Skip to content

Releases

kronik 0.5.0

Choose a tag to compare

@github-actions github-actions released this 15 Nov 02:52

Kronik v0.5.0 — "Isolation, standards and web" (M20–M26 of the build brief).

The largest release so far: long-lived connections, streaming bodies and app blob
storage are new capabilities, and the foreign-actor sandbox gains two new levels in
front of its manager, the nonroot daemon and the microVM.

Highlights

  • Long-lived connections (M24). WEBSOCKET and SSE routes; the owner actor
    accepts, sends and closes in recorded turns; send-now sends frames during a turn;
    replay and the diff compare what a fork sends; the API, CLI and MCP list, inspect and
    close connections; the four SDKs and the chat-ts example carry them.
    The actor WIT becomes kronik:actor@0.1.1 — additive, 0.1.0 components keep running.
  • Streaming bodies (M25). Request and response bodies above a kind's threshold
    become blobs: uploads and answers stream with constant memory (peak growth of a
    300 MB upload: +1.6 MiB; of a 1 GB response: +1.4 MiB on the test machine), wasi:http
    bodies go through the store, previews are capped and paced, and a blob store I/O
    failure stops the node like a log failure instead of losing a turn.
  • App blob storage (M26). Actors put, retain and release blobs, hand them to
    each other in messages, timers, calls and replies, and hold them against a per-actor
    quota. A collector deletes what nothing keeps, with dry-run reasons; a fork's replays
    keep their blobs; erasure deletes a subject's files locally and in the bucket, and
    privacy reseal seals what was written in the clear. Blobs are always on.
  • External secret backends (M23 B2–B6). Secret names linked to OpenBao or Vault, AWS
    Secrets Manager, GCP Secret Manager and Azure Key Vault, read through a cache with
    outage handling, and a PKCS#11 helper.
  • Isolation, opt-in. The nonroot daemon (M21, serve --init --rootless) and the
    microVM level (M20, isolation = "microvm", accepted in this release) — both proven
    on disposable hosted runners in the manual isolation workflow. gVisor remains the
    default and the root service stays the default.
  • Replay, forks and the upgrade check reach every point of the retention window;
    inputs an erasure made unreadable are reported unverifiable rather than divergent.
  • Durability. A node now makes durable whatever a dead process left in the page
    cache before it builds on it (segment tail, footer, segment name), a snapshot is
    durable before its marker, a reseal marks a segment durably before replacing it and is
    finished at the next start if it stopped in between, and a segment rewrite fails when
    its rename is not durable.
  • Upgrades on small machines. kronik upgrade check sized its Wasm instance pool by
    its workers, so the first replayed call ran out of instances and the check reported a
    divergence; on a one-CPU machine every window with a call failed and the start gate
    refused the new engine. Fixed, with headroom for nested calls everywhere a replay runs.
  • Bug fixes found by the team's own reviews: a guest could write a call reply the
    next start refused (the node then could not start), an accepted upload could be
    collected before its turn ran, rollback-by-replay left blobs in the fork, a preview
    upload of a new subject stayed plaintext in privacy mode, a turn's decision about
    another actor's connection made replay depend on timing, an erased subject's blobs
    kept counting against a quota, and a authentication flake on restart after a reseal.
    The CHANGELOG has each with its release evidence.

Upgrading

The engine is now kronik-0.5.0+wasmtime-50.0.1 (was kronik-0.4.1+wasmtime-50.0.1).
Data written by any earlier release needs a passing kronik upgrade check with this
binary before the first start; without one the start stops with
ENGINE_VERSION_MISMATCH, and a passing start appends one engine.upgrade record.

Going back to v0.4.1 needs its own check, and works only until the log holds a record
v0.4.1 cannot read — such as a connection's turn — so take a backup of the data
directory and a copy of the shipping bucket's prefix before the upgrade if you may
want to return: v0.5.0 keeps shipping into the same prefix, and v0.4.1's restore
refuses a bucket that v0.5.0 has written to. The operations guide's "Upgrading from
v0.4.1" has the steps, and says what to do about erasures that happened after the copy
(the copy keeps that time's keystore backups, so going back brings back the ability to
read what was erased since: erase again, then delete the copy).

Configuration files keep working. [blobs] takes only its own keys; there is no
enabled switch — blob storage is always on. A log format change is additive, but
earlier releases cannot read a log that holds the new records.

Install

curl -fsSL https://github.com/G-man3207/Kronik/releases/download/v0.5.0/install.sh | sh

Both archives are statically linked musl builds, x86_64 and aarch64, reproduced by two
independent builds in the release workflow (reproducibility.json), attested through
GitHub OIDC/Sigstore, and shipped with CycloneDX SBOMs and the original notices. See
docs/guides/install.md for verification without curl | sh.

Known limits

  • The microVM level and the nonroot daemon are proven on GitHub-hosted virtual machines
    with nested KVM, not on bare metal, aarch64 or other distributions.
  • WASI 0.3 components are still upstream-blocked (M22 criterion 3).
  • The cluster track (M16–M19), the WASI 0.3 remainder, fuzzing and the UI/fleet tracks
    are not in this release.
  • A call whose callee answers with an erased subject's tombstone, or with a body sealed
    under a key this node no longer holds, leaves the caller's turn open and retried at
    every start until the data is readable again or the turn is closed. Erased answers are
    recorded as a SUBJECT_ERASED call error, but the open-turn retry itself is unbounded
    (see the CHANGELOG and the issues).

Kronik v0.4.1

Choose a tag to compare

@github-actions github-actions released this 08 Nov 17:17

Kronik v0.4.1

Kronik v0.4.1 is a patch release for v0.4.0. It fixes a security issue in kronik restore and the defects found since v0.4.0 in restart, retention, audit export, secret handling and the upgrade check. No log format, WIT or breaking API/CLI change comes with it. The engine id is not: run kronik upgrade check with this binary before its first start on existing data.

Fixed

  • A restart could stop at the log's start. Once [retention] history (7 days by default) had passed an SQL actor instance that had not changed its SQL, the next restart, whatever its cause, stopped with STORAGE_CORRUPT: state is at lsn N, but the log starts at M (compacted), and the product could not repair it in place. Releases v0.1.0 through v0.4.0 were affected. Retention now raises such databases before deleting; a start on an affected directory installs its newest snapshot; a restore from a bucket in that state starts; and kronik upgrade check reports the directory the start would refuse.
  • Restart time grew with the log. A start read the log twice from its first record to certify the secret inventory, at a cost that grew with the active segment (24 s at the 326th cycle of the nightly crash test before). It now certifies from the coverage watermark. A first start also no longer logs watermark write: rename: No such file or directory.
  • A partial upload of a log segment stayed in the bucket for good. When retention deleted a shipper's active segment before a later round uploaded it as a closed one (a node that could not ship for longer than [retention] history), nothing removed the object, and a node restored from the bucket reported a log starting at it, with a hole of LSNs under its first real segment. The upload now removes it; objects an older release left are not found.
  • Reads that scrub answered 503 right after the node reported healthy. The first pass of the secret cycle rebuilt the certificate the start had just made.
  • Erasing a data subject left those reads at 503, also after a restart. The scan for secret references took a sealed request whose key was destroyed on purpose for a loss of coverage; another unopenable request still refuses it.
  • An operator's --reason could be replaced for good by [withheld: secrets could not be loaded] in the append-only log and every audit export, when a mutation began while the secret inventory was being rebuilt (8 of 160 runs of the mutation test with four processes in parallel; also in v0.4.0). A change now wakes the rebuild only if its own recertification failed, and a mutation waits for a rebuild in flight, at most [secrets] fetch_timeout plus 5 s, before its reason is scrubbed (10 to 24 ms in that window, nothing otherwise). A wait that runs out records the placeholder, as does a mutation while a backend is down and the rebuilds fail. Reads in the few milliseconds between a configuration change and its recertification can still answer 503.
  • The audit exporter stalled after a secret-inventory rebuild (kronik export status showed UNAVAILABLE for up to five minutes): a new certificate did not retire the one it displaced, so a batch was refused at admission and retried. No line is lost or sent twice.
  • The audit export, traces and event webhooks could stall for a minute. A configuration commit that landed between the capture of a log reader and its read made the read fail with UNAVAILABLE, and the follower paused 60 s instead of capturing a new reader: kronik export status showed UNAVAILABLE, a frozen cursor and a growing lag; nothing was lost or exported twice. The window is now one second.
  • A turn whose effect could not get its secret from an external backend left an open turn. The caller got 503, and the next start ran the turn again with nobody waiting (two executions for a retrying client; under an unsafe grant EFFECT_OUTCOME_UNKNOWN for a request never sent). The effect now fails with SECRET_UNAVAILABLE inside the turn, recorded before the guest sees it, and nothing is sent.
  • A provider that refused a secret ended the turn with SECRET_VERSION_GONE. A 401 or 403, an unusable answer, or a link to a backend no longer configured was taken for a lost recorded version; it now gets the recorded error of an outage.
  • A fork replay stopped at an effect whose secret had never resolved, and secrets.reveal in a fork answered the placeholder where production had failed (other records with an unresolved reference are still refused).
  • Under CPU contention a fork replay or propose with a model could answer a spurious 503 SECRET_UNAVAILABLE and ask the paid model again. Two threads of one replay collided on the secret cache's mutex, and the publication of a model answer was denied instead of retried (7 of 480 runs of three tests with four processes in parallel; also in v0.4.0; fail-closed, nothing written wrongly). It now retries for about a millisecond; a certificate that really is retired is still denied at once.
  • A secret mutation answered 503 "already committed" at once, so a retrying client created another version; it gave up after one failed inline refresh.
  • A resend racing a rollback via replay could run a second time (a window of about 100 microseconds); it no longer does.
  • A held resend whose dead letter a rollback removed ran again instead of answering DEAD_LETTER_NOT_FOUND.
  • The engine.upgrade record kept the data directory's path in the log. The audit event said engine upgrade checked by <absolute path>, in the append-only log and every audit export; new records name upgrade/<engine>.json relative to the data directory. Records written by v0.1.0, v0.3.0 and v0.4.0 keep the old text, and so do exports already made.
  • The daemon's log carried ANSI colours into a redirected, piped or journaled log. Escape sequences sat between a field's name, its =, and its value. The log is now coloured only when standard error is a terminal; NO_COLOR is still honoured on one.

Security: kronik restore trusted names and code objects from the snapshot bucket

A restore used the manifest's paths and the components' names as given, checked only against a hash from the same manifest, and installed any file under code/, including precompiled artefacts that the daemon loads as its own. It now accepts only plain relative paths below the snapshot directory (and no SQLite sidecar files) and only components named by the BLAKE3 hash of their content; it installs no precompiled artefact and, when it installs a snapshot, removes the ones the data directory holds. A bucket that breaks a rule is refused before anything is written.

The fix is three commits, in restore order:

  • kronik-ship: refuse manifest paths outside the snapshot directory
  • kronik-ship: install only components named by their hash
  • kronik-ship: refuse sidecar files in a snapshot manifest

Affected: >= 0.1.0, < 0.4.1 (v0.1.0 through v0.4.0). Patched: 0.4.1. The advisory is linked at the end of these notes; its Residual risk section is the part to read before relying on a restore from a bucket you do not control.

Upgrade notes

  • The engine id changed to kronik-0.4.1+wasmtime-50.0.1. Data written by any earlier release needs a passing kronik upgrade check with the v0.4.1 binary before the first start with it; without one the start stops with ENGINE_VERSION_MISMATCH before it serves anything. A passing start appends one engine.upgrade record. Going back to an older binary needs the reverse check.
  • Run the check against the configuration kronik serve uses (the same data directory and key source). The check reads the data directory, its privacy keystore and the key source; a KMS-backed check needs the provider and its credentials. An empty diff exits 0; a non-empty diff exits 1 with ENGINE_UPGRADE_DIVERGED.
  • The first start reads the retained log once, because v0.4.0's secret-coverage watermark could lie ahead of what its scan read; a restore causes the same scan. In testing 136,000 records took 1.8 to 3.0 s.
  • A check that diverges because a subject was erased after the newest turn is not a behaviour change. The erased turn's sealed request can no longer be opened, so a replay produces a failed turn where the recorded turn succeeded. If this has happened: make one new turn after the erasure, then run the check with --window shorter than the time since that erasure (for example --window 1m), and let that report be the last one written — serve reads the last report, not the newest check. The fork point is chosen from the root inputs, not from snapshots: snapshot create does not move it, and on an idle node no window value ages the erased turn out. The erased turn is unverifiable rather than different; v0.5.0 will report it as such. kronik upgrade check now names this reason and the way out when it happens.
  • A restore from a bucket that an older binary shipped also needs the check before the restored directory's first start.
  • Take a backup before upgrading, and keep it as long as a rollback may be needed. Once v0.4.1 has served v0.3.0 data, a v0.3.0 binary stops on it with STORAGE_CORRUPT: record at lsn N: unknown record kind 45 — a newer log, not corruption. A v0.4.0 binary accepts data that v0.4.1 has served, after its own reverse kronik upgrade check.
  • If this data directory was last restored by an earlier release from a bucket whose writers you did not fully control, delete <data-dir>/code/*.cwasm before the first start, and before running kronik upgrade check with it — see the advisory's Workarounds.
  • No on-disk format change: wit/ and proto/ are identical to v0.4.0, no SQL schema statement changed, no configuration key or dependency changed, and components built for v0.4.0 run unchanged. On disk there is a derived secret-coverage.json (schema 1 to 2), a cache, and a larger applied LSN in idle actor datab...
Read more

Kronik v0.4.0

Choose a tag to compare

@github-actions github-actions released this 03 Nov 21:35

Kronik v0.4.0

Kronik v0.4.0 adds to v0.1.0 and v0.3.0: standard HTTP fetch for actors through
wasi:http (M22), OpenTelemetry tracing with a complete, verifiable audit
export (M28), and the first slice of the KMS master-key lifecycle (M23 B1) —
OpenBao, AWS, GCP and Azure wrapping, with kronik keys wrap|rewrap|status and
the optional kronik-keyhelper distribution. It also moves the engine to
Wasmtime 50.0.1 (RUSTSEC-2026-0420 through 0424), replaces the secret
catalogue's clear-and-rebuild with atomic certificate publication under
crash injection, paces effect rate windows uniformly, and carries the
build and test repair work. The binaries report kronik 0.4.0,
with engine kronik-0.4.0+wasmtime-50.0.1.

Highlights

  • Standard fetch through wasi:http (M22). A component may import
    wasi:http/types and outgoing-handler 0.2.x. Each request is an
    http.fetch effect: recorded with the turn, replayed exactly, matched in
    forks and answerable by simulators. TypeScript SDK handlers may be async
    and use ordinary fetch, with {{secret:NAME}} header templates still
    resolved by the host so secrets never reach the guest or the log.
    examples/uptime-ts shows the pattern. M22's criterion 3 (WASI 0.3
    components) stays open: Wasmtime 50 marks p3 experimental (ADR 0070).
  • OpenTelemetry and audit export (M28). A call through three actors is one
    connected trace in an ordinary OTLP backend; an incoming traceparent links
    to the root. The exported audit log is complete against the log's records and
    kronik audit verify checks it, against S3 sinks included. Export cursors
    are durable; export failures stall the cursor without losing records.
  • KMS-backed master key (M23 B1). OpenBao transit, AWS KMS, GCP Cloud KMS
    and Azure Key Vault can wrap the 32-byte master seed; kronik keys wrap|rewrap|status and kronik-keyhelper (separate optional x86_64 GNU
    archive) ship with operator docs and generated schemas. B1 only: M23's later
    slices (B2–B6, PKCS#11 lifecycle end to end) are not in this release, and
    real-cloud IAM remains untested — provider proofs ran against pinned
    emulators.
  • Engine security patch. The five maintained Wasmtime dependencies are
    exactly 50.0.1, the first patched 50.x for RUSTSEC-2026-0420 through 0424.
    Features, deterministic controls and the deny policy are unchanged; the
    linked engine version remains part of every recorded identity and
    precompiled-cache path, so the upgrade procedure below applies.
  • Repaired and tidied. The workspace build and test suite are restored,
    generated docs are current, and the repository layout is tidied (the agent
    guide moved to docs/AGENT_GUIDE.md; stray tool folders removed).

Scope note: v0.4.0 does not contain the cluster

The original build brief mapped v0.4.0 to the cluster track (M16–M19). That
track is parked: it has a reviewed design pinned on a branch, but no
implementation on main, and it is not in this release. v0.4.0 instead contains
M22, M28 and M23 B1, which landed on main since v0.3.0. The park decision is
recorded on the project status page.

Crash-path hardening since v0.3.0

The v0.3.0 secret catalogue retired certification on every secret record and
rebuilt it from scratch; under crash injection that left windows where the
whole catalogue answered fail-closed 503 SECRET_UNAVAILABLE, and a crash
during a secret write could leave it uncertifiable. This release replaces
that design:

  • The certificate covers an applied log prefix and only the names a batch
    touched go stale, so the previous certificate keeps serving while the next
    one builds.
  • Builds publish atomically (shadow build, strict CAS on the applied head
    after a drain) instead of clear-then-rebuild, and a scan-frontier
    checkpoint makes a cold full scan resumable across restarts.
  • Rate windows are paced evenly instead of bursting at period boundaries, so
    every sliding window holds the granted rate across crash and clock-jump
    clusters.

The applier now degrades instead of wedging on availability-class
publication errors, abandoned applies restore their revision sentinel, and
audit reasons are scrubbed against the exact inventory the writer observed
(with startup certification retried, so the first records' reasons are not
lost to a transient applier lag).

DST now passes every workload (counter, effects, webhook, rebuild) in both
debug and release profiles, and the chaos test passes 100 kill-restart
cycles with an empty identity replay. No known deterministic DST failures
remain in this release.

Upgrading

Upgrading from a v0.3.0 (or older) data directory: 0.4.0 is a new engine
(kronik-0.4.0+wasmtime-50.0.1). Before restarting the service on existing
data, run the check with the new binary — the running daemon may stay up:

sudo kronik upgrade check --window 6h --config /etc/kronik/kronik.toml

Without a passing check the start stops with ENGINE_VERSION_MISMATCH and
changes nothing. The first start records the upgrade and recompiles every
deployed component, so it takes longer than usual; old
*.<old engine>.cwasm files in <data-dir>/code/ can be deleted afterwards.
A KMS-backed data directory needs the provider and its credentials for the
check, even while the old daemon keeps running. Going back to the old binary is
the same procedure the other way; a restore from a bucket an older binary
shipped also needs the check first.

Install

On Linux, x86_64 or aarch64, as root:

curl -fsSL https://github.com/G-man3207/Kronik/releases/download/v0.4.0/install.sh | sh
kronik doctor

The installer refuses unsigned archives: it checks each archive's Sigstore
keyless attestation — produced by this release's GitHub Actions workflow with
GitHub OIDC — against the exact workflow, tag and repository before anything is
read or executed. It bootstraps a source-hash-pinned official GitHub CLI for
verification; no GitHub account or token is needed, but fresh verification
needs HTTPS access to the verifier and the Sigstore TUF endpoints. A mirror
changes only the download location:

curl -fsSL "<trusted-installer-url>/install.sh" | KRONIK_BASE_URL="<mirror-url>" sh

Files

  • kronik-x86_64-unknown-linux-musl.tar.gz,
    kronik-aarch64-unknown-linux-musl.tar.gz: static stripped musl binaries
    with LICENSE, NOTICE, THIRD_PARTY_LICENSES.txt and a target-specific
    CycloneDX kronik.cdx.json
  • kronik-<target>.tar.gz.sha256: per-archive checksum (not authentication)
  • kronik-<target>.tar.gz.sigstore.json: keyless attestation bundle
  • install.sh: the installer, set to download from this release
  • SHA256SUMS: checksums of the unsigned build outputs
  • reproducibility.json: two independent builds compared byte for byte
    (ELF, archive payloads, distribution); differences fail the release
  • kronik-keyhelper-x86_64-unknown-linux-gnu.tar.gz: optional x86_64 GNU
    PKCS#11 helper, only when selected at release time

Full changelog

90e0c921 test: settle the fleet entropy baseline against the refresh cycle
8a14338c test: await the documented secret-coverage window in the privacy suite
c1c8138e test: assert the OTLP metric mapping against a shared registry
763b249a ci: prepare the pinned OIDC test runtime for the test step
e1b22a57 sched: keep audit demand visible while the inventory certifies
73a2c877 style: unused binding after the sentinel fix
66fbd034 sched: certify startup before audit writes and fix the sentinel
61741700 test: bound the process-exchange thread grace by attempts, not a clock
9b3df37e test: give the process-exchange thread exit a bounded grace
81f44484 test: settle fleet startup records before the no-history dry-run check
95d22eee style: cargo fmt
0257c491 sched: atomic certificate publication, restart custody and rate pacing
d889c49f sched: add staging-swap certificate publication and rate pacing
007f80bf release: report 0.4.0 and the Wasmtime 50.0.1 engine
d0596a46 api: allow the fleet control fixture's harness-shaped result
24432af5 test: compare compatibility across revisions by durable path
fef25337 privacy: the secret store identity the coverage watermark pins
3a31c860 docs: cover the current CLI, admin surface and key helper errors
f81009e3 ci: run the loadgen test in the examples step
3ff31ea7 test: restore lost coverage, bound the load-sensitive harnesses
db7e0a4b clippy: satisfy the workspace lint gate
f2735076 api: conform the oidc route dispatch to the segs convention
a1415334 testhost: a file-backed secret store and its original ownership
74621b1d api: committed mutations answer success with withheld reasons
6ba603ab sched: durable secret-coverage watermark, per-read probe, honest stalls
7fb823ee chore: tidy repository layout
c5c8aac6 docs: correct status claims and add the build-brief audit
f147d6a3 build: restore the workspace build and test suite
c798e4fc docs: add ADRs, runbooks and reference schemas
58644b12 chore: preserve the landed source and fixtures
425cc3fd ci: run bounded performance comparator self-tests
88ed2aed release: require verified keyless archives and reproducible builds
d9b946f2 audit: harden export confidentiality and durable publication
d01a8814 CI: check required selectors and actual notice boundaries
ecbfb85d release: share exact normal and supplied Rust notice collection
38ae48e0 CI: bind raw approval to the original gated artifact
cbe886a8 CI: prepare the real persisted-proposal M10 proof
2cdae06c release: verify build provenance and cluster CI boundaries
090c49ec install: preserve executables on failed replacement writes
fd0e2257 release: ship licensed archives and verify their consumers
65e848c6 docs: point custody proof at extracted release archives
46597ee5 runtime: patch Wasmtime and replay production upgrade history
5f835c85 ci: check every step and require pinned actionlint
b99a9ac0 Kronik M23 B1: KMS-wrapped master key lifecyc...
Read more

Kronik v0.3.0

Choose a tag to compare

@G-man3207 G-man3207 released this 29 Oct 13:18

Kronik v0.3.0. It adds to v0.1.0:

  • M10, remote control from CI, except its criterion 4 (a run in a public example repository). That missing criterion is why there is no v0.2.0.
  • M11, guard after promote and preview forks.
  • M12, effect simulators.
  • M13, rollback via replay.
  • M14, foreign actors in forks.
  • M15, the Python and Go SDKs.
  • An upgrade fix (d9a6e54): kronik upgrade check opens sealed values with the data directory's keystore and master key.

The binaries report kronik 0.3.0, with engine kronik-0.3.0+wasmtime-50.0.0.

Install

On Linux, x86_64 or aarch64, as root:

curl -fsSL https://github.com/G-man3207/Kronik/releases/download/v0.3.0/install.sh | sh
kronik doctor

To always get the latest release:

curl -fsSL https://github.com/G-man3207/Kronik/releases/latest/download/install.sh | sh

install.sh downloads the release's static binary for the machine and checks it against SHA256SUMS. It also offers to install gVisor's runsc, which only foreign actors need.

Upgrading

Upgrading from a 0.1.0-dev build: 0.3.0 is a new engine (kronik-0.3.0+wasmtime-50.0.0). Before restarting the service on existing data, run sudo kronik upgrade check --window 6h --config /etc/kronik/kronik.toml with the new binary (the running daemon may stay up). Without it the start stops with ENGINE_VERSION_MISMATCH and changes nothing. The first start compiles every component again; old *.kronik-0.1.0-dev+wasmtime-50.0.0.cwasm files in /code can be deleted afterwards. upgrade check now opens sealed values with the data directory's keystore and the configured master key (new --master-key-file); earlier binaries could not check data with privacy mode on.

Once a node has run 0.3.0, a 0.1.0-dev binary can't take its data back with privacy mode on, because the old binary's upgrade check lacks the fix.

Files

  • kronik-x86_64-unknown-linux-musl, kronik-aarch64-unknown-linux-musl: static binaries (musl)
  • kronik.cdx.json: CycloneDX SBOM of the binary's crates
  • install.sh: the installer, set to download from this release
  • SHA256SUMS

Evidence and known gaps

Project status

Kronik v0.1.0

Choose a tag to compare

@G-man3207 G-man3207 released this 29 Oct 11:04

Kronik v0.1.0: milestones M0 to M9. That covers durable turns on one node, effects and capabilities, fork, replay and behavior diff, the TypeScript SDK, policy, API and MCP, operations on a VPS, foreign actors in gVisor, privacy mode, and system proposals.

Version string. These binaries report 0.1.0-dev (kronik --version, engine kronik-0.1.0-dev+wasmtime-50.0.0), because the crate version was never bumped at this commit. Later releases carry their real version.

Install

On Linux, x86_64 or aarch64, as root:

curl -fsSL https://github.com/G-man3207/Kronik/releases/download/v0.1.0/install.sh | sh
kronik doctor

install.sh downloads this release's static binary for the machine and checks it against SHA256SUMS. It also offers to install gVisor's runsc, which only foreign actors need.

Files

  • kronik-x86_64-unknown-linux-musl, kronik-aarch64-unknown-linux-musl: static binaries (musl)
  • kronik.cdx.json: CycloneDX SBOM of the binary's crates
  • install.sh: the installer, set to download from this release
  • SHA256SUMS

Evidence and known gaps

Project status