Exodus and your coding agent audit your app, show what it costs today and on Google Cloud Run, and move it there step by step. Or plan a new product for Cloud Run from scratch.
Cloud Run Superfan AwardGoogle Cloud Run Hackathon on Product HuntYou can move lumen-api to Cloud Run after 4 changes.
src/bin/worker.rs:14Cloud Taskssrc/bin/clock.rs:12Cloud Schedulersrc/routes/upload.rs:41Cloud Storagesrc/app.rs:27Signed cookieslumen-api-k7qm.europe-west1.run.app> In this repository's folder, run exodus init --code K7QM-4XPT, then audit it for Cloud Run● Ran 4 commands⎿ exodus init⎿ exodus audit⎿ exodus ask "tokio worker on Cloud Run?"⎿ exodus send● Sent 41 findings to Exodus, no code. Report 1 is ready.· Worked 2m 14s>▶▶ auto · Opus 5.5Hand one prompt to your coding agent. Exodus gives it the stairs to Cloud Run, on Google's own infrastructure, whether the app runs today or is still an idea.
From Heroku, Railway, Render, Fly.io, Vercel or your own server. You see the bill there and on Cloud Run, and what changes on the way.
Tell your agent the idea. Exodus designs how it runs on Cloud Run: the services, the database, the stack and the bill at launch.
Run it in your terminal, once per machine.
Sign in with Google, copy the prompt into your agent, and it audits the app on your machine.
Check the bill and the changes, and the agent moves the app.
On Railway: the web, worker and clock processes run every hour beside Postgres, Redis and a volume. The bill gives one total, so no part has a price of its own.
Rebuilt for Cloud Run, each part runs only when there is work, and Redis is no longer needed.
On Cloud Run, traffic out is $1.44 a month; builds and secrets fit the free tier.
worker.rs:14Cloud Taskslet (_, job): (String, String) = redis.blpop("jobs", 0.0).await?;Cloud Run gives a copy CPU only while it answers a request, so a tokio loop waiting on Redis stalls, or under instance billing is paid for all month.
Each job becomes a POST to /tasks/<name>, an axum route on the same service, and the queue takes over retries and rate limits.
$0.00 on Cloud Runclock.rs:12Cloud Schedulerlet mut hourly = tokio::time::interval(Duration::from_secs(3600));Copies stop when idle and several can run at once, so a timer inside one fires late, never, or twice.
Cloud Scheduler starts a Cloud Run job at the same three times.
$0.00 on Cloud Runapp.rs:27Signed cookies.layer(SessionManagerLayer::new(RedisStore::new(pool)))Once the queue leaves, sessions are Redis's only job, and Memorystore bills every hour, busy or not.
axum-extra's SignedCookieJar signs the session into a cookie, with its key in Secret Manager.
$0.00 on Cloud Run: Redis goesupload.rs:41A Cloud Storage buckettokio::fs::write(upload_dir.join(&name), &bytes).await?;A copy's disk lives in its memory and is gone when the copy stops, and other copies never see it.
The browser puts the file straight into the bucket through a signed URL, and the service keeps only its key.
$0.02 on Cloud RunToday is lumen-api's Railway bill as the sample report shows it, one total. Cloud Run is the report's estimate at Google's list prices in Belgium as of October 2026, for 1,200 visitors a day.
VerdictYou can move lumen-api to Cloud Run after 4 changes.
worker.rs:14Must changeclock.rs:12Must changeupload.rs:41Must changeapp.rs:27Bills idlecac46de1442ecea4677445d9c91dExodus is a command line your agent runs, so it works with the one you already use. It brings the instructions each agent reads.
Anything that runs in a container runs on Cloud Run. Exodus picks the build for your stack: Google's buildpacks or a Dockerfile on the right base image.
exodus ask "why does my tokio worker stall on Cloud Run?"1. Background workers on Cloud Run - Why a queue worker breaks on Cloud RunA worker loop is not a request, so between requests its CPU is throttled to almost nothing and jobs sit in Redis.https://exodus.elim.io/blog/background-workers-with-cloud-tasks Exodus is built in Rust, the CLI included. Its knowledge base is the guides on our blog: your agent runs exodus ask and gets the passage that answers it, with the link to the guide, instead of guessing.
Go, Node.js, Python, Java, .NET, Ruby and PHP deploy straight from source.
It adds copies when traffic comes and removes them when it goes.
Each with a 99.95% uptime commitment in writing.
You pay for the time code runs. 2 million requests a month are free.
1 Gartner, Magic Quadrant for Cloud-Native Application Platforms, 3 August 2026; the recognition is for Google's platforms. Gartner does not endorse any vendor or product. GARTNER and MAGIC QUADRANT are trademarks of Gartner, Inc.
| Cloud Run | Heroku | Railway | Render | Fly.io | |
|---|---|---|---|---|---|
| You pay for | Seconds your app is working | Every hour a dyno is up | CPU and memory used, plus a plan fee | Every hour an instance is up | Seconds a machine runs |
| Stops when nobody uses it | On every service | Eco dynos only | Optional sleep mode | Free tier only | Auto-stop |
| Adds copies under load by itself | Every service, up to 1,000 requests per copy | Performance dynos only | Copies set by hand | Pro workspaces and up | Starts machines you created |
| Regions to deploy in | 43 | 12 | 4 | 5 | 17 |
| Uptime commitment | 99.95% on every service | Not published | Enterprise only | Enterprise only | 99.9% from $2,500 a month |
| Try a release on part of the traffic | Any percentage, back in one click | Not documented | Not documented | Not documented | Canary and blue-green deploys |
| Free every month | 2M requests, 180k vCPU-seconds | None | Trial, then $1 credit | 750 hours, sleeps when idle | None |
From each platform's own documentation and pricing pages, October 2026.
| Cloud Run | A VPSHetzner, DigitalOcean | A VM on GoogleCompute Engine | |
|---|---|---|---|
| You pay for | The seconds it works | The month, busy or idle | Every hour it runs |
| A small app, 1M requests a month | About $0.48 a month | $4 to $7 a month | $18.10: e2-small with a disk and an IP |
| Busy around the clock | About $63 a month for 1 vCPU | The same $6 to $24 | $26.91, e2-medium |
| Operating system | Google's, patched for you | Yours to patch | Yours to patch |
| HTTPS, scaling, load balancing | Built in | You set them up | A load balancer, billed apart |
| Releases and rollback | Revisions, any share of traffic | You build it | You build it |
| Local disk, always-on processes | Files go to a bucket, jobs to a queue | Yes | Yes |
| Traffic out | $0.12 a GB in Europe | Terabytes included | $0.12 a GB |
Google's list prices for europe-west1, Hetzner CX23 and DigitalOcean Basic, October 2026. Cloud Run billed per request, 1 vCPU and 512 MiB, with the free tier.
| Cloud Run | AWS Lambda | Cloudflare Workers | |
|---|---|---|---|
| What it runs | Your container as an ordinary web server | A function per event; a web app needs the Web Adapter | JavaScript, TypeScript, Python, Wasm |
| Requests one copy takes at once | Up to 1,000 | One | Billed per request and CPU time |
| Memory | Up to 32 GiB | Up to 10 GB | 128 MB |
| Longest request | 60 minutes | 15 minutes | 5 minutes of CPU |
| Cold start, median | 0.53 s | 0.28 to 0.31 s | 0.015 s |
| 10M requests of 400 ms a month | $13.69 | About $28.47 | Priced another way |
| Containers | Any image | Images up to 10 GB, with its runtime API | Cloudflare Containers, no autoscaling yet |
AWS and Cloudflare documentation and price pages, October 2026. Cold starts from the TU Berlin study below; the price row is Google's own pricing example and the same traffic on Lambda x86 with 512 MB.
Schirmer et al., TU Berlin, "New Kids: An Architecture and Performance Investigation of Second-Generation Serverless Platforms", ACM TOIT 2026. Calls from Frankfurt, November 2025 to January 2026. Cloud Run measured as Cloud Run functions in europe-west3 with startup CPU boost off; Fly.io as Fly Machines with fly launch defaults. Data and notebooks.
Pays only while it works. A server bills every hour of the day. Cloud Run bills the seconds your code answers a request, so a quiet night costs nothing and the free allowance covers a small app entirely.
Grows on its own. A spike, and Cloud Run adds copies in seconds, each taking up to 1,000 requests at once. When it passes, they go.
Close to your visitors. A request enters Google's network near the visitor and rides Google's own lines to your app.
Nothing to babysit. No server to log into and no certificate that expires on a Sunday. You deploy, and Google keeps the rest running.
Releases without fear. Every deploy is a revision. Show it to 5% of visitors, move everyone when it holds, go back in one click.
All of Google next door. Databases, files, queues, AI models and analytics in the same account and on the same bill, each one permission away.
The CLI is open source, Apache-2.0. The web cabinet is free.
Many services, a large database, no room for downtime. Our engineers have moved production apps to Cloud Run and will plan and run the move with your team.
Consult an expertEvery figure names its source. The CLI's exodus ask answers from the same guides.
Cloud Run runs your container and bills for the seconds it works. What it runs, its three shapes, how it bills, what changes in your code, cold starts, what it does not do, a first deploy.
Oct 10, 2026Google's list prices for Cloud Run and the services around it, the free tier, a small app priced line by line, the case where Cloud Run costs more, and how to check every price yourself.
Oct 10, 2026Google Cloud's free tier, product by product for a small app: Cloud Run, Firestore, Cloud Storage, one e2-micro. What is never free, and how the $300 trial ends.
Oct 19, 2026Cloud Run gives each billing account $4.32 of CPU, $0.90 of memory and 2 million requests a month. What that buys, who shares it, what quietly uses it up, and how to read what is left.
Oct 11, 2026One web app priced line by line at 100 thousand to 100 million requests a month: Cloud Run, database, egress, logs, images, every assumption written down, and how to check the prices yourself.
Oct 11, 2026What Fly.io charges for in 2026: Machines by the second, volumes, egress, IPv4, Managed Postgres and support. No free tier. The same small app priced on Cloud Run.
Oct 19, 2026A VPS bills the month; Cloud Run bills the seconds your code runs. Worked prices for a small app, a busy one and a steady one, and what each price leaves out: database, egress, disk, your time.
Oct 11, 2026The decisions to make before writing a product for Cloud Run: where state lives, the database, background work, payments, live updates, the region, and the bill at launch, in a year and busy.
Oct 11, 2026What each part of a Heroku app becomes on Google Cloud, the order of the move, the commands for each part, what it costs on both sides, what breaks, and the same map for Railway, Render and Fly.
Oct 10, 2026Railway and Render projects map to Google Cloud piece by piece. What each part becomes, the commands, the cost on each side, the order of the move, and when staying is the better answer.
Oct 10, 2026Fly.io runs your Dockerfile on VMs you size and count; Cloud Run runs the same image and counts for you. What each part of fly.toml becomes, secrets, volumes, Postgres, the domain, and both bills.
Oct 11, 2026A VPS runs nginx, the app, Postgres, cron and uploads on one machine. What each piece becomes on Cloud Run, the commands, what breaks, and why the bill usually goes up while the work goes down.
Oct 11, 2026App Runner takes no new customers, Lambda needs an adapter and Fargate bills all month. What each AWS piece becomes on Google Cloud, a small app's cost on each side, and the move in six steps.
Oct 11, 2026A dump and restore moves a small Postgres in minutes and loses rows written meanwhile unless writes stop. The order that keeps a way back, Cloud SQL's flags, proving both sides match.
Oct 12, 2026Queue workers, work after the response, timers, files on disk, in-memory sessions and open connections: why each breaks on Cloud Run, how to find it, what replaces it and what that costs.
Oct 10, 2026A worker waiting on a Redis queue stalls or bills all day on Cloud Run. How to turn each job into a Cloud Tasks request, with retries, rate limits, costs, and when a worker pool fits better.
Oct 11, 2026In-process timers do not fire on an idle Cloud Run app and fire twice on two instances. How to move each to Cloud Scheduler: endpoint or job, service account, run-once handlers, time zones, cost.
Oct 11, 2026In-memory sessions log people out on Cloud Run. How to use signed cookies with keys in Secret Manager, log out everywhere with one column, when a database table fits, and when Redis is worth it.
Oct 11, 2026A Cloud Run instance's disk is memory and dies with the instance. How to move uploads to Cloud Storage with signed URLs, serve downloads, move existing files, and what it costs.
Oct 11, 2026The day traffic moves to Cloud Run, step by step from a week before to a week after: the DNS TTL, freeze, copy, compare and switch, checks from outside, the way back, and the overlap's cost.
Oct 12, 2026Exodus is a free web app and an open-source command line. With the coding agent you already use, it audits your app, shows what it costs today and on Google Cloud Run, lists what to change and in what order, and then moves it there step by step. Exodus runs no AI of its own: your agent writes the code.
Nothing. The CLI is open source under Apache-2.0, and the web app, audits, reports, the plan and the knowledge base are free. You pay Google for Cloud Run and the other services your app uses, on your own account. The report estimates that bill from Google's list prices, with the date of the prices.
Not for the audit and the report: you sign in with Google, and your agent reads the repository on your machine. To move, you need a Google Cloud project with billing. Your agent works in it with your own gcloud, and a step only you can take, such as a login or a billing account, shows in the app with its steps. Exodus never signs in to Google Cloud.
No. The CLI reads the repository on your machine and sends the findings, not the code: for example a worker on a Redis queue, Postgres 180 MB, a timer at src/bin/clock.rs:12. There is no GitHub sign-in and no GitHub app. A report stays private until you share its link.
Any agent that runs commands in a terminal: Claude Code, Codex, Cursor, Gemini CLI or Copilot. exodus init writes the instructions each one reads: skills for Claude Code, AGENTS.md for Codex and the others, rules for Cursor. We test with Claude Code and recommend it.
The report says so. Its verdict is move, move after changes, or stay, with the reason. Cloud Run runs containers that answer requests and scale to zero, so an app under steady heavy load, one that keeps state on its own disk, or one that serves no HTTP may be better where it is. The audit and the report cost nothing either way.
There is no subscription to cancel. Delete a project in its Settings tab, or your whole account in Settings under Danger zone: its reports, share links and CLI keys go with it. On your machine, delete the exodus binary from ~/.local/bin and the .exodus folder in your repository.
Use Exodus on our platform for free, or take the open-source CLI and run the whole move yourself.