The best Firebase alternatives, compared honestly
Firebase is still the fastest way to get an authenticated, real-time, offline-capable app in front of users, and Crashlytics and App Check have no real equivalent anywhere on this page. What sends teams looking is the billing model, and it is more specific than "Firebase got expensive". Firebase is not one product on one meter: it is roughly ten products on roughly thirteen different billing units, and on the flagship the unit is the document read, not the query. Google's own documentation says there is a minimum charge of one document read for every query even when it returns nothing, that index entries bill one read per 1,000, and that a query with an offset of 10 returning a single document is charged 11 reads. So the meter tracks your interface, not your data, and Google's own worked example, priced at multi-region rates, puts $261.90 of a $292.02 monthly bill in read, write and delete operations. Then comes the part almost no comparison mentions: on Blaze, you cannot cap it. This page compares seven alternatives on the one column that decides the migration.
The best Firebase alternative depends on which part of the model you are escaping:
- You want a hard ceiling on the invoice → Supabase: the Pro plan ships a Spend Cap enabled by default, which is precisely what Blaze does not offer for Firestore, Storage or Auth.
- You want a permissive licence with no carve-out → Appwrite: BSD-3-Clause, licence file untouched since 2019, no enterprise directory in the repository, and a published per-100k database read and write meter.
- You specifically want Firestore's reactive queries → Convex: the closest analogue on the list, but it is FSL licensed rather than open source, and seat fees sit outside your spending limit.
- You want no vendor meter at all → PocketBase: MIT, one binary, self-hosted only, and its own docs say it is not yet recommended for production-critical apps.
Why teams look elsewhere
What pushes teams off Firebase
Every claim below comes from Google's own documentation rather than from a complaint thread, because the interesting thing about Firebase's billing is that none of it is hidden. It is all published, in six different places, in units that do not convert into each other.
The meter is a document read, not a query
A list screen showing 500 rows is 500 reads, every time it loads. On top of that: one read per batch of 1,000 index entries, a minimum of one read for a query returning nothing, 11 reads for a query with an offset of 10 that returns one document, extra reads for every get() in your Security Rules, and a listener re-billed as a brand-new query after a 30-minute disconnect with offline persistence on, or on every single reconnect with it off. Google's own example: 90% of the bill is operations.
On Blaze you cannot cap the products that bill you
This is the strongest single fact on the page and it is Firebase's own wording. Budget spend caps exist for exactly four products: AI Logic, App Hosting, Cloud Functions and Extensions. "For the other products that have a paid tier of usage, you cannot cap your usage when on the Blaze pricing plan." Firestore, Realtime Database, Cloud Storage, Hosting and Authentication have no cap at all.
Budget alerts are a notification, not a control
Verbatim from Google: "Budgets and budget alerts do not cap your usage or charges." And the four caps that do exist "are not 'hard caps'. Due to latency in usage and cost reporting, spend cap enforcement isn't instantaneous", with the docs adding that "you remain responsible for any overage costs accrued due to reporting latency".
Every product is on a different unit
Firestore bills reads and writes per 100k. Realtime Database bills $5 per GB stored and $1 per GB downloaded, and where Firestore exempts rule-denied requests from network bandwidth only, Realtime Database bills "all traffic to and from your database, including operations denied by security rules". Auth bills two separate MAU tiers, one per account and one per project, plus SMS priced per country from $0.01 in the US to $0.35 in Indonesia. Firestore Enterprise edition bills Read Units in 4 KiB tranches, so even the same product name has two incomparable rate cards.
One irreversible decision at project creation
A multi-region Firestore database costs exactly double per operation: $0.06 per 100k reads against $0.03 in us-central1. And per Google's docs, "once you provision a database instance, you cannot change its location setting". You pick the multiplier on day one, before you have any traffic data, and you keep it forever.
A steady stream of forced migrations
Dynamic Links shut down on 25 August 2025 and every link now returns HTTP 404, with only commercial third parties offered as the migration path. Since 3 February 2026 a Spark project with a legacy bucket loses console access and gets HTTP 402 and 403 from the API. Studio sunsets 22 March 2027, Extensions 31 March 2027, Firebase ML 15 June 2027. Remote Config is a metered product as of 1 September 2026.
The shortlist
7 Firebase alternatives worth evaluating
Nothing on this list replaces Firebase completely, and any comparison claiming otherwise has not checked. Crashlytics, App Check, Analytics, Test Lab and App Distribution have no equivalent in any of these products, so a realistic migration keeps a Firebase project for those and moves the database, auth and functions. The ranking below therefore optimises for the backend triangle of data, auth and functions, and each entry lists one honest strength and one real weakness rather than a feature grid.
It answers the specific complaint at the top of this page: "The Pro Plan has a spend cap enabled by default to keep costs under control." The units are Postgres-native and mostly volume-based rather than per-operation, so the bill tracks data and traffic instead of screen renders, and the licence audit is clean: Apache-2.0, MIT and the PostgreSQL Licence, with no BSL or SSPL anywhere. Free gives 500 MB of database, 5 GB egress, 50,000 monthly active users and a limit of 2 active projects, with projects paused after a week of inactivity. Pro is from $25 a month, billed monthly, with the first project included and additional projects from $10 a month, including $10 of compute credits that "reset monthly and do not accumulate". Weakness: nothing here reaches $0 when idle. "Compute is charged by the hour ... If a project runs for part of an hour, you are still charged for the full hour", and "Pro-Projects at the moment cannot be paused", so a dormant staging project still costs the $25 plan fee, since the $10 Micro compute is simply absorbed by the included credits. Its own docs also list what self-hosting does not include: branching, managed backups and PITR, ETL and the platform management API.
The licence is the cleanest on the list and it is verifiable rather than claimed: BSD-3-Clause, with the licence file touched only twice, both times in May 2019, and byte-identical between the current stable tag and the 2.0 release candidate. There is no ee/ directory in the repository. It is also the only alternative here that meters database reads and writes explicitly, at $0.060 and $0.10 per 100k on Pro, which makes it directly comparable to Firestore rather than approximately comparable. Pro is "from $25 a month", billed in 30-day periods. Weakness: single sign-on is not available below Enterprise, and Appwrite's own pricing page cannot agree with itself about Enterprise: the plan-comparison table shows a dash for Free, a dash for Pro and "Coming soon" for Enterprise, while the Enterprise card lists Single Sign-On as an included bullet. Nothing scales to zero, the self-hosted support tiers have no published rate, and the enterprise page claims 99.99% uptime where the SLA document says 99.95%.
If what you actually love about Firestore is that a query is a live subscription, this is the nearest thing to it, with a published rate for every unit and an unusually honest idle policy: "The deployment fee is waived for paused and idle deployments. Idle deployments are defined as those that have not had any function calls for 30 days." Weakness, and there are two. The licence is FSL-1.1-Apache-2.0, which is source-available with a competing-use ban and becomes Apache-2.0 only two years after each release, so this is not an open-source escape from a proprietary one. And the meter is a per-developer seat at $25 a month that sits outside your own cost controls: "if you send the spending limit to $0/month, you will be billed for the seat fees only". Note also that pricing is geo-served and stated as such, with EU deployments at 1.3 times US rates and no way to change a deployment's region later.
One binary, SQLite inside, MIT licensed, and the README says plainly "You are free to do whatever you want with it, even offering it as a paid service". There is no vendor cloud and therefore no published rate for anything: your bill is whatever your own server costs. For a small app that is the whole appeal. Weakness: the project says the quiet part itself. "PocketBase is NOT recommended for production critical applications yet, unless you are fine with reading the changelog and applying some manual migration steps from time to time." It is also a one-person project by its own description, with 2,444 commits from the maintainer against 5 from the next contributor, no company behind it, donations declined, pull requests currently closed to non-collaborators after LLM spam, scaling that is "only on a single server, aka. vertical", and a release cadence to match: v0.40.1 on 24 August 2026 is both the latest tag and the latest commit.
MIT licensed, actively shipping rather than winding down: in June 2026 it released Constellation, an open-source Hasura-compatible GraphQL engine, with its last commit on 1 September 2026. If your team wants Postgres plus a GraphQL API plus auth and storage as one managed unit, this is the shortest path to it, at "from $25 a month", billed monthly, with $15 of compute credits included. Weakness: the included $15 of compute credits buys exactly one shared vCPU core at the published $15 per vCPU per month, so Pro's floor is a single shared core; a dedicated core is $50 a month ($0.0012 per vCPU per minute against $0.00034 shared), and there is no tier between Pro at $25 and Team at $599. At 9,281 GitHub stars it is also an order of magnitude smaller than Supabase or Appwrite, which matters for community answers at 2am.
The client libraries are Apache-2.0, and the underlying services genuinely do approach zero when idle: DynamoDB on-demand and Lambda bill per request. If your organisation already has an AWS account, a security review and a bill, this removes a vendor rather than adding one. Weakness: Amplify's pricing page prices only hosting and builds, and routes every other backend line to somebody else's rate card, with "Auth powered by Amazon Cognito: see Cognito pricing details", data to DynamoDB, APIs to AppSync, functions to Lambda. Its own documentation answers "How much does it cost to operate Amplify Gen2?" by linking back to that hosting page. The free tier is a moving target: the "Deploy an app" row is labelled "Free for 12 months" while the per-unit allowances are monthly and unqualified, and AWS's own note gives accounts created since 15 July 2025 up to $200 in credits on a free plan lasting 6 months. DynamoDB's non-expiring free tier applies only to provisioned capacity and not to on-demand tables, and Amplify Gen 1 "has entered maintenance mode and will reach end of life on May 1, 2027".
Parse is the historical answer to "what happens when a backend-as-a-service is switched off": Facebook retired the hosted service on 28 January 2017, the community kept the server, and it is still Apache-2.0 with 21,412 stars, and parse-server, Parse-SDK-JS and parse-dashboard were all pushed on 1 September 2026. Managed hosting is available from Back4App from $25 a month per app billed monthly, or $15 billed annually. Weakness: the meter is broad rather than hidden. Back4App counts every SDK or REST call as a request, and its own definition includes logins, cloud-function calls, push-send requests and even loading your classes in the dashboard, so a chatty client burns the 500k allowance far faster than a naive read count suggests. Its own pricing page also contradicts itself on how many free projects you get, one in the table and five in the FAQ, and the Parse server's default branch is alpha, with an LTS policy stating "we may leave certain vulnerabilities up to the community to fix".
Deliberately left out: Firebase SQL Connect (renamed from Data Connect in April 2026), because it is not an alternative to Firebase, it is Firebase; it appears in the table below only as the one Firebase product that cannot scale to zero, since it needs a Cloud SQL instance that "starts as low as $9.37 / month". Hosted Parse at parse.com has been dead since January 2017, so only the open-source successor is rankable. Ory is authentication only and Directus is a headless CMS over a database you already run, so neither answers this query. Firebase Studio is a date in the section above rather than a comparison row.
Side by side
Firebase alternatives compared
Prices as of September 2026 from each vendor's own pricing page, with the billing cycle labelled on every figure because two of these vendors default their pricing page to an annual toggle. The column that matters is "What stops the bill": a plan with a real cap and a plan with a budget alert feel identical until the day traffic spikes, and only one of them protects you.
| Platform | Billing unit | What stops the bill | Free tier | Cheapest paid (cycle) | Licence |
|---|---|---|---|---|---|
| Firebase | Document reads, writes, GB, MAU, invocations | ✗ Nothing on Firestore, RTDB, Storage, Hosting, Auth | 50k reads/day, 1 GiB, 50k MAU | Blaze: pay as you go | ✗ proprietary |
| Supabase | Hourly compute, MAU, GB | ✓ Spend Cap, on by default on Pro | 500 MB db, 50k MAU, 2 projects | $25/mo, monthly | ✓ Apache-2.0 / MIT / PostgreSQL |
| Appwrite | DB reads/writes per 100k, GB, executions, MAU | Free freezes read-only; Pro uses prepaid credits | 75k MAU, 500k reads, 2 projects | from $25/mo, 30-day periods | ✓ BSD-3-Clause |
| Convex | Per-developer seat, function calls, GB-hours | Free errors on new calls; paid limit excludes seats | 1M function calls, 0.5 GB | $0 + usage (Starter); $25/dev/mo (Pro), monthly | ✗ FSL-1.1-Apache-2.0 |
| PocketBase | None: your own server | ✓ Your hardware, and no vendor invoice | Entirely free, self-host only | No vendor cloud exists | ✓ MIT |
| Nhost | Compute credits, GB | $15 credits = 1 shared vCPU, then $0.20/GB and $0.10/GB overages | 1 project, 1 GB db, paused at 1 week | from $25/mo, monthly | ✓ MIT |
| AWS Amplify | Build minutes, GB, plus 4 other services | ✗ Nothing; fixed items only add to it | Monthly per-unit allowances; account free plan 6 months | Pay as you go | Clients Apache-2.0, services proprietary |
| Parse Platform | Requests per month per app (Back4App) | Overage bands; a request is any SDK or REST call, logins included | 25k requests/mo, USA region only | $25/mo per app, monthly | ✓ Apache-2.0 |
Numbers that do not fit in cells: Firestore us-central1 is $0.03 per 100k reads, $0.09 per 100k writes and $0.15 per GiB-month, and exactly double per operation in a multi-region you can never change. Realtime Database is a different meter again at $5 per GB stored and $1 per GB downloaded after 360 MB a day. Firestore Enterprise edition bills Read Units in 4 KiB tranches at $0.05 per million. Cost of an idle project with no traffic: Firebase $0, Supabase $25 a month for the first project, then about $10 for each additional one, Appwrite $25, Convex $25 per developer, Nhost $25, Amplify $0.023 per GB a month, a charge that "recurs until the app is deleted", plus a $15 monthly per-app surcharge if WAF is on, before AWS WAF's own charges, Back4App $25 per app, PocketBase whatever your VPS costs. Pricing and free tiers change often; check each vendor for current terms. Compiled September 2026.
Official pages: Firebase pricing · Firestore pricing · Firebase FAQ (spend caps) · Avoid surprise bills · Firebase release notes · Supabase pricing · Appwrite pricing · Convex pricing · PocketBase FAQ · PocketBase docs · Nhost pricing · Amplify pricing · Parse Server
A fair call
When Firebase is still the right choice
The honest accounting is not "Blaze is uncapped so leave". It is that no product on this page replaces Firebase whole, the free daily quota is genuinely generous, and one of the criticisms above turns into a defence when you look at it from the other side.
Firebase is still right if…
- You are on Spark and want a guarantee, not a cap. This is the disadvantage that is actually an advantage: exceeding a Spark quota shuts that product off for the rest of the calendar month, so Spark literally cannot bill you. No paid plan on this page offers that, and Supabase's default Spend Cap is the closest anyone comes.
- You rely on Crashlytics, App Check, Analytics, Test Lab or App Distribution. None of the seven alternatives has an equivalent, and none of those five products carries a deprecation notice. Most realistic migrations keep a Firebase project for exactly this.
- You are mobile-first and need offline persistence that just works. The SDKs' local cache and conflict handling are more mature than anything else here, and rebuilding that behaviour is application work you will not enjoy estimating.
- Your reads are small documents fetched by key. 50,000 free reads a day is a lot when your access pattern is not "render a 500-row list on every screen".
Look elsewhere if…
- You have spiky public traffic on Blaze. There is no ceiling to set: Supabase is the only managed option here that turns a spend cap on for you by default.
- Your read volume is driven by list and search screens. Per-document billing punishes exactly that pattern, and a Postgres-backed option bills the query rather than the rows it returned.
- You need a permissive licence on paper. Firebase is proprietary end to end: that points at Appwrite (BSD-3-Clause), PocketBase (MIT) or Parse Platform (Apache-2.0), and away from Convex, whose backend is source-available rather than open source.
- You need SSO or SAML. Firebase charges $0.015 per monthly active user above 50 on its OIDC and SAML tier, per project. Check the alternatives carefully here too: Appwrite offers no SSO below Enterprise, and its own pricing page contradicts itself on whether Enterprise has it, and Convex gates it behind a $2,500 monthly minimum.
- You built on Dynamic Links, Extensions or Firebase ML. Those decisions have already been made for you, with dates: August 2025, March 2027 and June 2027.
Common questions
Common questions about Firebase alternatives
What is the best Firebase alternative in 2026?
Supabase, for most teams, because it is the only managed option on this list that ships a spend cap enabled by default on its paid plan, which is the exact control Firebase's own FAQ says Blaze does not offer for Firestore, Storage or Authentication. If your priority is a permissive licence with no enterprise carve-out, Appwrite is BSD-3-Clause and its licence file has not changed since 2019. If you want Firestore's reactive queries specifically, Convex is the closest analogue, though it is FSL licensed rather than open source. If you want no vendor meter at all, PocketBase is MIT and self-hosted only.
Why do teams leave Firebase?
Because the meter tracks the user interface rather than the data. Firestore bills per document read, and Google's docs are explicit that there is a minimum charge of one document read for every query even when it returns no results, that index entries are billed one read per 1000, and that a query with an offset of 10 returning one document is charged 11 reads. Google's own worked example for a 100,000 daily-user chat app puts $261.90 of a $292.02 monthly bill in read, write and delete operations, and it is priced at multi-region rates. The second reason is that on the Blaze plan you cannot put a ceiling on any of it.
Can I set a spending limit on Firebase?
Not on the products that generate most bills. Firebase's support FAQ lists budget spend caps for exactly four products: Firebase AI Logic, App Hosting, Cloud Functions and Extensions. For everything else it says that on the Blaze pricing plan you cannot cap your usage. Firestore, Realtime Database, Cloud Storage, Hosting and Authentication have no cap. Google also documents that budgets and budget alerts do not cap your usage or charges, and that even the four spend caps are not hard caps because of reporting latency, leaving you responsible for overage accrued in the gap.
Is the Firebase free Spark plan safe from surprise bills?
Yes, and that is genuinely Spark's strongest feature: it cannot bill you, because exceeding a no-cost quota shuts that specific product off for the remainder of the calendar month rather than charging overage. The trade-off is that Spark is not a smaller version of Blaze. Cloud Functions, Cloud Storage and App Hosting all read Not applicable on Spark, so several core products simply do not run there, and since 3 February 2026 a Spark project with a legacy storage bucket loses console access and its API calls return HTTP 402 and 403 errors.
Is there a truly open-source Firebase alternative?
Yes, and the licences vary more than the marketing suggests. Appwrite is BSD-3-Clause with no enterprise directory in the repository and a licence file untouched since 2019. PocketBase is MIT and says explicitly that you may offer it as a paid service. Parse Platform is Apache-2.0. Supabase is a mix of Apache-2.0, MIT and the PostgreSQL Licence with no BSL or SSPL component, though its CLI repository has no licence file at the root, so GitHub reports none, and the MIT text sits one level down at apps/cli-go/LICENSE. Convex is the exception: its backend is FSL-1.1-Apache-2.0, which is source-available with a competing-use ban and converts to Apache-2.0 only two years after each release.
Which Firebase products are being shut down?
Dynamic Links shut down on 25 August 2025 and every link now returns HTTP 404, with only third-party vendors offered as a migration path. Firebase Studio stopped accepting new signups on 22 June 2026 and sunsets on 22 March 2027. Firebase Extensions sunsets on 31 March 2027, after which deployed extensions keep running but can no longer be updated, reconfigured or uninstalled. Firebase ML shuts down on 15 June 2027 and is already closed to new projects. Core products including Firestore, Realtime Database, Authentication, Hosting, Crashlytics and App Check carry no deprecation notice.