×Proceed only with an explicit, recorded override.
!The reviewed publication step may continue.
✓Auditable all the way down
Profiles, findings, plans, readiness assessments, generated files, and publication decisions use explicit schemas. Digests make repeated runs comparable and reviewable.
Measured, not marketed
40/40 supported fixtures classified; seven safety fixtures failed closed.
40/40 schema-valid plans with reproducible digests in the measured corpus.
Prebuilt binaries, doctor checks, and verified scanner provisioning.
Implemented capabilities are being validated on real projects and guided flows.
Evaluation numbers describe the repository’s documented test corpus, not a claim that LaunchGuard proves software secure or production-ready.
Local and free by default
The core workflow is designed to run on your machine with user-supplied tools and provider accounts. Costs outside the software—hardware, domains, signing, and provider usage beyond free tiers—remain visible rather than hidden behind “free forever” language.
Open source · MIT