Cloud Tasks alternative
The open-source alternative to Google Cloud Tasks.
Cloud Tasks got the model right: enqueue an HTTP request, and the queue delivers it to your endpoint with retries and rate limits. Workly keeps that model and takes it off GCP. Run it on your laptop with no configuration, as one container next to your apps in any cloud, or hosted at workly.run. And it keeps every attempt, so a failed task tells you why.
Workly and Cloud Tasks compared.
The same idea: your app only needs an endpoint, and the queue does the delivering. The differences are where it runs and what it remembers.
| Google Cloud Tasks | Workly | |
|---|---|---|
| Delivery | HTTP push to your endpoint, or App Engine | HTTP push to your endpoint |
| Runs on | Google Cloud only | Your laptop, any cloud or cluster, or hosted at workly.run |
| Local development | No official emulator; community stubs | workly dev: SQLite, dashboard and an echo target, no configuration |
| Source | Proprietary | Open source, Apache-2.0 |
| Retries | Attempts, backoff, doublings and a maximum retry duration | Attempts and backoff with jitter; honours Retry-After; you choose whether 4xx is retried |
| Rate limits | Dispatches per second (up to 500 a queue) and concurrent dispatches | Per queue: rate_per_second and max_concurrency (up to 10,000) |
| Delayed tasks | schedule_time, up to 30 days ahead | delay_seconds or run_at |
| Deduplication | Task names, for up to 24 hours | idempotency_key on enqueue |
| Handler time | Up to 30 minutes | Up to 30 minutes (timeout_seconds) |
| Endpoint auth | OIDC or OAuth token from a service account | HMAC-SHA256 signature; the SDKs check it in one call |
| After success | Removed from the queue; attempts only in Cloud Logging, if enabled | Kept with every attempt: status, error, response body, timing |
| After the last retry | Deleted; no dead-letter queue | Kept as failed, grouped by error; retry one or all once the fix is in |
| Visibility | Cloud Console and Cloud Monitoring | Built-in dashboard, MCP for coding agents, and Prometheus metrics when self-hosted |
| Cron | A separate product, Cloud Scheduler | Schedules in the same server, with run history and run-now |
| Workflows | A separate product, Google Cloud Workflows, with its own YAML syntax | Workflows in the same server: a DAG of HTTP steps, each a task in your queues |
| Payload | Up to 1 MiB | 1 MiB by default; up to 100 MiB self-hosted |
| Price | 1M operations free a month, then $0.40 per million; every attempt counts | Free to self-host. Hosted: 100k tasks a month free; a task counts once, however many attempts |
Cloud Tasks limits and prices from Google's documentation, October 2026.
A task that failed should say why.
Cloud Tasks deletes a task once it succeeds, and again once it runs out of retries; what happened in between is in Cloud Logging if you turned it on. Workly keeps the task and every attempt, with the status code, error, response body and timing. Failed tasks are grouped by error, so after a bad deploy you fix the handler and retry the whole group.

Moving over is a few lines.
Queues map one to one, with the same settings under plainer names. Workly creates a queue on first use, or keep them in a workly.yaml and apply it on every deploy.
import json
from google.cloud import tasks_v2
client = tasks_v2.CloudTasksClient()
client.create_task(
parent=client.queue_path("my-project", "europe-west1", "emails"),
task=tasks_v2.Task(
http_request=tasks_v2.HttpRequest(
http_method=tasks_v2.HttpMethod.POST,
url="https://api.example.com/send-email",
headers={"Content-Type": "application/json"},
body=json.dumps({"user_id": 123}).encode(),
oidc_token=tasks_v2.OidcToken(
service_account_email="tasks@my-project.iam.gserviceaccount.com",
),
),
),
)from workly import Workly
workly = Workly() # WORKLY_URL, WORKLY_API_KEY
workly.tasks.enqueue(
"emails",
url="https://api.example.com/send-email",
payload={"user_id": 123},
)gcloud tasks queues create emails \
--location=europe-west1 \
--max-dispatches-per-second=5 \
--max-concurrent-dispatches=10 \
--max-attempts=5 \
--min-backoff=2s \
--max-backoff=3600s# workly.yaml, applied with `workly deploy`
version: 1
queues:
emails:
rate_per_second: 5
max_concurrency: 10
max_attempts: 5
min_backoff_seconds: 2
max_backoff_seconds: 3600In your handler, check a signature instead of a token.
Cloud Tasks proves a request is genuine with a Google-signed OIDC token. Workly signs every delivery with HMAC-SHA256 over the timestamp and body, and the Python and TypeScript SDKs check it in one call. A Workly-Task-Id header stays the same on every attempt, for deduplicating, much like the task name header Cloud Tasks sends.
from workly import SignatureVerificationError, verify_signature
@app.post("/send-email")
def send_email():
try:
# Instead of checking a Google OIDC token
verify_signature(request.get_data(), request.headers)
except SignatureVerificationError:
abort(401)
...
return "", 204Replace Cloud Scheduler too.
On GCP, a cron job that calls an endpoint is a second product. In Workly it's a schedule on the same server: a cron expression in your timezone, retried and signed like a task, with every run recorded and a button to run it now.
Read about schedules →When Cloud Tasks is the better choice.
You're all in on Google Cloud
Cloud Tasks authenticates to Cloud Run and Cloud Functions with IAM: no secret to keep, nothing to run. If that matters more than history and portability, stay.
You target App Engine
App Engine targets route by service and version. Workly only speaks HTTP to a URL.
You need a vendor SLA today
Hosted Workly has no SLA yet. Self-hosted Workly is as available as the Postgres and pods you run it on.
Questions
- Is Workly a drop-in replacement for the Cloud Tasks API?
- No. The model is the same (queues, HTTP tasks, retries, rate limits), but Workly has its own small HTTP API and SDKs. Moving is mostly replacing the create_task call with enqueue and the OIDC check in your handler with Workly's signature check.
- Can I run Workly on Google Cloud?
- Yes. Run it on GKE or a VM with Cloud SQL for Postgres, next to your services. It needs to run continuously, since it is the dispatcher.
- Can Workly deliver to services on a private network?
- Self-hosted, yes: endpoints inside your VPC or cluster are just URLs. Hosted Workly only delivers to public addresses.
- How do I test Cloud Tasks-style code locally?
- Run workly dev. It starts the server and dashboard on localhost:7337 with SQLite and no configuration, and delivers to handlers on localhost. The same code runs against production by changing WORKLY_URL.
- What happens to a task that keeps failing?
- After its last attempt it's marked failed and kept, with every attempt's status, error and response body. Failures are grouped by error, so you can fix the handler and retry one task or all of a queue's failures at once.
Try it in five minutes.
pip install "workly[cli]" and workly dev give you a server and dashboard on your laptop. Or sign in to hosted Workly and get a workspace and an API key in a minute, free on the Hobby plan.