Docs / Roadmap
Roadmap
Direction of travel for Kraftverk — with honest status labels. Capabilities vs experimental work are called out deliberately.
Milestone 1 — complete
- Cargo workspace + docs
- Inspect, KraftBench, stats, Kraft Index
- SQLite experiments
- Safe optimize + rollback + recovery
- Core CLI surface
Expansive Platform Phase (0.2) — complete
- ≤10 crates (
system,optimizer,data,sdk,desktopconsolidation) - KraftBench v2, telemetry/noise model, optimizer goals/sessions/profiles/receipts
- HTML/JSON reports, doctor/capabilities/insights, desktop instrument UI
- Branding, Pages site, CI + release workflows
- AMD-exclusive hardware enforcement (
amd-only-v1)
What works vs experimental (0.2)
| Area | Status |
|---|---|
| AMD eligibility gate | Working |
| Inspect / doctor / capabilities | Working |
| KraftBench v2 + baseline | Working |
| Safe optimize + sessions | Working |
| Reports / history / explain | Working |
| Desktop instrument UI | Working |
| Privileged agent | Scaffold |
| GPU benches | Unsupported |
| Temp / power sensors | Unsupported |
Milestone 2 (planned)
- Privileged agent with authenticated IPC
- Power-plan / selected system tunables where reversible
- Richer telemetry sensors (where APIs exist)
- Additional search strategies (ε-greedy, Bayesian — still seedable)
- Optional Tauri packaging for the desktop instrument
Milestone 3+
- GPU backends with real kernels — never fake
- Cross-run regression dashboards (still local-first)
- Optional export formats for papers / CI
Explicit non-roadmap
- Placebo “cleaner” features
- Undocumented silent tweaks
- Cheating benchmarks to inflate marketing numbers