NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #439 by repository stars
Latest stable
v0.120.23
Last release today
21 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Nearly every release is documented
notes for 58 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
4 years old
2616 releases · first in 2023
Drag across the ruler to compare, or tap a release to open it.
Tenants registered together shared one ticker phase, so every refresh checked out connections at the same instant.
Tenants registered together shared one ticker phase, so every refresh checked out connections at the same instant.
fix(frontend): resolve exchange token tenant from the request path (#…
fix(frontend): resolve exchange token tenant from the request path (#…
Allows configuration of a separate OLAP database for task and event sinks and aggs.
Allows configuration of a separate OLAP database for task and event sinks and aggs.
Your coding agent can read these notes before it upgrades. Set up the MCP server →
fix(scheduler): keep the scheduler partition id across a recreate ( #5 …
fix(scheduler): keep the scheduler partition id across a recreate (#5…
On smaller / constrained Postgres instances, the 60 second periodic jobs lead to a pretty massive connection spike which eat up a lot of resources tem
On smaller / constrained Postgres instances, the 60 second periodic jobs lead to a pretty massive connection spike which eat up a lot of resources temporarily, putting some pressure on regular queueing. This adds some jitter so these aren't all processed at once.
initial commit small refactor
initial commit
small refactor
refactor: expand sqlchelpers to limit iteration
feat: add reduce/3 enumerable
refactor: expand sqlchelpers to limit iteration
refactor: naming
chore: rm max timestamp helper
refactor: update queries to use more partition pruning
refactor: update v1_retry_queue_item_delete_function
chore: gen
chore: wiring
refactor: add a bunch more pruning
refactor: dynamic flush of durable event buffer
refactor: dynamic flush of durable event buffer
refactor: use mq_buffer_core instead of re-implementing from scratch
fix: greptile
fix: add cleanup
UI refinements and table instead of raw JSON
UI refinements and table instead of raw JSON
SDK README and provider doc
gen client again
fix(engine): resolve pool debugger callers through fairpool frames (#…
fix(engine): resolve pool debugger callers through fairpool frames (#…
fix: stream tests, deprecation warnings
chore: add pyright and ty to ci linters
chore: version, changelog
fix: get rid of any
chore: gen
fix: builtins imports
chore: set up linters for the examples and the tests
fix: raise on cancellations
fix: more test failures
fix: stream tests, deprecation warnings
chore: docs
chore: docs
chore: lint
fix: proper termination in hatchet-lite , reduce partition heartbeat…
fix: proper termination in hatchet-lite, reduce partition heartbeat…
pin to ubuntu 24.04 instead of latest
pin to ubuntu 24.04 instead of latest (#5562)
fix(engine): make the task activity gauge an index probe shared acros…
fix(engine): make the task activity gauge an index probe shared acros…
fix: read allocated resource counts from a per-tenant table
Recomputing the shard aggregation on every lookup is the expensive path.
Cron and schedule count changes take the workflow's latest version lock
before reading the rows they adjust, and schedule deletes lock the rows
that still count. Workflow delete and invalid-cron delete apply their
deltas. The rebuild holds an exclusive table lock across the delete and
insert. The schedule-fire trigger skips schedules on a deleted workflow.
The migration moves after the ones already on main.
A workflow run can only have one trigger row, so the delete-workflow count test was failing on the second insert.
Goose skips a migration whose version is older than ones already applied, so the November 5 timestamp would never run on databases that have moved past it.
Concurrent schedule inserts and fires queued behind each other: each took
an exclusive lock on the workflow's latest version and then updated the
tenant's count row, and the hourly rebuild held an exclusive table lock
across a full scan.
Cron, schedule, and webhook writes now append a row to
tenant_allocated_resource_count_delta in their own transaction. A count is
the tenant's count row plus its deltas, and a ticker job folds the deltas
into the count rows. Schedule writes take a key-share lock on the version,
so they wait only for a version registration or workflow delete. The
unfiltered read appends a correcting delta where the stored count differs
from the live rows, in place of the locked rebuild.
fix migration
fix: count a schedule once when its delete waits on a workflow delete
A schedule delete that waited on a workflow delete for the schedule's row
read the version and workflow from before the wait, so both subtracted the
schedule. The version share lock only prevented this when the two saw the
same latest version.
The delete now locks the schedule rows first and reads which of them still
count in a later statement. It takes no version lock, so a fire no longer
waits on a version registration.
The ids for a bulk delete come from the request, so the lock taken before counting is now filtered by tenant.
fix(db): bound engine pool dial and connect separately
fix(db): bound engine pool dial and connect separately (#5547)
fix(frontend): show pending invites on tenant 403 page
fix(frontend): show pending invites on tenant 403 page
fix(frontend): retry tenant route once access is granted from 403 page
Nothing published for this version
fix(engine): read the step completion entitlement on the caller's tra…
fix(engine): read the step completion entitlement on the caller's tra…
fix: one more source of dag crash looping
fix: one more source of dag crash looping (#5510)
fix: stranded matches chore: rm tests chore: clean up comments fix: one more comment
fix: stranded matches
chore: rm tests
chore: clean up comments
fix: one more comment
FlattenExternalIds derived its inserted_at bound from a CTE in the same statement, so partitions of v1_task, v1_dag and v1_dag_to_task could only be p
FlattenExternalIds derived its inserted_at bound from a CTE in the same
statement, so partitions of v1_task, v1_dag and v1_dag_to_task could only
be pruned at execution time, after every one of them had been locked.
Resolve the ids through v1_lookup_table first and pass the range to the
v1_task and DAG statements as plain parameters under a custom plan, so
the planner prunes before locking. The DAG join only runs when a lookup
row points at a DAG.
Add ResolveTaskExternalIds, which maps task external ids to their primary
keys from the lookup table alone behind an LRU cache, and use it in the
dispatcher's worker-status handler and the scheduler's dead-lettered
durable callback handler, which never needed the task row.
Nothing published for this version
feat: batch step triggers in operator
feat: batch step triggers in operator
fix: pr comments, naming
Merge branch 'main' into nafees/sign-sbom-images
Merge branch 'main' into nafees/sign-sbom-images
Merge branch 'main' into nafees/sign-sbom-images
Merge branch 'main' into nafees/sign-sbom-images
Nothing published for this version
Nothing published for this version
Nothing published for this version
fix: release concurrency topic leases when a tenant is handed off ( #5 …
fix: release concurrency topic leases when a tenant is handed off (#5…
fix(dispatcher): tolerate transient worker stream backpressure
fix(dispatcher): tolerate transient worker stream backpressure
fix(dispatcher): keep worker send tests out of synctest
go-deadlock's pooled timers abort a synctest bubble in the load jobs. Drop the skipped stall repro, fix the config table padding, and record that an in-flight send stays assigned.
feat(operator): run each tenant's DAG operator on every dispatcher (#…
feat(operator): run each tenant's DAG operator on every dispatcher (#…
Merge remote-tracking branch 'origin/main' into t3/dag-operator-replicas
Merge remote-tracking branch 'origin/main' into t3/dag-operator-replicas
Callers enqueue audit log entries in a pgoutbox topic in the core database, and the OLAP controller copies them into v1_audit_log. A request that reco
Callers enqueue audit log entries in a pgoutbox topic in the core database,
and the OLAP controller copies them into v1_audit_log. A request that records
audit logs no longer waits on the OLAP database. A batch that fails to copy
stays queued and is retried.
Nothing published for this version
Nothing published for this version
Nothing published for this version
feat(frontend): add a browser-local banana mode in Labs
Keep the theme, logo, and rain out of tenant entitlements so enabling it does not change the product for anyone else.
add tables fix nans
add tables
fix nans
feat: send cloud logins through control-plane MFA
Shards with a control-plane exchange client reject local login and cookie
auth, and a pending second factor is sent to /auth/mfa with an account
security page for managing it.
Keep recovery codes on screen until they are saved, and treat a closed passkey prompt as a cancellation.
Removing a method asks for a current factor, and logout reloads the login page so a cancelled session check cannot skip it.
Replacing recovery codes now asks first, and the no-tenants settings button navigates again.
Adding, removing, and replacing now share one confirmation form.
feat: add an MFA column to the organization members table
fix: recover MFA sign-in and authenticator setup after a failed step
A consumed magic link is not submitted again, a rejected authenticator code can be restarted, and an account with no organization leaves the verify page.
A missing control-plane route was failing password and magic-link login after the session already existed.
Login no longer calls the MFA status route, and the MFA page reads stable error codes.
Return the saved post-login path after the factor proof, ignore the
mfa_required response on password and magic-link forms, and wait to
load the current user until the control plane is known so the shard
does not replace the pending session cookie.
feat(durable-streams): count bytes published and read for billing ( #5 …
feat(durable-streams): count bytes published and read for billing (#5…
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
feat(tenants): let owners opt into Labs preview features
Durable streams, sandboxes, and serverless are tenant entitlements owners can toggle from settings. Turning one off asks for confirmation, and free plans and non-owners cannot change them.
deslop
greptile feedback
docs(api): note admin access and token restriction on Labs update
docs: document BoolFromMaybeBoolean
generate
fix sandbox operator entitlement propagation
v1_sandbox and related tables
some cleanup
specific kinds
UI
daytona
fix UI gate
update dialog
move TS SDK for sandbox
fix entitlement
rename migration
regen clients
fix changelog
fix gemfile
freestyle support
e2b and archil
README wording
default TTL
fix sandbox operator entitlement propagation
rename migration file
rename otel
operator required OAPI just like serverless API
fix lint
update readme
Nothing published for this version
Nothing published for this version
fix: release action links of a claimed operator's superseded worker (…
fix: release action links of a claimed operator's superseded worker (…
chore: drop unneeded indexes fix: migration version
chore: drop unneeded indexes
fix: migration version
Nothing published for this version
Adds a magic-link email template (Postmark alias magic-link, plus the matching SMTP template on the shared layout) and SendMagicLinkEmail to the email
Adds a magic-link email template (Postmark alias magic-link, plus the matching SMTP template on the shared layout) and SendMagicLinkEmail to the email service. The login page gets an Email me a login link button that opens the email form, and /auth/magic-link confirms the link with an explicit Sign in click so link scanners cannot spend the token.
fix: dont leak
fix(auth): show a dedicated state for invalid magic links
fix(auth): refresh current user after magic link confirm, gate on lookup, allow retry
feat(auth): disable magic link resend during cooldown
chore: rename all the migrations
chore: rename all the migrations
chore: give em names
chore: rename one more
Nothing published for this version
fix: migration ver fix: migration number chore: split migrations fix: migration name
fix: migration ver
fix: migration number
chore: split migrations
fix: migration name
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.