Kendr Optimizer v0.1.4 (pre-alpha)
Pre-releaseKendr Optimizer v0.1.4
This pre-alpha release adds a read-only terminal dashboard and strengthens the
public documentation and release evidence around the existing CLI. It does not
replace the frozen benchmark bundle with a fresh v0.1.4 optimizer evaluation.
Saffron terminal dashboard
kendr-opt dashboardopens the dashboard;kendr-opt tuiis a visible alias.- Five panels cover Quick start, Commands, Harnesses, Updates, and Trust. Use
Left/Right or Tab/BackTab to change panels, Up/Down orj/kto scroll,?
for help, andqor Escape to exit. The layout adapts to small terminals and
asks for a resize when there is not enough room to present panel content. - Running
kendr-optwithout a subcommand opens the dashboard only when both
standard input and standard output are terminals. In a non-interactive
context it prints long help and exits successfully. Asking explicitly for the
dashboard without a terminal fails with a clear error. - Alternate-screen and raw-terminal mode are confined to the dashboard and are
restored on exit.kendr-opt runand the hosted CLI are never wrapped by the
dashboard. - Human-only
setup,setup --list, runtime-error, and dashboard output uses
Kendr saffron (#E2712A) on Kendr ink (#2B2925), with deep saffron
(#B8551A) borders, when the relevant stream supports it. Labels remain
meaningful without color;NO_COLORselects a plain dashboard, while
TERM=dumbdisables the dashboard and selects plain output for other
commands. - JSON and other machine-readable output, update JSON,
--version, child
harness streams, and non-terminal service logs are never decorated.serve
logs are written to standard error.
The dashboard is an informational guide, not an interactive configurator. It
does not install or update anything, call a provider, modify configuration, or
perform network requests.
Reference and downloadable whitepaper
The repository now carries a standalone CLI reference
for command behavior, outputs, environment controls, and automation boundaries.
The generated publication view of the technical whitepaper is also available
directly from this release as
kendr-optimizer-verification-gated-token-reduction-whitepaper.pdf.
The v0.1.4 release contains 16 assets. SHA256SUMS covers the other 15 assets,
including the PDF, and publication checks require those checksums to agree with
GitHub's recorded SHA-256 digests. CI checks that the release PDF is byte-for-byte
identical to the repository's canonical generated PDF and that its embedded
source digest agrees with the Markdown whitepaper. The
Markdown remains the authoritative source; the PDF is a generated publication
view.
Update and compatibility continuity
The default installer records the preview update channel, so an eligible
v0.1.3 installation can discover this prerelease with kendr-opt update --check
and install it with kendr-opt update. The stable channel deliberately ignores
pre-alpha prereleases. The published v0.1.2 executable still predates the updater
and must first be bootstrapped with a later official installer.
Host compatibility remains bounded by the explicit upstream versions and
commits in the compatibility ledger. The current
ledger was last audited on 2026-08-07, and its recorded verification run is dated
2026-08-08 with 59 tests. This release does not turn those bounded observations
into a claim of universal or future host compatibility.
Security and evidence limits
Normal optimization remains local and the core crates remain network-free. The
dashboard performs no network action. An explicit or passive update check sends
only repository and release requests to GitHub; release assets are downloaded
only for an explicit installation. Kendr does not send prompts, tool results,
recovery data, provider settings or credentials, or traffic to Kendr.org.
GitHub and its asset delivery network can still observe the client's IP address,
request timing, and version-bearing user agent.
GitHub release immutability, SHA256SUMS, GitHub-recorded digests, and the PDF
source-identity check detect corruption or publication drift within the GitHub
trust boundary. They do not independently authenticate the publisher. The
release is not protected by a maintainer signature, Sigstore identity, or
operating-system code signature; compromise of the GitHub organization,
workflow, or initial upload remains a supply-chain risk.
Updater replacement still requires the eligible immutable release, exact asset
set, checksum, archive, candidate-smoke, receipt, anti-downgrade, and
post-install validation gates documented for v0.1.3. Backup-backed rollback is
not a journaled, power-loss-safe transaction. Package-manager-owned binaries
should continue to be updated through their package manager, and bundled
adapters are refreshed by the next kendr-opt setup or kendr-opt run.
The optimizer ranking and per-case results remain the frozen
v0.1.0-benchmark.5 evidence. They are not a fresh v0.1.4 benchmark and should
not be read as proof of broad model non-inferiority or guaranteed savings on a
new workload.