NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules · #957 by repository stars
Last release 6 days ago
02 Oct 2026
Ships fairly regularly
a new release about every 1 weeks
Some releases are documented
notes for 8 of 14 stable releases
Nothing withdrawn
no release was ever pulled
2 years old
326 releases · first in 2024
One column per month.
fs: new exporter options skip_ownership , skip_permissions , skip_times and skip_root_perms_and_time .
skip_ownership, skip_permissions, skip_times and skip_root_perms_and_time.C:\ entry name in the fs importer.WARNING: Starting on Friday September 18, authentication requests not done with the new authentication flow will be rejected.
Friendly reminder: if you're updating from plakar 1.1.4 or older, now PLAKAR_INSECURE_PLAINTEXT is required to open non-encrypted kloset repositories.
This release introduces a hardened authentication flow to our API server for registered users.
This release introduces a hardened authentication flow to our API server for registered users.
When running plakar login, the confirmation page on our server will now display a code that has to be typed in the terminal to ensure the person who is confirming authentication is also the person who has the terminal opened and prevent some phishing attacks.
WARNING: Starting on Friday September 18, authentication requests not done with the new authentication flow will be rejected.
Friendly reminder: if you're updating from plakar 1.1.4 or older, now PLAKAR_INSECURE_PLAINTEXT is required to open non-encrypted kloset repositories.
refuse to open an unencrypted Kloset store unless PLAKAR_INSECURE_PLAINTEXT is set in the environment
PLAKAR_INSECURE_PLAINTEXT is set in the environmentplakar pkg add -allow-unsigned opts outauth_token over plaintext HTTP now requires insecure=true, https->http redirects are rejected, plus assorted protocol hardening-token requires clients to present an Authorization Bearer token; requests are now loggedplakar pkg add / plakar pkg show: -devel to use the integration devel treeauth_token, insecure, timeout and tls_no_verify options, documented in plakar-store(1)/tmp) instead of a bare cancellationO_NONBLOCK so a FIFO no longer stalls a backupFull Changelog : v1.1.3...v1.1.4
Full Changelog: v1.1.3...v1.1.4
prune -group-by was being dropped when merged with a retention policy, causing per-day/per-minute caps to apply globally instead of per group. Now carries through correctly.restore: use a different name for the default restore directory by @omar-polo in #1792
plakar diag errors from doc by @brmzkw in #1896Note truncated.
This is the largest release we have ever published. v1.1.3 is the final, stable outcome of the v1.1.0-beta cycle that started in January, hardened through three release candidates and several thousand additional snapshots of real-world testing.
We jumped straight to v1.1.3. We cut a few quick follow-ups on the way out the door; v1.1.3 is the one you actually want. Several of the extra version bumps also reflect us finding our footing with a new integrations monorepo workflow.
Since v1.0.6 (roughly six months of work):
| Repository | Commits | Pull requests |
|---|---|---|
| plakar | 421 | 250 |
| kloset | 599 | 274 |
| integrations | 1,090 | 206 |
stdio renderer (old behaviour, unchanged) and a new tui renderer that provides a clean, live view during backup and restore.plakar backup /etc /home now works on a single source. Multi-source snapshots (spanning multiple independent data origins) are in progress for the next release.plakar mount can now target specific snapshots or individual directories, and can expose a mount over HTTP. Added -allow-others flag to pass fuse.AllowOther() at mount time.cached introduced. The agent process (present since v1.0.0, auto-managed since v1.0.4) is gone. It is replaced by cached, a lightweight background process responsible only for shared cache maintenance and locking. Commands now run directly in the CLI, shrinking the failure blast radius and unlocking features that were previously awkward.repair now takes an exclusive lock.cached runs correctly in the background on Windows and no longer trips over syslog setup.prune -group-by is implemented and documented; Kloset's locate also gained -group-by along with dataset and data-class filters.Measured with Korpus (1,000,000 items) on a 14-core Mac Mini with 64 GiB RAM and NVMe storage, compared to v1.0.6:
| Operation | v1.0.6 | v1.1.3 | Change |
|---|---|---|---|
| Backup | ~3 minutes | ~2 minutes | −33% |
| Sync | ~5 minutes | ~4 minutes | −20% |
| Restore | ~60 minutes | ~3 minutes | −95% |
| Check | ~1 minute | ~1 minute | — |
The restore improvement comes from a better algorithm, smarter parallelism, improved prefetch utilization, and removal of unnecessary system calls.
Peak RAM, same test environment:
| Operation | v1.0.6 | v1.1.3 | Change |
|---|---|---|---|
| Backup | ~3.0 GiB | ~1.3 GiB | −43% |
| Sync | ~3.6 GiB | ~1.7 GiB | −52% |
| Restore | ~2.3 GiB | ~800 MiB | −66% |
| Check | ~1.3 GiB | ~800 MiB | −40% |
Sources: fixed a gRPC-level memory leak affecting long-running backups over SFTP and S3; reworked the caching subsystem; default to spilling temporary data to disk rather than holding it in RAM.
Default on-disk cache footprint (1,000,000 items): reduced from ~4 GiB to ~1.8 GiB (−55%) by removing the on-disk VFS cache in favour of store queries. The old behaviour is available via flag for environments where bandwidth matters more than disk.
plakar mount handles direct access to filesystem paths without browsing from the root.restore: use a different name for the default restore directory by @omar-polo in #1792
plakar diag errors from doc by @brmzkw in #1896Note truncated.
Full Changelog : v1.1.0...v1.1.1
Full Changelog: v1.1.0...v1.1.1
Your coding agent can read these notes before it upgrades. Set up the MCP server →