Vertra Cloud Knowledge Base

Mirror: plans-and-limits publishes the plan, runtime, database, Data tab and deploy limit tables from this page in summarized form. Changed a number here? Update plans-and-limits, overview and api-reference/limitations in the same pass.

​

1. Overview

• Site: https://vertracloud.app
• Dashboard: https://vertracloud.app/dashboard
• API: https://api.vertracloud.app
• Documentation: https://docs.vertracloud.app
• CLI (guide): curl -fsSL https://cli.vertracloud.app/install | sh on macOS/Linux, irm https://cli.vertracloud.app/install | iex on Windows (PowerShell)
• MCP server (guide): remote endpoint https://mcp.vertracloud.app/mcp (OAuth), or local with npx -y @vertracloud/mcp and a VERTRA_API_KEY
• SDKs (guide): npm install @vertracloud/sdk-api · pip install vertracloud-api · go get github.com/vertracloud/sdk-api-go
• VS Code extension (guide): code --install-extension VertraCloud.vertra-cloud
• GitHub Actions (guide): vertracloud/github-action

​

2. Supported Languages

​

Language detection

​

Deno, .NET and Elixir

​

3. Databases

​

How to connect to each database

• Host: <database-id-without-hyphens>.db.usa1.vertraweb.app. Port: unique per database (not 5432/27017/6379/3306) — copy it from the database page or GET /v1/databases/{id}.
• Files (Certificates tab or GET /v1/databases/{id}/credentials/certificate): certificate-<id>.ca.pem = the CA that signs the server certificate (the CA file everywhere), certificate-<id>.crt = client certificate (not the CA), certificate-<id>.key = client private key, certificate-<id>.pem = bundle of client key + client certificate + the CA (MongoDB’s certificate-key file), certificate-<id>.p12 = client identity as PKCS#12 with an empty password (Prisma 6). Older databases may not offer the .ca.pem; there, the .pem works as the CA file.
• Common errors: self-signed certificate in certificate chain = wrong or missing CA (usually the .crt passed as the CA); key values mismatch = .crt and .key are not a pair; connection requires a valid client certificate = .crt/.key not sent. Full table in Databases › Why does my database connection fail?
• The server certificate matches the host, so full verification works.

​

Data manager (the “Data” tab)

​

4. Plans and Pricing

​

vCPU per Application

• Growth: on plans with up to 2 vCPU, +1 vCPU for every 512 MB of RAM; on plans with more than 2 vCPU, +1 vCPU for every 1 GB (1,024 MB).
• Floor: a project with 512 MB or more gets at least 2 vCPU; below 512 MB, at least 1 vCPU.
• Ceiling: never more than the plan’s vCPU. Starter 1 (1 vCPU) gives every project 1 vCPU.

​

AI Agents (MCP)

​

Plan Names

​

Features by Plan

Previous versions kept counts the versions besides the one running now. Version validity counts from the moment a version stops being the running one; the running version never expires.

Manual backups per day has two limits, counted on the calendar day in Brasília time. Per project (application or database): up to 5. Per account, across all projects: 2 per GB of plan RAM, at least 5 and at most 64, which is the value in the table: 5 on Starter 1 and 2, 8, 12 and 16 on Pro 4, 6 and 8, 24, 32 and 48 on Business 12, 16 and 24; every Enterprise size has 64. Daily backups are automatic and safety backups are taken by the platform; neither uses these limits. The 5-minute interval between manual backups of the same project still applies.

Automatic rollback and Zero-downtime deploys are deploy settings, changed in the application’s settings or with PATCH /v1/apps/{id}/config (deploy). Asking for either on a plan that doesn’t include it fails with 403 PLAN_DOES_NOT_SUPPORT_AUTO_ROLLBACK or 403 PLAN_DOES_NOT_SUPPORT_ZERO_DOWNTIME; turning them off is always allowed.

The file manager is the dashboard’s Files tab and its HTTP routes; ownership and path rules follow the same rules across the API, the CLI and the MCP.

• Publish (every plan) — having a site: the application gets a *.vertraweb.app subdomain and becomes reachable over the internet. The site’s HTML pages show the “Powered by Vertra Cloud” badge in the bottom-right corner on Starter 1 only; from Starter 2 up it doesn’t appear. Switching plans turns the badge on or off immediately, without a redeploy.
• Choose the subdomain (every plan) — choosing the subdomain name, at creation or later (PATCH /v1/apps/{id}/network/subdomain).
• Custom domains (Starter 2+) — pointing domains you own to the application, as many as the plan allows (see Custom Domains).

​

Account Without an Active Plan

• Projects offline for 30+ days are automatically deleted, with a snapshot taken first
• Email notifications on days 7, 23 and 27 before deletion

​

Billing Periods and Payment

• Periods: 1, 3, 6 or 12 months, on every plan. A 3-month purchase gets 10% off the monthly price, a 6-month purchase 15% off and a 12-month purchase 25% off.
• Payment methods: PIX or credit card (Visa and Mastercard), on every plan, Starter 1 included. A coupon can restrict an order to one method.
• Prices are always in R$. The website shows an approximate US dollar amount next to the price (≈ $) for reference only; the charge is in reais.
• The price before you pay. POST /v1/orders/quote returns what will be charged for a plan and period, with an optional coupon: the list price, the period discount, the coupon discount, the total and whether the price is the transition price below. It creates no order and doesn’t use the coupon. The checkout shows this same value.
• Card = automatic renewal. A card is saved to the account and the plan renews on the default card, with no loyalty period. Three days before the plan expires the card is charged and an email is sent the same day. If the charge fails it is retried on each of the next 2 days and on the expiration day; if every attempt fails, the plan expires normally. If the bank asks for confirmation, you receive an email to confirm the payment. To stop automatic renewal, use Cancel renewal on the Plan card in Billing: the plan stays active until its expiry date and the card stays saved.
• PIX doesn’t renew. You only get the usual expiration notice and pay again when you want to.
• Unpaid orders expire. An order that isn’t paid stays open for 72 hours after it’s created, then it expires and can no longer be paid. You get an e-mail when an order is left open, and reminders with 48 hours and 24 hours left.
• PIX QR codes last up to 30 minutes. If a QR code expires while the order is still open, generate a new one for the same order.
• Billing details are required to buy a plan. They are needed to issue the service invoice (NFS-e). In Brazil: name, CPF or CNPJ and the full address (CEP, street, number and neighborhood; city and state are filled from the CEP). Outside Brazil: name, street, city, country and postal code. The phone is optional. Set them under Billing; if they are missing, the checkout asks for them before payment. Through the API, an order without them fails with BILLING_DETAILS_INCOMPLETE.
• Where payment data lives. Card number, expiry and CVV are typed into a Stripe field and tokenized in the browser; they never reach Vertra. Billing details (name, address, phone, CPF/CNPJ) are stored by Stripe, not by Vertra. Vertra keeps only the card brand, last 4 digits and expiry, to list saved cards.
• Refunds. The first purchase of a paid plan can be refunded in full within 7 days. An automatic card renewal can also be refunded in full within 7 days of the charge; the plan goes back to its previous expiration date. Renewals you make yourself, upgrades and plan changes aren’t refundable. Requests go to [email protected].
• Invoices (NFS-e): the Brazilian service invoice (NFS-e) is issued automatically for every paid order and delivered by email, issued in the name and CPF/CNPJ of your billing details. A full refund cancels the invoice.

​

Move from the Previous Plan Grid

• Nothing about the time you paid changes. The expiration date stays the same, and time paid in advance stays valid. No new amount is charged before it.
• Memory and vCPU. Each application and database keeps its memory: you choose it. The vCPU ceiling of the new plan reaches each one on its next restart or deploy (see vCPU per Application).
• Renewing in the transition period. Until the date given in the notice e-mail, you can renew the equivalent plan for 1 month at the monthly list price of your previous plan (without any period discount or coupon you used) or at the new monthly price, whichever is lower, already with the features of the new plan. An automatic card renewal before that date follows the same rule and covers 1 month, even if your previous period was 3, 6 or 12 months; the following automatic renewals are monthly, at the current price, until you buy another period. Purchases of 3, 6 or 12 months you make yourself use the new price, with the discount of the new periods, shown at checkout. After that date, every renewal is charged at the new price.
• Transition coupon. The accounts that received the notice e-mail also received, in it, a coupon link for the first monthly purchase of the new plan made between the transition date and the coupon’s expiry date given in the e-mail. Before the transition date the monthly renewal already uses the transition price, so the coupon isn’t accepted yet (COUPON_NOT_STARTED). The coupon is valid only for that account, only on the monthly period, for one use. See Discount Coupons.
• Unpaid orders. An open order created before the change keeps the amount it had, and activates the equivalent plan of the new grid when paid.

​

5. Application Types

• Turning it on and off recreates the container, so the application restarts — network rules and the exposed port only exist from the moment the container is created.
• When turning it off, the subdomain is released: there is no reservation, someone else can claim it. Every custom domain is also removed, and its CNAME and validation TXT records stay at the user’s DNS provider until the user deletes them.

• Build context: ZIP up to 100 MB, or a GitHub repository, with a file named exactly Dockerfile at the root. Up to 2 images per account on Pro, 6 on Business and 20 on Enterprise, counting every state. Platform linux/amd64.
• Build: isolated build environment, time limit of 26 minutes, built image up to 10 GB uncompressed (IMAGE_TOO_LARGE). .env* files are removed from the context.
• Runtime: always user 1000:1000. /app is the persistent disk; /tmp, /var/tmp and /run are writable but temporary; everything else is read-only. The port is the first EXPOSE or 80, injected as PORT.
• Refused: a # syntax= other than the official frontend, RUN --security=insecure, RUN --network=host, --mount=type=ssh, --mount=type=secret. Warned: WORKDIR /app, .env* in the context, COPY/ADD of .env, VOLUME outside /app.
• Updates: rebuild by upload, commit (on top of the current context) or push to a linked GitHub branch. A rebuild restarts apps only when the image option auto_update_apps is on (on for images created from GitHub, off for the others; change it on the image page or with PATCH /v1/images/:image_id): then every successful build restarts the image’s running apps on the new version, stopped apps stay stopped and a failing app does not block the others. Each app keeps 30 releases for 30 days, and rollback uses an already built version.
• Snapshots hold the data of /app and the image version; restore can optionally go back to that version.
• Deleting an image deletes every application created from it. Starter plans have no image slots and cannot create or rebuild images; a downgrade to a plan with fewer image slots than the images in use (own and image templates) is refused: delete images first.

​

Storage

​

Bandwidth Limits by Memory

​

Lifecycle

Creating (building) → Online, or Stopped if the install or build fails
Online → Stopped (manual) or Error (crash)
Error → Starting (auto-restart) or Stopped (crash loop: 5x in 10min)

​

Build Command (build_command)

​

Configuration File (vertracloud.config)

NAME=My App
MAIN=index.js
BUILD=npm run build
START=npm start
SUBDOMAIN=my-app
VERSION=22
MEMORY=512

​

Subdomains

• Format: your-app.vertraweb.app
• Unique per application
• Web publishing (getting a subdomain) is available on every plan
• Choosing the name of the subdomain is available on every plan, at creation or later
• Accepted format when choosing: 3 to 50 characters, ^[a-z0-9][a-z0-9-]*[a-z0-9]$
• Renamable later via PATCH /v1/apps/{id}/network/subdomain (minimum 5 minutes between changes)

​

Custom Domains

• Available from the Starter 2 plan. Domains per application (the automatic www or root companion does not count): Starter 1: 0; Starter 2: 1; Pro: 2; Business: 5; Enterprise: 20
• The application must be published on the web. Adding, changing or removing a domain does not restart or redeploy it
• Configure the DNS records shown for the domain (a CNAME and, usually, a certificate validation record). Automatic DNS setup is not available yet: POST /v1/apps/{id}/domains/{domain_id}/connect always answers DOMAIN_CONNECT_UNAVAILABLE
• SSL provisioned automatically after DNS validation
• Automatic www: adding example.com also adds www.example.com (and the other way around). The companion redirects to the main domain with a 308, doesn’t count toward the limit, is removed with the main domain, and can be disabled with www: false. Either one can be made the main domain (primary: true)
• Root domains need a DNS provider that supports CNAME flattening, ALIAS or ANAME; the dashboard warns about it
• Port per domain: any port from 1 to 65535 (default 80), up to 5 different ports besides 80 per application (DOMAIN_PORT_LIMIT_REACHED beyond that)
• Path prefix per domain (literal path such as /api, starts with /, no trailing /): added in front of every request. Recommended for APIs; a website has to run under that path
• Redirect per domain to another domain of the same application, with status 301, 302, 307 or 308 (default 308), keeping the path and query string. A redirect can’t point to another redirect
• Status per domain: pending (waiting for DNS), active, moved (DNS stopped pointing at Vertra Cloud) or deleted; plus the certificate state (pending, active, failed)
• A domain that stays pending, moved or deleted for 14 days is removed automatically and the owner is notified by email
• Unpublishing the application removes all of its domains
• Clean up your DNS after a domain goes away (removed by you, by unpublishing, or by the 14-day cleanup): delete its CNAME and its validation TXT records at your DNS provider. Removing a domain does not touch your DNS; if the records stay, the name keeps pointing at Vertra Cloud and another customer could claim it
• Routes: GET/POST /v1/apps/{id}/domains, PATCH/DELETE /v1/apps/{id}/domains/{domain_id}, GET /v1/apps/{id}/domains/{domain_id}/dns, POST /v1/apps/{id}/domains/{domain_id}/connect. Reads need the apps:read scope; writes need apps:write

​

6. Auto-Restart and Crash Loop Detection

• Containers that crash are restarted automatically — the default on every plan, always on, with no option to turn it off
• Exit code 0 (normal termination) does NOT trigger a restart. Neither do errors a restart can’t fix: syntax errors, missing modules, out of memory, a failed install or build, or a Vertra Shield pause
• Every other crash is restarted, even if the application was up for only a few seconds, with a growing wait between attempts: immediately, then 10 s, 30 s, 60 s and 2 min
• Crash loop: up to 5 automatic restarts in a 10-minute window. The 6th crash in that window suspends auto-restart for 24 hours: the application stays stopped and the dashboard says why
• Starting or restarting the application manually clears the crash count and lifts the suspension
• An application that stays up for 10 minutes leaves the window, so older crashes stop counting

​

7. Platform Protections (Vertra Shield)

​

Request rate limit per visitor IP

​

Shield Cooldown (visible pause)

​

Regional access restrictions (*.vertraweb.app)

• Blocked (sanctions or regulatory restrictions): Afghanistan (AF), Belarus (BY), Cuba (CU), China (CN), Iran (IR), North Korea (KP), Russia (RU), Syria (SY), Antarctica (AQ) and all Tor exit nodes (T1).
• Browser check (security reasons): Bangladesh (BD), India (IN), Iraq (IQ), Philippines (PH), Ukraine (UA), Venezuela (VE), Vietnam (VN), Indonesia (ID) and Malaysia (MY). Only a person in a browser passes the check; API clients and webhook senders from these countries receive 403.

​

8. SSL/TLS

• All sites get automatic SSL/TLS certificates
• Automatic renewal before expiration
• HTTP → HTTPS redirect enabled by default
• Custom domains also get automatic SSL
• All database connections require TLS

​

9. Snapshots

• Daily backup of applications and databases, on every plan, kept for 30 days like every snapshot. It doesn’t use the manual limit below
• Manual snapshot creation for apps and databases, on every plan — at most 1 every 5 minutes on the same project, up to 5 a day per project, and a daily limit across the whole account that depends on the plan: 2 per GB of plan RAM, at least 5 and at most 64 (5 on Starter 1 and 2, 8 to 16 on Pro, 24 to 48 on Business, 64 on Enterprise). The day is counted in Brasília time and the limits reset at midnight. Exceeding any returns SNAPSHOT_RATE_LIMITED (429) with details: { scope: "resource" | "resource_daily" | "account", retry_after }; resource is the 5-minute interval, and resource_daily and account also carry limit and resets_at
• Restore from any earlier snapshot (30-second cooldown)
• Download the snapshot content as a ZIP, of any size. Downloading the project directly (GET /v1/apps/{id}/download) is limited to 300 MB of files (DOWNLOAD_TOO_LARGE): above that, download a snapshot
• Resetting a database automatically takes a safety snapshot before destroying the data, on every plan. If the backup fails, the action is blocked (SNAPSHOT_FAILED) and the database is not reset. This forced snapshot has its own ceiling, separate from the manual-creation one: at most 2 attempts per hour on the same project, and 10 per hour across all projects on the account — exceeding either returns FORCED_SNAPSHOT_RATE_LIMIT and also blocks the action.
• Deleting an application or a database does not create a snapshot on its own. The panel’s delete dialog has the option Create snapshot before deleting, off by default (shown on plans with manual backups): when on, the panel creates the snapshot and deletes only if it succeeds; when off, it deletes right away. Through the API, DELETE on an application or database deletes right away: to keep a backup, create a snapshot first.
• Deleting a whole stack or a custom Docker image (which deletes its apps) still takes the safety snapshot automatically, with the same ceiling. Deletions the platform makes on its own (expired plan, account without a plan) also snapshot every project first; if the snapshot fails, the resource is stopped and kept, not deleted.
• Snapshot size limit: 4 GB (the size of the snapshot file, already compressed). Above that, creation is refused with SNAPSHOT_TOO_LARGE (413) instead of a generic error. It applies to manual, daily and safety snapshots. Downloading has no such limit: any existing snapshot can be downloaded.

​

10. Workspaces

​

The 20 permissions

​

Seed roles

​

Custom role ceiling by plan

​

Members by plan

​

Invites

• Email: tied to a specific address, expires in 7 days. WORKSPACE_INVITE_EMAIL_MISMATCH if accepted by an account with a different email.
• Link: multi-use, fixed 24h expiration, optional max_uses. Revocable at any time.

​

Folders and favorites

• Personal: the account’s individual view, returned by GET /v1/users/me.
• Workspace: the member’s individual view inside a workspace, returned by GET /v1/workspaces/:id. It is not shared with the other members.

​

Action requests

​

Retention

​

Route → permission

​

Error codes

​

11. File Manager

• Available on every plan
• Browse and edit files directly in the dashboard
• Syntax highlighting for all languages
• Upload via drag-and-drop (files or ZIP); the upload dialog’s Commit destination starts at the open folder and can point to any other folder, and a ZIP is merged into it
• File tree up to 6 levels deep
• Create, edit, move, delete files and folders
• Lists every directory of the project, including dependency folders such as node_modules; only the file tree endpoint (GET /v1/apps/{id}/files/tree) skips dependency and build folders to stay small
• vertracloud.config doesn’t exist on the volume: it’s generated from the database on download and on snapshot (see §Configuration File)
• Storage: 10 GB fixed per application (see §5 Storage)

​

12. Deploy

​

Methods

1. ZIP upload via dashboard or API
2. GitHub integration — repository selection (including private ones)
3. GitHub webhooks — automatic deploy on push to the chosen branch, main or master by default (every plan)
4. VSCode extension — deploy straight from the editor

​

Process

​

Upload Limits

• Maximum ZIP size: 100 MB
• Maximum dependency install time: 15 minutes (INSTALL_TIMEOUT if exceeded)
• Maximum build command time: 10 minutes (BUILD_TIMEOUT if exceeded)
• 1 exposed port per application

​

Automatic Rollback (Pro+)

​

Zero-Downtime Deploys (Starter 2+)

• off (default): stop the running version first, then start the new one.
• auto: zero downtime only for applications that receive web traffic.
• on: always, including applications without web traffic, such as bots. Two processes of the application run together for a while, so a bot can answer the same event twice.

• Both versions use the same folder, and the two processes run together for up to about 2 minutes. Data written by both at once, such as a JSON file, can be corrupted; SQLite usually copes. A database migration run at startup runs while the previous version is still serving.
• An application that reads files on every request (PHP, static sites, a built Next.js or Nuxt app, late imports in Python) can serve a mix of old and new files while the deploy installs and builds. For those, use off.
• If the new version does not become healthy within about 2 minutes, it is discarded and the previous version keeps running. The owner gets a notice with the log of the discarded version.
• If the second copy can’t be started at that moment, the deploy falls back to stopping the running version first.
• While the new version is prepared, the terminal still shows the running version, not the build. A request the old version was already answering when it was asked to stop can fail once; handling SIGTERM gracefully avoids it.

​

Deploy-per-Hour Limit

​

Files Excluded from the ZIP

• JavaScript/TypeScript: node_modules, .npm, package-lock.json, .next
• Python: venv, .venv, .cache, __pycache__, .env
• Go: vendor
• Rust: target
• Java: target, .gradle
• PHP/Ruby: vendor

dist/ and build/ (build output) are not removed automatically — if your main/start points inside them (common in projects that bundle before deploying), the platform needs these files. Only exclude them from the zip if the real entry point is outside of them.

​

13. Plan Downgrade

• Available via dashboard or API: POST /v1/users/me/downgrade
• Formula: newDays = remainingDays × (currentPrice / targetPrice)
• Restrictions: Starter 1 cannot downgrade. A downgrade always moves to the immediately previous plan: Starter 1 → Starter 2 → Pro 4 → Pro 6 → Pro 8 → Business 12 → Business 16 → Business 24 → Enterprise 32 → … → Enterprise 512
• Checks: Memory usage and the Docker images in use (own and image templates) must fit the target plan (INSUFFICIENT_MEMORY_FOR_DOWNGRADE, INSUFFICIENT_IMAGE_SLOTS_FOR_DOWNGRADE), rate limit (1x every 3 hours)
• Impact: Reduced memory limit, higher-plan features disabled, workspaces lost when leaving Starter 2 for Starter 1. The saved automatic rollback and zero-downtime settings stay on the application but only act on a plan that includes them

​

14. VSCode Extension

• Available on the VS Code Marketplace (search “Vertra Cloud”)
• Features: Quick deploy, app management, file management, real-time monitoring
• Requirements: VS Code 1.80+, active Vertra account, internet connection
• Deploy via the command palette: Ctrl+Shift+P → “Vertra: Deploy”

​

15. Authentication Methods

• Discord OAuth — Login and account linking
• GitHub OAuth — Login and account linking
• Google OAuth — Login and account linking
• Magic Link — Passwordless authentication by email
• API key — For programmatic access (managed from the dashboard settings); up to 10 keys per account, each with a name, scopes and optional allowed IPs. New format vc_live_ + 48 hex characters, shown only once; older keys (64 hex) remain valid with all scopes. Full detail in §16.

​

16. API Reference

​

Authentication

Authorization: Bearer YOUR_API_KEY

​

API Keys

​

Rate Limiting

​

Response Format

• Success: { "response": { ... } }
• Error: { "code": "ERROR_CODE", "message": "..." }

​

Common Error Codes

​

Main Endpoints

​

Applications

​

Databases

​

Users and account

​

Workspaces

​

Billing

​

Redeem

​

Services

​

17. Environment Variables

• Injected into the container at startup
• Managed via dashboard or API
• Limits: Maximum 256 variables, key max 255 characters, value max 3,072 characters
• Encrypted at rest
• Forbidden keys: PORT, HOST, PATH, HOME, USER, SHELL
• The platform sets PORT=80 and HOST=0.0.0.0 in the container

​

18. Automatic Cleanup (Accounts Without a Plan)

​

19. Common Problems and Fixes

​

Platform

• App crashing / OOMKilled: Insufficient memory. Increase the allocated RAM. Java/Puppeteer need 512 MB+.
• Crash loop (auto-restart disabled): 5 crashes in 10 min. Check the logs, fix the root error, redeploy.
• App stops immediately (exit code 0): The process has nothing keeping it alive. Make sure the bot is logged in or the server is listening.
• ZIP too large: Exclude node_modules, venv, .git, vendor, target, __pycache__.
• Dependency installation exceeds 15 minutes (INSTALL_TIMEOUT): Remove unnecessary dependencies.
• Build command exceeds 10 minutes (BUILD_TIMEOUT): Simplify the build_command or the framework’s build step.
• Site times out: Bind to 0.0.0.0 (not localhost). Use the PORT value.
• Custom domain doesn’t work: Starter 2+ plan, correct CNAME, wait for DNS propagation. A root domain needs a DNS provider with CNAME flattening. A domain with no DNS for 14 days is removed; delete its CNAME and TXT records at your DNS provider afterwards.
• Database connection refused: TLS is required. Download certificates from the dashboard.
• PLAN_REQUIRED: The account has no active plan. Subscribe to a plan to create applications and databases or publish a site.
• Email not received: Check spam. Try another method (Discord, GitHub, Google).

​

Node.js / JavaScript

• Cannot find module 'xxx': The package isn’t in package.json’s dependencies (it may be in devDependencies and the application has no build command, since devDependencies are installed only when there is one).
• missing script: start: Add "start": "node index.js" under scripts in package.json.
• Cannot use import statement outside a module: Add "type": "module" to package.json, or use require().
• ERR_REQUIRE_ESM: The package only supports ESM. Use ESM or an older version (e.g.: node-fetch@2).
• ERESOLVE dependency tree: Create .npmrc with legacy-peer-deps=true.
• EADDRINUSE: Port already in use. Only one app.listen() in the code.
• Native modules fail (better-sqlite3): Don’t upload node_modules/. The platform compiles for Linux.

​

TypeScript

• tsx / ts-node: command not found: Add tsx to dependencies.
• TSError: Unable to compile: Include tsconfig.json. Fix TS errors before deploying.

​

Python

• ModuleNotFoundError: The module isn’t in requirements.txt. Generate it with pip freeze > requirements.txt.
• requirements.txt not found: Check the exact name and extension (watch out for .txt.txt on Windows).
• SyntaxError: invalid syntax: The selected Python version doesn’t support the feature used. Upgrade the version.
• UnicodeDecodeError: Save files as UTF-8. Use encoding='utf-8' when opening files.
• Flask/Django not reachable: Use Gunicorn: gunicorn app:app --bind 0.0.0.0:80.

​

Go

• go.mod not found: Run go mod init and go mod tidy. Include go.mod and go.sum.
• Incompatible binary: Compile for Linux: GOOS=linux GOARCH=amd64 go build.

​

Java

• no main manifest attribute: Configure the build tool to create a fat JAR with Main-Class.
• OutOfMemoryError: Java needs 512 MB+. Use flags: -Xmx256m -Xms128m.

​

Discord Bots

• Invalid token / Improper token: Generate a new token in the Developer Portal. Set it as an environment variable.
• DisallowedIntents: Enable privileged intents in the Developer Portal (Message Content, Server Members, Presence).
• ClientMissingIntents: Specify intents in your code: new Client({ intents: [...] }).
• Bot goes offline randomly: Insufficient memory or an outdated discord.js/discord.py.
• on_message doesn’t fire (discord.py): Missing intents.message_content = True + enabling it in the Developer Portal.

​

Environment Variables

• Undefined/null values: Configure via the dashboard, don’t rely on a .env file.
• .env doesn’t work: The platform doesn’t read .env. Use the dashboard’s variables interface.

​

20. Discount Coupons

​

21. Data Retention

​

22. Blob Storage

​

Quota and limits by plan

​

Features that depend on the plan

​

Limits for every plan

• Simple upload: one request up to 100 MB per file.
• Direct upload: up to 10 GB per object, sent in parts straight to storage, with resume and abort.
• Path: 1 to 1,024 characters, up to 16 folders, each part up to 255 characters of letters, digits, ., _, - and space. Every file needs an extension.
• Blocked types: exe, msi, dll, bat, cmd, com, scr, cpl, pif, hta, vbs, vbe, jse, wsf, wsh, msc, reg, lnk, sys, drv, ps1, apk, xpi, dmg, pkg, deb, rpm and appimage. Executables are also detected from their content.
• HTML, HTM, SVG, XML and XHTML are always downloaded (attachment), never shown inside the page.
• Folder rules set defaults for a folder: visibility, expiration, maximum size, extensions, Cache-Control, and deletion after a number of days (1 to 3,650).

​

Quota, downgrades and grace period

• A write that would take the account past its quota is blocked with STORAGE_QUOTA_EXCEEDED. Deleting objects frees space.
• After a downgrade, your files stay. If they use more than the new quota, new uploads are blocked and the excess is charged at R$0.20 per started GB on the next renewal.
• When the plan expires, the account enters a 72-hour grace period: uploads and other writes are blocked (BLOB_NAMESPACE_SUSPENDED), while public files, temporary links and share pages keep working. Renewing within those 72 hours reactivates everything. A suspension for abuse is different: the files go offline right away.
• After the grace period the files are deleted. Deletion is permanent and there is no backup.
• Share pages open on the website. A share can have an expiration, a download limit and, from Business, a password. A share doesn’t protect a public object.

​

Access

• API key scopes: blob:read and blob:write. Workspace permissions: blob:read and blob:manage. A key with only blob:* scopes can be restricted to one folder (blob_prefix, a path ending in /): it reaches only objects, uploads and shares inside that folder, and the S3 credentials derived from it inherit the limit. Everything else (rules, domains, statistics) answers 403 BLOB_PREFIX_DENIED. A stack creates its own restricted key for its Blob folder.
• Upload token: a short-lived vcup_ token for browsers. It fixes the path or folder, size, extensions and visibility, and it is accepted only on the upload routes.
• S3-compatible access: GET /v1/s3/credentials, limited to 10 calls per hour. Revoking the API key that created it cuts S3 access at once. S3 requests per 10 seconds per account: Starter 1 100, Starter 2 200, Pro 400, Business and Enterprise 600; CopyObject/UploadPartCopy 20. Above it: 503 SlowDown.
• Every route is documented on its own page under API Reference > Blob Storage.

​

Blob error codes

​

23. Templates and stacks

​

Origins

​

Code and image templates

​

Required plan

​

Using a template

• POST /v1/templates/{id}/deploy starts a job (202) that validates the plan and memory, creates databases and waits for them, creates the applications, writes the variables, starts everything and waits for the health check. Any failure undoes everything the job created (TEMPLATE_DEPLOY_FAILED, details.step); before deleting an application or a database the undo takes a safety snapshot. A template of a workspace deploys only inside that workspace, and only by its owner. Send an Idempotency-Key to make it safe to repeat. Only one deploy of the same template runs at a time (TEMPLATE_DEPLOY_IN_PROGRESS).
• Questions have types (text, secret, select, boolean, number, url, email); secrets can be generated for you. Outputs, such as the address or a generated token, are shown at the end; a secret output is shown once.
• Updates are never automatic: a new version shows New version available; updating takes a safety snapshot first, updates the services one by one and, if any fails, restores all of them. Permissions never become stricter in an existing application. Some templates lock the blue-green deploy (two copies running at once would break them) and require rollback to restore the data too (TEMPLATE_ROLLBACK_REQUIRES_DATA).
• Changing the code of an application from an open template detaches it from the template; the API asks for confirm_detach first (TEMPLATE_DETACH_CONFIRMATION_REQUIRED).

​

Stacks

• Applications in a stack reach databases through the public address, with client certificates managed by the platform and left out of downloads, snapshots and “save as template”. A network mode other than public returns STACK_NETWORK_MODE_UNAVAILABLE.
• A connection links an application to a database, a Blob folder or another application. Its variables are managed: updated when the database password, certificate or host changes, and marked in the dashboard. Editing one by hand turns it into a normal variable. The API returns restart_required and apps_to_restart when applications must restart.
• A resource belongs to at most one stack, in the same account or workspace as the stack. A stack can be created from a template, from the /new flow, by adding a database to an application, or by transforming a folder that holds an application and a database.
• Leaving a stack: keep the connection (variables become normal ones with the public address) or remove them. Deleting a stack asks whether to only undo it (the resources stay) or delete everything; for the Blob folder the choice is keep (default) or delete.
• Folders organize, stacks connect: a stack has its own card in the list and its own box with the connections in the Flow, and it can live inside a folder.

​

Share links

​

Closed templates

​

Manifest limits

​

Template and stack error codes

​

24. Glossary