Nov 17, live webinar: ECS to EKS without freezing the roadmap for a quarter

Supabase CLI v3 Broke Your CI: Patch It and Stay, or Move to Your Own Postgres on Kubernetes?

Supabase CLI v3 made database commands local by default and requires --yes for hosted changes, which breaks unattended CI pipelines. Here is the ten-minute fix to stay, the signals that justify moving to your own Postgres on Kubernetes instead, and what each path costs.

Guillaume Da Silva
Staff Solutions Architect
NOV 17, 2026 · 16 MIN
Supabase CLI v3 Broke Your CI: Patch It and Stay, or Move to Your Own Postgres on Kubernetes?

Key points:

  • Supabase CLI v3.0.0 shipped on November 2, 2026, and npm, Homebrew, Scoop, the Linux packages and supabase/setup-cli now install it by default. Database commands (db push, db pull, db dump, migration list, gen types, storage) target the local stack unless you pass --remote, and every command that changes a hosted project asks for confirmation with No as the default (Supabase changelog).
  • That is why CI breaks. A pipeline that ran supabase db push or supabase functions deploy can't answer a prompt, so v3 fails with HostedMutationConsentRequiredError unless you pass --yes (Supabase discussion). The fix is --remote on database commands, --yes on hosted changes, and renaming [inbucket] to [local_smtp] in config.toml.
  • One breaking change is a bad reason to migrate. Supabase made the change so that "a command run without a flag" can no longer "reach a production database." Patch the pipeline, pin a version range, and stay if Supabase otherwise fits.
  • Move to your own Postgres when the reasons are structural. Data residency or a customer contract that requires your own cloud account, compute add-ons that cost more than equivalent managed Postgres with your discounts, a company standard on Kubernetes, VPC-private networking, or a release process that needs to own the toolchain. Two or more of those, and the migration usually pays for itself.
  • Leaving swaps one dependency for others. Your own stack has Kubernetes releases about every four months, Helm charts and migration tools to upgrade. Qovery runs the cluster and its upgrades in your own AWS, GCP, Azure or Scaleway account so a small team doesn't need a DevOps hire to absorb that.

Qovery · Infrastructure control plane
Kubernetes, operated through one governed API
Learn more

If your deploy job went red in early November with HostedMutationConsentRequiredError on a supabase functions deploy step that hadn't changed in months, you were probably running supabase/setup-cli without a pinned version. It picked up the new major release the day after it shipped. The fix is small. The question it tends to trigger is bigger: "Should we get off Supabase?"

I'm a Staff Solutions Architect at Qovery. I help teams move apps and databases into their own cloud accounts, and I also tell a fair number of them to stay where they are. This page is the decision guide: what broke, the fix if you stay, the signals that justify leaving, and the cost on each side. The step-by-step migration lives in How to migrate a Supabase app (Postgres, Auth, Edge Functions) to Kubernetes, and I only summarize it here.

What changed in Supabase CLI v3, and why does it break CI pipelines?

Supabase CLI v3 changed two defaults. Database commands now run against your local stack unless you pass --remote, and every command that changes a hosted project asks for confirmation with No as the default answer. In CI the CLI can't ask, so those commands fail unless you pass --yes (Supabase changelog).

The release went out on November 2, 2026, and Supabase's discussion notes that supabase/setup-cli with version: latest installs v3 by default from November 3 (GitHub discussion). That timing explains why so many pipelines broke the same week. The CLI is mostly a CI tool. When Supabase launched CLI v2 in December 2024, it said "nearly 85%" of its 180,000+ weekly installations came from CI/CD environments such as GitHub Actions (Supabase blog).

Here is everything in v3 that can change a pipeline's behavior:

  • Local by default. db push, db pull, db dump, migration list, gen types and storage commands target the local stack. A v2 supabase db push that reached your linked project now tries to reach a local database that doesn't exist on the runner.
  • Confirmation for hosted changes. functions deploy, secrets set, config push and db push --remote prompt first, default to No, and fail in CI without --yes.
  • Flag consolidation. --linked, --project-ref and --db-url are deprecated in favor of --remote (or --remote=<ref or URL>).
  • New configuration precedence. Settings resolve as flag, then shell environment, then project env files, then config.toml, then the default. "Flags and environment variables override config.toml, including [remotes.<name>] blocks." A stray SUPABASE_* variable on a runner can now win over your file.
  • A new diff engine. pg-delta is the default for schema diffs, so the SQL that db diff and db pull generate can change.
  • Removed keys and commands. The [inbucket] section must be renamed to [local_smtp] or v3 refuses to load the config. Deprecated commands such as db remote commit and db branch exit with code 1, and secrets unset now needs --all to remove every secret.

Supabase gives a clear reason for the change: "In v2, some database commands defaulted to the linked project and others to the local stack, and four different flags chose a target. A command run without a flag could reach a production database." I agree with the change. A CLI that can push to production because someone forgot a flag is worse than one that breaks your CI once.

How do you fix a CI pipeline broken by Supabase CLI v3?

Add --remote to database commands that should reach your hosted project, add --yes to every unattended command that changes it, rename [inbucket] to [local_smtp], and pin a version range in supabase/setup-cli so the next major version doesn't land on a Monday morning. Most pipelines need ten minutes.

The migration guide maps the common v2 commands to v3 (Supabase docs):

v2 command in your pipelinev3 equivalent
supabase db pushsupabase db push --yes --remote
supabase db push --db-url "$DB_URL"supabase db push --yes --remote="$DB_URL"
supabase migration list --linkedsupabase migration list --remote
supabase gen types typescript --project-id "$REF"supabase gen types typescript --remote="$REF"
supabase db dump --linked -f schema.sqlsupabase db dump -f schema.sql --remote
supabase functions deploysupabase functions deploy --yes
supabase secrets set --env-file .env.productionsupabase secrets set --env-file .env.production --yes
supabase config pushsupabase config push --yes
echo y | supabase <command>supabase <command> --yes

A typical GitHub Actions deploy job after the fix:

YAML
steps:
  - uses: actions/checkout@v4
  - uses: supabase/setup-cli@v3
    with:
      version: ~3.0.0   # a semver range, so v4 never arrives unannounced
  # keep your existing auth and project link setup, or pass --remote="$DB_URL"
  - run: supabase db push --remote --yes
  - run: supabase functions deploy --yes

If you need a day to test, pin v2 with version: v2.stable or version: ^2. Supabase says v2 "receives security fixes and fixes for broken behavior, but no new features," so treat the pin as a short bridge to v3.

Four checks I'd add before you merge the fix:

  1. Run the last v2 release first. v2.122.0 and later warn about removed commands, flags and config keys, and about commands whose default target changes in v3 (Supabase changelog). The warnings list your pipeline's breakage for you.
  2. Diff a schema before you trust pg-delta. Run db diff on a branch with v2 and v3 and compare the SQL before a v3 diff touches production.
  3. Audit runner environment variables. With the new precedence, an old SUPABASE_* variable in your CI settings overrides config.toml. Delete what you no longer use.
  4. Tell your coding agents. The guide says agents "never get a prompt, even in a terminal, and the error tells the agent to ask you before it passes --yes." That's a good default, so don't put --yes in an agent's standing instructions.

If this is the only thing that hurt, you're done. Stay on Supabase.

Is one breaking CLI change a good reason to leave Supabase?

No. A breaking change in a major version, announced with a migration guide, warnings in the previous release and a version you can pin, is normal maintenance for any tool your pipeline depends on. Leaving Supabase is a multi-week project; fixing v3 is a ten-minute pull request.

Your own stack has its own upgrade treadmill. The Kubernetes project ships a new minor version about every four months, and EKS gives each one 14 months of standard support before extended support kicks in at extra cost (Amazon EKS docs). Your migration tool, your Helm charts, your ingress controller and your Postgres major version all have release notes too. Somebody has to read them.

On your own stack, you choose when to take each upgrade, you test it in a cloned environment, and no vendor default changes under a pipeline that pulls latest. On a hosted platform, the vendor sets the pace and you adapt. Some teams want that trade. Some outgrow it.

So ask whether the structural reasons to leave apply to you.

What are the real signals that it's time to move to your own Postgres on Kubernetes?

Move when the reasons are about ownership, cost or compliance, and when two or more apply at the same time. The common ones: data must live in your own cloud account, compute add-ons cost more than managed Postgres would with your discounts, your company standardizes on Kubernetes, the database must sit in your VPC, or your release process needs to own every tool in the pipeline.

Here is how I score it with customers:

SignalWhat it looks likeWeight
Data residency or contract termsAn enterprise customer requires data in your AWS, GCP or Azure account, in a named region, with your own encryption keysDecisive on its own
VPC-private databaseSecurity wants the database unreachable from the internet and inside your VPC peering or private link setupHigh
Compute add-on spendYou're paying for XL to 4XL compute ($210 to $960 a month per project) on several projects, plus PITRHigh once it's a real line in the budget
One cloud billFinance wants spend on one invoice where Savings Plans or committed-use discounts applyMedium
Company Kubernetes standardThe rest of your services already run on Kubernetes and Supabase is the exceptionMedium
Toolchain ownershipReleases need pinned, audited tooling and change windows you controlMedium
One breaking CLI changeA ten-minute pipeline fixLow
Small team, early product, no compliance asksSupabase saves you real operations workStay

The cost signal comes from Supabase's own price list. Pro starts at $25 a month and Team at $599, both include $10 of compute credits, and compute add-ons run from $10 (Micro) to $210 (XL), $410 (2XL) and $960 (4XL) a month per project. Point-in-time recovery is $100 a month per 7 days of retention, disk is $0.125 per GB past the included 8 GB, and egress is $0.09 per GB (Supabase pricing). Those prices are reasonable for what they include. They stop looking reasonable when the same company already has a committed AWS spend with discounts that never apply to the Supabase invoice.

The signals to stay matter too. If your app leans on PostgREST from the browser, Row Level Security policies, Realtime and Storage, leaving means rebuilding pieces your frontend calls directly. Supabase's own self-hosting page is clear that self-hosted Supabase is community-supported, without branching, managed backups, PITR or the Management API (Supabase self-hosting docs). If nobody is asking for your own account and your bill is in the hundreds of dollars, stay.

Run your Postgres and your API in your own cloud account.
Qovery deploys your services and a managed PostgreSQL into your own AWS, GCP, Azure or Scaleway account, or your existing Kubernetes cluster, with a preview environment per pull request. Start deploying in under 10 minutes.

What does staying on Supabase cost compared with your own Postgres on Kubernetes?

For a small project, Supabase is cheaper than any self-run stack, because the fixed cost of a cluster and an engineer's time outweighs a $25 to $135 monthly bill. The cost lines cross when you run large compute add-ons on several projects, pay for PITR, and already have cloud discounts that a managed Postgres in your own account would pick up.

Cost lineSupabase hosted (Pro/Team)Your own Postgres on Kubernetes in your cloud account
Platform base$25/month (Pro) or $599/month (Team)Kubernetes control plane: $0.10 per cluster-hour on EKS standard support, about $73/month (AWS)
Database computeAdd-ons from $10 to $960+/month per projectManaged Postgres (RDS, Cloud SQL, Azure Database for PostgreSQL, Scaleway) at your cloud's rates
DiscountsNot applicableSavings Plans up to 66% (Compute) or 72% (EC2 Instance) on AWS compute (AWS)
Backups and PITRPITR $100/month per 7 days of retentionProvider-managed automated backups and PITR in RDS and its equivalents
Egress$0.09/GB past the included quotaYour cloud's rates
Kubernetes upgradesNoneEvery ~4 months; extended support on EKS is $0.60 per cluster-hour (AWS)
PeopleClose to zero opsSomeone owns the cluster, unless a platform runs it for you

The people line decides most of these comparisons. If moving off Supabase means hiring a DevOps engineer to run Kubernetes, the math rarely works below a large database bill. If something else runs the cluster and its upgrades, the comparison is mostly compute against compute, and your discounts start to count.

EKS extended support costs six times the standard control plane rate, so a team that falls behind on Kubernetes upgrades pays for it. That's the self-hosted version of a breaking change you didn't plan for, and it's why I don't recommend leaving Supabase unless someone or something owns the upgrade cadence.

What does the move off Supabase look like, in short?

Treat it as three migrations: Postgres (including the auth schema), Auth (GoTrue and its JWT secret), and Edge Functions. Move them in separate windows with separate rollback plans. The full walkthrough is in our Supabase to Kubernetes migration guide; here is the outline.

  1. Inventory first. Count users and auth providers, list extensions, RLS policies, storage buckets, Edge Functions, cron jobs and triggers. Supabase is open source (Postgres, PostgREST, GoTrue, Realtime, Storage, the Deno edge runtime), so you're moving known components.
  2. Move Postgres. A dump and restore needs a planned write freeze; logical replication gets the cutover down to seconds. Restore into managed Postgres in your account, not a StatefulSet.
  3. Keep or drop PostgREST deliberately. If the frontend calls supabase-js .from().select(), run PostgREST as a container in front of the new database, or budget for an API rewrite.
  4. Move Auth without password resets. The bcrypt hashes in auth.users are ordinary table data. Run supabase/auth yourself or import the hashes into another identity provider.
  5. Turn Edge Functions into services. Each Deno function becomes an API route or a small container; supabase.functions.invoke() becomes a plain fetch.
  6. Replace the CLI in CI. Database migrations move to a tool you pin and own (the Supabase migrations folder is plain SQL, so tools such as Flyway, Atlas, Sqitch or golang-migrate can apply it), and deployments come from your platform instead of supabase functions deploy.

After the move, your pipeline still runs migrations, but you pick the tool, pin its version, and upgrade it on your schedule.

Where does Qovery fit if you decide to leave Supabase?

Qovery runs the Kubernetes side of the move in your own AWS, GCP, Azure or Scaleway account, or on the Kubernetes cluster you already have, so a small team can leave Supabase without hiring someone to operate a cluster. It doesn't host your app or your data: the control plane is Qovery's, the cluster, database and bill are yours.

Concretely, for a Supabase migration:

  • Your API, PostgREST if you keep it, the GoTrue container and each former Edge Function become Qovery services that deploy on git push.
  • The database runs on a managed cloud service (Amazon RDS, Cloud SQL, Azure Database for PostgreSQL, Scaleway Managed Database), with the provider's backups and failover.
  • Every pull request can get a preview environment with the full stack, which is where you test migration scripts and the auth flow before the cutover. Environment auto-stop shuts those copies down outside working hours.
  • Qovery runs managed cluster upgrades, which covers the Kubernetes upgrade cadence from the cost table above, and adds per-environment RBAC and audit logs.
  • The Qovery Terraform provider and Helm services let you keep the whole setup as code, with versions you pin.

If you're one developer on an early product, stay on Supabase and apply the CLI fix. This path is for teams with a contract, a compliance requirement or a bill that says the database belongs in their own account.

How do I run this with Qovery from Claude Code or Cursor?

Install the Qovery Agent Skills once, optionally connect the MCP Server, then describe what you want in plain language. Every action waits for your approval. These are the only commands you need:

BASH|Install the Qovery Agent Skills globally (Claude Code, Cursor, Codex, OpenCode, Gemini CLI)
curl -fsSL https://skill.qovery.com/install.sh | bash

# Same install, scoped to the current repository
curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project

# Connect Claude Code to the Qovery MCP Server (OAuth, read-only by default)
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242

# Write access, only when you want the agent to change infrastructure (also enable it in the Qovery Console)
claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242

The skills (qovery-onboard, qovery-deploy, qovery-preview and others) do the setup and deploying; the MCP Server reads and operates what already exists, and its OAuth step asks you to approve access. A good first prompt: "deploy this API and a managed PostgreSQL to my AWS account with Qovery, and create a preview environment per pull request." Swap AWS for GCP, Azure, Scaleway or your existing Kubernetes cluster. Start with the AI agent quickstart or the Claude Code integration.

Frequently asked questions
What changed in Supabase CLI v3 and why did it break my CI pipeline?

Supabase CLI v3.0.0, released November 2, 2026, makes database commands such as db push, db pull, db dump, migration list and gen types target the local stack unless you pass --remote. Commands that change a hosted project, including functions deploy, secrets set and config push, now ask for confirmation with No as the default, and fail in CI with HostedMutationConsentRequiredError unless you pass --yes. Pipelines using supabase/setup-cli without a pinned version picked up v3 automatically.

How do I fix my GitHub Actions workflow for Supabase CLI v3?

Add --remote to database commands that should reach your hosted project, add --yes to unattended hosted changes (supabase db push --remote --yes, supabase functions deploy --yes), and rename [inbucket] to [local_smtp] in supabase/config.toml. Pin a semver range on supabase/setup-cli@v3, or pin v2 temporarily with version: v2.stable. Run v2.122.0 or later first to see warnings for everything v3 will change.

Is a Supabase CLI breaking change a good reason to move to my own Postgres on Kubernetes?

Not on its own. The v3 change fixes a real safety problem, ships with a migration guide, and takes minutes to adapt to. Move to your own Postgres on Kubernetes when structural reasons apply: data residency or a contract requiring your own cloud account, large compute add-on and PITR spend, VPC-private networking, a company Kubernetes standard, or a release process that must own its tooling.

When does it make sense to move off Supabase to Postgres on Kubernetes?

It makes sense when two or more ownership, cost or compliance signals apply at once, and when something runs the cluster for you. Typical triggers are an enterprise customer requiring data in your AWS, GCP or Azure account, several projects on XL to 4XL compute ($210 to $960 a month each), and cloud discounts that a managed Postgres in your account would pick up. Small teams with no such pressure should stay on Supabase.

Does leaving Supabase get rid of breaking changes in my toolchain?

No. Your own stack has Kubernetes minor releases about every four months, plus Helm charts, a migration tool and Postgres major versions to upgrade, and EKS extended support costs $0.60 per cluster-hour if you fall behind. The difference is that you pin versions and choose when to upgrade. A platform such as Qovery runs the cluster upgrades in your own cloud account so that work doesn't land on a team without a DevOps engineer.

Guillaume Da Silva
About the author
Guillaume Da Silva

Guillaume is a Staff Solutions Architect at Qovery. He helps customers troubleshoot production infrastructure and writes about Kubernetes operations, AI agents, and reliable cloud platforms.

Next step

Run your Postgres and your API in your own cloud account.

Qovery deploys your services and a managed PostgreSQL into your own AWS, GCP, Azure or Scaleway account, or your existing Kubernetes cluster, with a preview environment per pull request. Start deploying in under 10 minutes.