TypeScript SDK Changelog

v1.37.0 - 2026-10-07

Added

• Adds RetryAfterError, which allows for custom retry strategies.

Fixed

• Duration strings now accept fractional units, e.g. '0.5s', which the Duration type already allowed but failed to parse at runtime.

v1.36.0 - 2026-10-06

Added

• hatchet.streams.publish accepts payloads up to 64 MiB.
• Added hatchet.streams.topicMetadata(topic, { namespace }), which returns a topic's tenant, message count, latest cursor and last publish time, counting only messages within the tenant's retention.

v1.35.1 - 2026-10-05

Fixed

• A stopping worker keeps its action stream open until its running tasks finish, and only then closes it and unsubscribes from the engine. Since 1.34.1 it closed the stream before waiting for those tasks, so the engine rejected every heartbeat sent in the meantime and the worker logged Heartbeat rejected: worker stream is not active every four seconds until shutdown completed. Task cancellations now also reach tasks that are still running during shutdown.

v1.35.0 - 2026-10-05

Added

• Added pause and unpause methods to the workflows client (hatchet.workflows.pause(workflow, opts)) and to workflow declarations (workflow.pause(opts)). opts requires a queueTTL for how long runs stay queued while the workflow is paused, and optionally takes the behavior (QUEUE or DROP) for cron and scheduled runs triggered while paused, both defaulting to QUEUE. The WorkflowPauseScheduledCronRunQueueBehavior enum is now exported from the package root.

v1.34.1 - 2026-10-05

Fixed

• Durable tasks that fan out to thousands of children no longer flood the engine with worker status requests. The durable listener sent one worker status request per awaited child, each listing every pending child, which was quadratic in the fan-out size; it now sends one request per event loop iteration listing only the newly awaited children.
• The periodic worker status request lists only children that have been pending for more than two seconds, is sent every five seconds instead of every second, and is split into requests of at most 10,000 entries so it stays under the gRPC message size limit.
• Re-delivered child completions are matched against pending callbacks without scanning the pending queue.
• Children spawned from a durable task over the gRPC durable listener are sent to the engine in requests of at most 100 children, instead of one request for the whole fan-out.
• Concurrent signals and worker.stop() now share one shutdown operation, keeping the durable listener open until waiting runs finish eviction. A worker stopped while it is still starting no longer registers a listener or leaves a health server running afterwards.
• Concurrent eviction requests for the same task invocation now share their acknowledgement, including timeout and disconnection errors, instead of leaving one caller to time out. The eviction timeout timer is cleared once the acknowledgement arrives.
• A stopping worker keeps sending heartbeats until its running tasks finish, and only then unsubscribes from the engine. It used to stop the heartbeat and unsubscribe before waiting for those tasks, so the engine could treat the worker as dead and reassign tasks that were about to complete.

v1.34.0 - 2026-10-02

Added

• Added hatchet.streams, a client for durable streams. In preview, and subject to change.

[1.34.0-alpha.2] - 2026-09-25

Added

• The @hatchet-dev/typescript-sdk/core entry now exports HatchetCore (also as Hatchet), the application client for runtimes without gRPC: Cloudflare Workers, Vercel Functions, Deno, Bun, browsers and Node. It issues unary Connect calls over the runtime's fetch through createFetchTransport (binary protobuf bodies, HTTP/1.1 or HTTP/2 as the runtime negotiates) and is configured from an explicit object only: { token, serverUrl?, hostPort?, tls?: { strategy: 'none' | 'tls', serverName? }, namespace?, transport?, fetch?, maxReceiveMessageBytes?, maxSendMessageBytes?, retrier?, logger?, logLevel? }. A serverUrl must be an absolute http:// or https:// URL with no username, password, query string or fragment, and its scheme must agree with tls: http:// is refused unless tls is { strategy: 'none' }, since the token travels on every call, and https:// is refused with 'none'; hostPort takes its scheme from tls as the Node client does. With only a token, the engine address comes from the token's grpc_broadcast_address claim; no other claim is needed, and a token without it fails at construction with an error naming the claim and the serverUrl/hostPort fields to set instead. A caller-supplied transport owns its base URL, auth and limits, so no address is resolved or validated for it. It owns run, runNoWait, runMany and runManyNoWait (over TriggerWorkflow and BulkTriggerWorkflow, with the Node client's options and namespacing, accepting the declarations from /edge; a bulk trigger that fails after one of its batches was accepted rejects with BulkTriggerPartialError, whose refs are the runs created so far, failedBatchIndex the batch that failed and cause its error, so a caller can keep or cancel those runs instead of retrying the whole request and creating them twice), events.push and bulkPush, workflows.put, rateLimits.put, runs.get, getDetails, cancel and replay (GetRunDetails, CancelTasks and ReplayTasks on the v1 AdminService), logs.put and streams.put. runMany honors returnExceptions on its RunManyOpt entries the way a declaration's runMany does: with it set, a failed run's slot holds an Error instead of rejecting the whole wait. logs.put rejects a line over 1,000 characters with a HatchetError stating the length, the limit the Node client's putLog drops such lines at; since the core call acknowledges the write, it does not drop silently. Its WorkflowRunRef.result() polls GetRunDetails with backoff (250 ms doubling to 5 s, jitter included), takes { timeoutMs, signal }, and reads outputs and failures the way the Node client's result() does: an absent output is {}, a stored null is null, an output that is not JSON rejects with its SyntaxError, a run rejects with the tasks' error messages when any task reported one, and a failed or cancelled run whose tasks carry no error resolves with their outputs. runs.get also returns each task's rawOutput, the stored text, next to the decoded output. The REST-backed features of the Node client (crons, schedules, runs.list, workflows.get, events.list) are not on this entry, since the generated REST client depends on axios.
• Added createFetchTransport to clients/transport. Its hostPort takes the gRPC target forms the Node client takes (host:port, dns:///host:port, ipv4:, ipv6:), the bearer token is checked to be header-safe the same way, requests and responses are held to the Node transport's 4 MiB message limits (a request over maxSendMessageBytes is refused before it is sent and a response over maxReceiveMessageBytes, counted after HTTP decompression, is cancelled; both reject with a ConnectError of code ResourceExhausted; a caller-supplied transport owns its own limits), and the core client's calls go through the same generated client adapters as the Node client's, so runs.get/getDetails accept the per-call signal, deadline and metadata options and WorkflowRunRef.result() cancels the poll in flight when its signal fires or its timeoutMs runs out, passing the remaining time to the call as a Connect timeout.
• scripts/check-edge-entry.mjs also rejects packages named with --forbid and SDK modules that use the process or Buffer globals, setImmediate, a timer's .unref(), AbortSignal.timeout or AbortSignal.any without feature-detecting them; pnpm check:core runs it against the core entry in CI, next to check:edge. A Go e2e (pkg/testing/e2e/tscore, go test -tags e2e) runs the core client from the built package over HTTP/1.1 against the harness engine.

Changed

• runs.getDetails on the Node client issues GetRunDetails through the Connect transport, the same code path the core client uses; its request and response types are unchanged. The trigger and event requests are built by modules both clients share (clients/admin/trigger-request, clients/event/rpc), with no change to what is sent.
• util/batch measures payloads with TextEncoder instead of Buffer, so it runs outside Node.

[1.34.0-alpha.1] - 2026-09-25

Fixed

• HatchetClient and HatchetLogger no longer write log lines when the log level is OFF.

[1.34.0-alpha.0] - 2026-09-22

Added

• Added a package exports map, so import { HatchetClient } from '@hatchet-dev/typescript-sdk/v1' now resolves under native Node ESM instead of failing with ERR_UNSUPPORTED_DIR_IMPORT (fixes #3613; supersedes #3650 by @dhimanAbhi, whose . and ./v1 entries it folds in). The map lists ., /v1, /edge, /core and every directory that ships an index.js explicitly, and wildcard entries keep every existing deep import resolving with or without the .js suffix, including a type import of an explicit .d.ts path. scripts/check-exports.mjs (pnpm check:exports) resolves each documented path from an ESM and a CommonJS scratch module against the built package in CI and compiles a TypeScript consumer of the declarations under the node16 and bundler resolvers.
• Added the @hatchet-dev/typescript-sdk/core entry point. It currently exports only CORE_ENTRY_VERSION; the application client for runtimes without gRPC is built on it.
• Added protobuf-es bindings for the SDK's protos (src/protoc-es, generated by protoc-gen-es through pnpm run generate-protoc-es) next to the ts-proto bindings, which remain the SDK's public message types.
• Added a Connect Transport seam in clients/transport: createNodeTransport(config) speaks the gRPC protocol over HTTP/2 through @connectrpc/connect-node, derives its TLS settings from tls_config (none, tls, mtls, ca_file, cert_file, key_file, server_name) with the same trust precedence as the grpc-js channel (ca_file, else the bundle GRPC_DEFAULT_SSL_ROOTS_FILE_PATH names, else the platform roots; GRPC_SSL_CIPHER_SUITES applies when set) and attaches the bearer token to every call. It dials the endpoint host_port names in the target forms grpc-js accepts (host:port, dns:host:port, dns:///host:port, ipv4:address:port, ipv6:[address]:port); a unix: socket target is refused at the first call with a clear error. HatchetClientOptions.transport injects a transport into the client.

Changed

• The admin, event and v1 admin clients issue their unary RPCs through the Connect transport. The public client properties (admin.client, admin.v1Client, events.client, and admin.workflowsGrpc and admin.adminGrpc on the v1 client) keep their generated WorkflowServiceClient, AdminServiceClient and EventsServiceClient types: every RPC of each service is served, and the nice-grpc call options (signal, metadata, onHeader, onTrailer) plus an absolute deadline (Date or epoch milliseconds) are honored on each call, with an aborted signal rejecting with the same AbortError as before. The wire protocol, the request and response types and the status-code classification are unchanged; the streaming clients, the worker and the run listeners stay on nice-grpc. Errors from these calls are ConnectErrors: code is the same numeric gRPC status code, the server's message is in rawMessage, and message reads [already_exists] ... where it used to read /WorkflowService/TriggerWorkflow ALREADY_EXISTS: ....
• EventClient and the legacy AdminClient keep their positional constructor parameters and take an optional transport last.
• HatchetClientOptions.credentials applies to the streaming nice-grpc channel only; the unary transport derives its TLS settings from tls_config.
• A token containing characters that cannot travel in an HTTP header (a stray line break in HATCHET_CLIENT_TOKEN, for example) now fails the first unary call with a fixed message instead of an error that quoted the token.

v1.33.2 - 2026-09-24

Fixed

• Importing the SDK no longer prints the v0 deprecation warnings. The root entry re-exports the v0 workflow and step modules, and the SDK itself imported them internally, so every v1 user saw HATCHET_V0_REMOVED twice plus a note about ConcurrencyLimitStrategy on startup. The warning is now emitted once, when a v0 workflow is registered or put to the engine.

v1.33.1 - 2026-09-09

Changed

• Embedded mode now reports first-run progress on stderr: a line when the sidecar download starts (with version and destination) and completes, a notice when a slow release or checksum fetch blocks startup, a startup line before waiting for the engine, and a heartbeat every 30 seconds while the engine is still becoming ready. Warm starts print at most one startup line.
• HatchetEmbeddedClient.init() now warns once when HATCHET_CLIENT_TOKEN is set in the environment, since Hatchet clients created with the standard constructor in the same process will not use the embedded engine.

v1.33.0 - 2026-09-08

Added

• Added /readyz and /livez endpoints to the worker health server, matching Kubernetes' own probe naming convention. /readyz returns 200 only when the worker is HEALTHY; /livez returns 200 whenever the process can respond, independent of Hatchet connectivity, so a transient upstream disconnect doesn't trigger a container restart. /health is unchanged.

v1.32.0 - 2026-09-08

Added

• Added the @hatchet-dev/typescript-sdk/edge entry point: the declaration factories and classes, Context, DurableContext, conditions, durations, errors and the wire types needed to declare and run tasks from a runtime without a Hatchet client (Cloudflare Workers, Vercel Functions). It imports nothing from Node, which scripts/check-edge-entry.mjs enforces in CI, and ships a declarations() factory that returns task, durableTask, workflow and batchTask bound to no client. The subpath resolves through the package's file layout, like the existing v1/embedded subpath.
• Added workflowToProto(definition, { namespace }), a pure function that builds the CreateWorkflowVersionRequest a workflow registers under, lifted out of the worker together with normalizeWorkflowDefinition and the mapping helpers. The worker registers through it.
• Added the ContextRuntime and DurableTransport interfaces. Context and DurableContext now perform every engine-facing operation through them; a worker supplies adapters over its client and durable listener, and the existing (action, client, worker) constructors keep working.
• Added createActionId(workflowName, taskName) and ParentRunContextManager.useStorage(...) so runtimes without AsyncLocalStorage can still declare and serve tasks.

Changed

• Action ids are registered fully lowercased (<workflow>:<task>), matching the ids the worker keys its registries by. Behaviour on the wire is unchanged: the engine already normalised ids (ParseActionID lowercases the verb and UpsertAction stores LOWER(actionId)), and the SDK now sends the normalised form itself.
• Memo keys for ctx.now() are computed with WebCrypto (crypto.subtle.digest) instead of crypto.createHash. The bytes are unchanged, so recorded durable event logs keep replaying; globalThis.crypto (Node 20 or newer) is required for durable tasks.
• bindAbortSignalHandler moved from util/abort-error to util/abort-signal, keeping the abort helpers free of Node imports.

v1.31.1 - 2026-09-03

Security

• Bumped qs to 6.16.0 to address GHSA-4mjr-xmp4-gh2g and GHSA-x5fp-wj9c-mxmx.

v1.31.0 - 2026-09-03

Added

• Added support for tenant-scoped shared concurrency strategies. Declare a Concurrency entry with isTenantScoped: true and a name, and reference the same name from tasks in different workflows so they share a single concurrency limit. Tenant-scoped entries can be mixed with ordinary workflow-scoped entries on the same task.
• Concurrency.maxRuns now accepts number | string: a string is a CEL expression over task input computing the max runs for each concurrency group, so different groups (e.g. pricing tiers) can have different limits.

v1.30.1 - 2026-09-01

Fixed

• Fixes another durable callback ordering bug which would cause NonDeterminismErrors to be raised on replay in the case where e.g. children were spawned recursively, concurrently.

v1.30.0 - 2026-08-26

Added

• Added support for CANCEL_QUEUED_EXCEPT_NEWEST and CANCEL_QUEUED_EXCEPT_OLDEST concurrency strategies.

v1.29.3 - 2026-08-31

Changed

• The missing-token configuration error now explains how to run Hatchet embedded for local development, via HatchetEmbeddedClient.init(), with a link to the embedded mode docs.

v1.29.2 - 2026-08-10

Fixed

• Improved error handling around failed heartbeats

v1.29.1 - 2026-08-25

Added

• Added stopEmbedded() on clients returned by HatchetEmbeddedClient.init() and a standalone stopEmbeddedSidecar() export, which gracefully stop the embedded engine (including its bundled Postgres) and resolve once it has fully exited.

v1.29.0 - 2026-08-24

Added

• Added embedded mode (early beta): HatchetEmbeddedClient.init() from the @hatchet-dev/typescript-sdk/v1/embedded entry point downloads and spawns the hatchet-embedded sidecar (checksum-verified, cached under ~/.hatchet/embedded), runs a full Hatchet engine locally with a bundled Postgres by default (pass databaseUrl to use your own), and returns a client wired to it. The separate entry point keeps embedded support out of production bundles.

v1.28.2 - 2026-08-10

Fixed

• Changed hatchet.task constructor concurrency options handling so that it does not set concurrency options at the workflow level, which would cause slow concurrency strategy handling in the engine

v1.28.1 - 2026-07-30

Fixed

• Fixed heartbeat worker logging to ignore Node watch-mode worker reload messages that don’t match the heartbeat message protocol.
• Improved heartbeat worker and action listener logging to ignore first network error, warn log on second error, and only error log past 3 errors to reduce false alarms due to transient network errors.

v1.28.0 - 2026-07-23

Changed

• Adds beta batchTask methods to both tasks and workflows, allowing for dynamic batching based on either time or batch size.

v1.27.0 - 2026-07-22

Added

• Adds support for terminal status-based idempotency keys, which are released when the task holding the key reaches a terminal state (either completed, cancelled, or having failed and exhausted all retries).

v1.26.2 - 2026-07-21

Added

• Added a retrier config option to ClientConfig (and corresponding HATCHET_CLIENT_RETRIER_MAX_ATTEMPTS, HATCHET_CLIENT_RETRIER_INITIAL_INTERVAL, HATCHET_CLIENT_RETRIER_MAX_JITTER env vars) to control retry behavior for user-facing gRPC calls — event pushes and workflow triggers. Internal engine communications (action events, stream events, workflow registration) are unaffected and continue using hardcoded defaults.

v1.26.1 - 2026-07-20

Added

• Added ctx.workflowNameV1() to return the current workflow name.

Deprecated

• Deprecated ctx.workflowName(), which continues to return the task name for backward compatibility. Use ctx.workflowNameV1() for the workflow name or ctx.taskName() for the task name.

Fixed

• Fixed workflow name values in context log metadata and OpenTelemetry attributes.

v1.26.0 - 2026-07-16

Added

• Adds support for defining idempotency keys on workflows and standalone tasks via an idempotency option, which ensures that they're only run once in a provided time window, based on a CEL expression. Triggers that collide with an existing run throw an IdempotencyCollisionError containing the existing run's ID.

v1.25.0 - 2026-07-09

Added

• Added slotCost to task options, so a task that needs more memory or CPU can consume more than one worker slot and a worker runs fewer of them at once. Durable tasks do not accept it, and on older engines it has no effect. See Task Slot Cost.

v1.24.3 - 2026-06-17

Removed

• Removed the unused _isV1 field and isV1 getter from HatchetClient. The getter always returned a hardcoded value and was not referenced anywhere in the codebase.

v1.24.2 - 2026-06-15

Fixed

• Fixed a bug where the durable event listener's request iterator could survive a stream reconnect and drain items from the new queue into the dead stream, causing durable tasks to hang indefinitely after an engine restart. The iterator now captures its queue and abort signal at creation time and terminates cleanly when the connection is replaced.

v1.24.1 - 2026-06-12

Fixed

• Fixed an issue where errors raised by child tasks spawned inside a durable parent task were not propagated back to the parent. The parent can now catch the child's error and handle it gracefully.

v1.24.0 - 2026-06-11

Added

• Added a getDetails method to hatchet.runs to retrieve task details.

v1.23.1 - 2026-06-09

Added

• Added an individualRunSpansForBulkRun OpenTelemetry config option. When enabled, a child hatchet.run_workflow span is created for each item in a bulk run (runWorkflows), nested under the parent hatchet.run_workflows span, with each item's traceparent pointing at its own span. Defaults to false to preserve the existing span structure.

Fixed

• WorkflowsClient.get() now finds the exact workflow name match from list results instead of taking the first result, preventing incorrect workflow ID resolution when a name prefix-match returns multiple workflows.

v1.23.0 - 2026-05-27

Added

• Fixes cancellation_grace_period and cancellation_warning_threshold not being propagated from client config to Hatchet config.
• Adds grpc_max_recv_message_length and grpc_max_send_message_length to client config, also configurable via env vars. Defaults to 4MB.

Fixed

• SDK import deprecation warnings to route via process.emitWarning with code HATCHET_V0_REMOVED.
• EventClient.BulkPush call uses options argument as fallback when no input is present.

v1.22.4 - 2026-05-22

Fixed

• Bumped @anthropic-ai/claude-agent-sdk to ^0.3.148 so Claude agent SDK integrations resolve the correct Linux native binary on glibc systems.
• Updated the TypeScript Claude agent example to load the ESM-only Claude Agent SDK dynamically.

v1.22.3 - 2026-05-18

Fixed

• Fixed @openai/agents import that was not inside try block and caused errors when installing with Bun.

v1.22.2 - 2026-05-13

Fixed

• Fixed DurableContext.waitForEvent overload ordering so calls without a payload schema infer the untyped event payload return type.

v1.22.1 - 2026-05-05

Fixed

Moved optional dependencies from optionalDependencies to peerDependencies.

v1.22.0 - 2026-04-28

Added

• Adds mcpTool method to Workflow objects for integration with Claude and OpenAI agent SDKs. Requires Zod v4.
• Bumps minimum Zod version to 3.25.0. Zod schemas provided to the SDK must be Zod 4 schemas, but you can still use Zod 3 in your application code.

v1.21.2 - 2026-04-22

Added

• Adds triggeringEventId and triggeringEventKey to the Context

v1.21.1 - 2026-04-21

Fixed

• Adds an optional label on durable event waits, which will propagate through to the dashboard

v1.21.0 - 2026-04-08

Added

• runMany and runManyNoWait APIs for workflows and standalone tasks to support bulk runs with per-run options.
• RunManyOpt input shape containing an input object and an options object.

Changed

• Bulk docs to include runMany and runManyNoWait examples.

v1.20.1 - 2026-04-07

Fixed

• Fixed duplicate child run deduplication when mixing ctx.runChild() and workflow.runNoWait() (or deeply nested recursive spawns). Context.spawnIndex and ParentRunContextManager.childIndex were tracked independently, causing both APIs to emit overlapping childIndex values and silently deduplicate children that should have been unique. The two counters now share a single source of truth via AsyncLocalStorage, and incrementChildIndex mutates the context object in place instead of replacing it with enterWith, which lost updates across await boundaries.

v1.20.0 - 2026-04-07

Added

• Adds scope and lookbackWindow arguments for the DurableContext.waitForEvent, which allows durable tasks to look back in time for events that may have been emitted before the task started.

v1.19.1 - 2026-03-25

Changed

• Event source info (hatchet__source_workflow_run_id, hatchet__source_step_run_id) is now injected into event metadata at the EventClient level, so cross-workflow trace linking works even without the OTel instrumentor enabled.

v1.19.0 - 2026-03-25

Fixed

• Fixed OpenTelemetry version mismatch causing TypeError: Cannot read properties of undefined (reading 'name') when exporting spans. The SDK now requires OpenTelemetry JS SDK 2.x (@opentelemetry/sdk-trace-base@^2.0.0, @opentelemetry/core@^2.0.0) to match the @opentelemetry/exporter-trace-otlp-grpc@^0.208.0 dependency.

Changed

• Updated OpenTelemetry optional dependencies to the unified 2.x release set.

v1.18.0 - 2026-03-18

Added

• OpenTelemetry instrumentation via HatchetInstrumentor with automatic tracing for workflow runs, event pushes, and step executions
• OpenTelemetry example demonstrating automatic and custom span instrumentation (examples/opentelemetry_instrumentation)

v1.17.2 - 2026-03-17

Added

• Added getTaskStats and scrapePrometheusMetrics methods to the metrics client.

v1.17.1 - 2026-03-17

Changed

• Updates the DurableTaskRunAckEntryResult interface to include workflowRunExternalId field, to enable spawning children from durable tasks fire-and-forget style.

v1.17.0 - 2026-03-16

Added

• Added a DurableContext.waitForEvent helper which returns the payload of the awaited event.
• Added an EvictionPolicy, which allows durable tasks to be evicted from the worker when idle.

Changed

• Makes a bunch of internal-facing changes for new durable execution features

v1.16.0 - 2026-03-11

Added

• Added logs client for retrieving task run logs.

v1.15.2 - 2026-03-06

Fixed

• waitFor and task conditions (e.g. user event keys) are correctly namespaced when using a non-default namespace.
• Cron expressions now support an optional leading seconds field (6-part expressions), e.g. 30 * * * * * to trigger at 30 seconds past every minute.

v1.15.1 - 2026-03-04

Fixed

• Fix npm publish so the package includes compiled JavaScript at the correct paths.

v1.15.0 - 2026-03-03

Added

• Adds a desiredWorkerLabels option to RunOpts to allow dynamically routing task runs to a specific worker at trigger time

v1.14.0 - 2026-02-28

Deprecated

• v0 SDK is now deprecated. Migrate to the v1 API for ongoing support.

Added

• Internal legacy transformer for backwards compatibility with existing v0 workflows and workers.

v1.13.1 - 2026-02-27

Changed

• Updated internal dependencies to address security advisories.

v1.13.0 - 2026-02-23

Added

• Introduced client middleware support with composable before/after hooks to customize request handling and response processing.
• Added middleware examples and recipes to demonstrate practical client-side patterns.

v1.12.1 - 2026-02-18

Fixed

• Restored ctx.taskRunId() as a deprecated alias for ctx.taskRunExternalId() on both v0 and v1 worker contexts, so existing code calling ctx.taskRunId() continues to work after the proto naming changes in 1.11.0.

v1.12.0 - 2026-02-13

Added

• Webhooks client for managing incoming webhooks: create, list, get, update, and delete methods for webhooks, so external systems (e.g. GitHub, Stripe) can trigger workflows via HTTP.

v1.11.0 - 2026-02-05

Internal Only

• Updated gRPC/REST contract field names to lowerCamelCase for consistency across SDKs.

v1.11.0 - 2026-02-04

Changed

• Updated the metrics client for the latest server metrics APIs (including adding getTaskStatusMetrics for tenant task/run status counts).
• Removes deprecated metrics methods.

v1.10.8 - 2026-02-02

Changed

• Improved cancellation log messages: cancellation-related logs now use debug level instead of error level since cancellation is expected behavior, not a failure.
• Updated terminology in log messages from "step run" to "task run" for consistency.
• Added link to cancellation docs (https://docs.hatchet.run/v1/cancellation) in error messages when task completion fails.

v1.10.7 - 2026-01-27

Added

• Adds support for an inputValidator prop on the various workflow definitions, e.g. hatchet.workflow and hatchet.task, which accepts a Zod schema to validate the input to the workflow or task. Used on the dashboard to provide autocomplete on the trigger workflow form.

v1.10.6 - 2026-01-27

Changed

• Improves handling of cancellations for tasks to limit how often tasks receive a cancellation but then are marked as succeeded anyways.