Airbrake has been doing one job for well over a decade: catching exceptions and telling you which route they hurt. PostHog is a product analytics platform that added error tracking as one more product. You'd expect the veteran to win on its own ground, and in my test it didn't. PostHog reopened the one bug that came back after a release and kept two different crashes apart, while Airbrake merged those crashes and reopened every error it had after a deploy.
The specialist still won where it specialises. Airbrake is a hosted error tracker whose Node notifier also times every server route, so each error shows what share of its route's traffic it broke, and its search breaks any error down by user, release and route with counts. It has no free plan, and the trial asks for a card.
PostHog is a product platform where error tracking shares a project with replay, logs, tracing and analytics, so an error can open its recording, its server trace and its logs. Getting there takes more switches than any tracker I've tested.
The two covered different ground in my test, and that shapes every verdict below. PostHog watched the storefront and the API through three releases. Airbrake sat on the API only, for releases 1.0.0 and 1.1.0; its browser notifier stayed out of the test. Server errors and the regression release are the shared ground. Browser results and the slow third release are PostHog's alone.
The sections run from setup and capture through the server error page, the browser, grouping, search, releases, performance, logs, alerts and AI, ending with cost.
PostHog vs Airbrake: Results at a glance
Every verdict comes from one Vite storefront and one Express API, with each tool set up from its own docs. Airbrake's browser notifier, or a source path it can download from, could change a couple of rows.
Area
Better in my test
Why
Setup
PostHog, narrowly
Free with no card, more switches; Airbrake wants a card before the trial
Server error capture
Airbrake, narrowly
It caught the unhandled rejection PostHog missed
Debugging a server error
PostHog
Code lines from the server's files plus a linked trace; Airbrake shows raw JSON
Browser errors and replay
PostHog only
Airbrake's browser notifier wasn't tested
Error grouping
PostHog
Two unrelated crashes stayed apart; Airbrake merged them
Search and breakdowns
Airbrake
Filter chips that show one user's share on each row
Releases and regressions
PostHog
Reopened only the returning bug; Airbrake reopened all five
Performance
PostHog
Found the slow query by release; Airbrake times routes with no breakdown
Logs
PostHog
A logs product linked to traces; Airbrake keeps none
Alerts
Not compared
No channel connected in either tool
AI-assisted debugging
PostHog only
Airbrake has no assistant that I found
Price
PostHog
Free up to 100,000 exceptions monthly; Airbrake starts at $17
How I tested PostHog and Airbrake
Every tool I review runs against monitoring-lab, a bookshop that breaks on purpose. Its storefront is Vite with hash routing, its API is Express on SQLite, a stand-in price service fails some of the time, and a reporting job fires every minute. Tools plug in through a browser adapter and a server adapter, and the app's code stays the same.
The API's planted failures reached both tools:
A pair of unrelated TypeErrors: checkout looks up .discount on a coupon that isn't there, and the book page looks up .title on a book that isn't there.
A handled error: the app reports sold-out books itself.
A flaky neighbour: the price service answers 503.
Async failures: a recommendations promise rejects unhandled, and failing report runs include their run number in the message.
PostHog also heard from the storefront: the wishlist crash, failed API calls the page reports, and a failed tracking request.
Releases and coverage
Release
Scenario
PostHog
Airbrake
1.0.0
Install, first errors, maps, route timing
✔: storefront and API
✔: API
1.1.0
Coupon crash marked fixed, same code deployed again
✔: tested
✔: tested
1.2.0
Slow book query at checkout, every fifth charge rejected
✔: tested
✘: not run
1.1.0 is the release both tools faced: each was told the coupon crash was fixed, then received it again.
SDKs and coverage
Platform
PostHog
Airbrake
API
posthog-node 5.57.2 with OpenTelemetry alongside
@airbrake/node 2.1.9
Browser
posthog-js 1.441.0
Not tested
Plan
Free, US Cloud
30-day trial, card required
Headless shoppers:posthog-js skips headless Chromium unless told otherwise, and I told it otherwise.
Shared revision: both Airbrake deploys carried one revision, which left its fixed and introduced columns untested.
Server paths: my API's stack frames used file:// paths, which Airbrake can't download source for.
No repository connected, so PostHog's AI read only stack frames.
No alert channel in either tool.
Terms in each tool
Idea
PostHog
Airbrake
Grouped errors
Issue
Error
A single report
Exception event
Occurrence
Fixed
Resolved
Resolved
Release
Created by source map uploads, or $app_version
A deploy sent through CLI or cURL
Steps before an error
Timeline tab
Breadcrumbs, passed as a parameter here
Breakdowns
Breakdown panel
Aggregations tab
Request timing
Operations view
Performance routes
Setup and onboarding
PostHog asks you to switch things on. Airbrake asks you to pay up front.
PostHog: A goal, then a list of switches
PostHog opens by asking what you want to do. "Find and fix issues" pre-selected three products:
Error Tracking wasn't one of them, so I added it, with Logs and Metrics, under "Show all products". With no Express choice in the AI install wizard, I followed the manual Node.js instructions:
The page's note about setupExpressErrorHandler is the important line, since Express swallows errors before the SDK sees them. My installed version was too old for that helper, and upgrading to the latest posthog-node fixed it. Three more settings changed what I saw:
Headless traffic is ignored by posthog-js until you set opt_out_useragent_filter: true.
Replay was off in project settings, although onboarding showed it ticked.
Browser exception autocapture is a project setting, or capture_exceptions: true in code.
Airbrake: Payment details, then a notifier that also times routes
Airbrake's trial starts at a form with card fields:
After that, the API reported errors and route timings the moment it started. Two oddities: events landed in a "Monitoring" project while the default one stayed empty, and the Deploys page popped up a dialog with the project key in plain view.
The notifier:
src/monitoring/adapters/airbrake.js
Copied!
import { Notifier } from '@airbrake/node';
const airbrake = new Notifier({
projectId: Number(process.env.AIRBRAKE_PROJECT_ID),
projectKey: process.env.AIRBRAKE_PROJECT_KEY,
environment: 'development',
});
With performanceStats left at its default of on, Airbrake's Performance page fills itself.
Worth turning on: in PostHog, enable replay in project settings and check your posthog-node version supports setupExpressErrorHandler before you trust server capture. In Airbrake, report a deploy with a new revision from your pipeline on every release, because those deploys drive its fixed and introduced counts.
Verdict: PostHog, narrowly. You can start free, and every switch it needed was documented. Airbrake's notifier was the smoother install, once you've handed over a card.
Server error capture
Both tools watched the same API, so this is a direct comparison.
PostHog: Everything except the rejection
PostHog caught the coupon crash, the missing book, sold-out orders, the price service failure and the report job. Two things stood out:
The unhandled promise rejection never arrived.posthog-node appears to wait for Node to rethrow it, and the app's own rejection handler prevents that.
Every anonymous request is a new user. Without a signed-in shopper, each request got its own ID, so 24 price-service events showed 24 users.
Airbrake: Everything, with one thin entry
Airbrake received every planted failure, including the rejection, which arrived ten times with a single frame:
That's loadRecommendations at server.js:144 and nothing about the request. No badge marks it unhandled, and the sold-out error my code reported on purpose carries the same severity as a crash.
Server capture comparison
Capability
PostHog
Airbrake
Express crashes
✔: all
✔: all
Unhandled rejection
✘: missed
✔: one frame
Report job
✔: captured
✔: captured
Handled versus unhandled
✔: severity per issue
✘: same severity
Users per error
Inflated by anonymous requests
✔: signed-in users only
Verdict: Airbrake, narrowly. It caught the one failure PostHog missed. PostHog's issues carried more per row, but a missed rejection is the bigger gap on a Node API.
Debugging a server error
Same crash, the coupon TypeError, in both tools.
PostHog: The code and the trace
posthog-node reads source files from disk, so the error showed the failing line with the code around it:
Line 150 sits highlighted, and "Show 1 vendor frame" folds away Node's internals. View trace on the same error opened the request:
It held eight spans in 2.67 ms: Express middleware plus the app's own SELECT book for checkout. There was no request body in the properties.
Airbrake: A route error rate and a page of JSON
Airbrake's overview carries a number PostHog doesn't show:
Its ROUTE line puts the crash at 11.76% of 85 checkout requests, a direct measure of the damage. Everything else is raw JSON, where my own fields sit among the notifier's os and process data:
Breadcrumbs only showed up because my adapter packed them into a parameter. The backtrace stayed codeless; a sourceMapErrors entry said Airbrake can't download from file:// paths.
Server debugging comparison
Capability
PostHog
Airbrake
Code on the frame
✔: from the server's files
✘: file:// paths unsupported
Request trace
✔: View trace
✘: not offered
Request params
✘: not captured
✔: as JSON
Share of the route failing
✘: not shown
✔: on the overview
Readable breadcrumbs
✔: Timeline tab
Raw JSON parameter
Verdict: PostHog. The code and the trace answered the crash faster. Airbrake's route error rate is worth having, and its page makes you read JSON to get the rest.
Browser errors and session replay
Airbrake has a browser notifier, but I didn't test it, so this section covers PostHog alone.
Source maps and the browser stack
PostHog offers three paths: unminified code, public maps it fetches, or a private upload at build time. After posthog-cli sourcemap inject and upload, the wishlist crash read cleanly:
The frame points at ../../src/pages.js@121:13, with line 122 highlighted. One catch: because the fingerprint comes from resolved frames, the crash before and after mapping counted as two issues. Expect that once per bug when maps first go in.
Replay on the error
Replay was off at the project level until I switched it on. After that, each browser error had a Recording tab:
The pointer rests on "Add to wishlist", and the shopper's name is on screen. PostHog's Normal privacy level hides inputs and leaves page text visible. A Timeline tab beside it lists the session's pageviews, exceptions and logs.
Capability
PostHog
Airbrake
Browser errors
✔: after turning off the headless filter
Not tested
Private source map upload
✔: CLI at build time
Not tested
Session replay
✔: off by default
✘: not offered
Text masking by default
Inputs only
Not applicable
Verdict: PostHog only. The browser side, with maps and a recording linked from the error, is a reason to pick PostHog on its own.
Error grouping
The two TypeErrors in my API look almost the same as messages: "Cannot read properties of undefined", reading discount in one and title in the other.
Airbrake: One pattern, two bugs
Airbrake turns the variable parts of a message into placeholders, and both crashes collapsed into "Cannot read properties of undefined (reading {})":
Its message-param breakdown shows title and discount side by side in that group. Six server bugs made five rows, resolving one crash resolved the other, and I found no way to split them. The same rule worked well elsewhere: every failed report run shared one row.
PostHog: Separate at first, split later
PostHog fingerprints the stack, not the message, which kept the two TypeErrors as separate issues. The weakness appeared in 1.2.0, where the payment bug became two issues:
Both carry the same two lines from my code. One also holds Node timer frames that are hidden behind "Show 2 vendor frames" but still count toward the fingerprint, and the slow query changed which frames appeared. The coupon and sold-out errors gained sibling issues the same way. Grouping rules and a manual merge exist to fix it.
Grouping comparison
Behavior
PostHog
Airbrake
Groups by
Stack frames
Message with placeholders
The two TypeErrors
✔: kept apart
✘: merged
Values in messages
✔: ignored
✔: replaced with placeholders
Timing changes in Node
✘: new issues in 1.2.0
Not tested
Manual fixes
✔: grouping rules, merge
✘: none found
Verdict: PostHog. It kept the two crashes apart, and you can correct its splits. Airbrake hid one bug inside another and gave me no tool to separate them.
Airbrake merged two crashes on message shape, and PostHog split one bug on Node's internals. Better Stack kept the 1.2.0 payment failures in a single group in my test:
The ten exceptions share one group with the throwing line underneath, and other bugs kept their own rows. One new bug added one new row.
Both tools answer "who is affected, on which release". Airbrake answers it on the list itself.
Airbrake: Chips with counts
Clicking into search lists 27 keys you can filter on, User, Version, Route and Message param among them:
Choose one user and each row reports that user's share of its occurrences:
carol owns 5 of the 16 sold-out errors and 4 of the 23 crashes. Each error's Aggregations tab slices occurrences by user, version, route and message param. Typed text didn't match; you work through chips.
PostHog: A Breakdown panel per issue
PostHog's issues list filters by property, and each issue has a Breakdown panel that splits occurrences by release or any custom property. There's no per-user share across the list, and anonymous requests inflate user counts.
Capability
PostHog
Airbrake
Filter the list by user
✔: property filter
✔: chip with N of M counts
Breakdown inside one error
✔: any property
✔: over 20 aggregations
Route filter
✔: property filter
✔: Route chip
Free-text search
✔: yes
✘: chips only
Verdict: Airbrake. Its N of M counts on every row answered "how much of this is carol" without opening anything.
Releases and regressions
Both tools were told the coupon crash was resolved. 1.1.0 then went out carrying it.
PostHog: The one bug came back
PostHog offers three statuses, Active, Resolved and Suppressed, and none takes a version. Marking it resolved popped up an offer to email the affected signed-in shoppers that the bug was fixed. I didn't send it. On 1.1.0, the issue reopened on its own:
It returned to Active, going from 44 to 64 occurrences, with no "regressed" label to set it apart in the list. The Breakdown panel showed which release brought it back:
Of 64 occurrences, 44 came from the first release and 20 from the second. Release records come from source map uploads and default to the git commit name; passing --release-version at the inject step, not only at upload, fixed the naming.
Airbrake: Every error came back
Airbrake resolves with one click and an Undo toast, with no version. A project setting that resolves every error in the environment at each deploy ships switched on:
Reporting 1.1.0 cleared all five groups at once, and every one reopened as shoppers came back. The crash group signalled its return with a number alone:
Of 46 occurrences, 23 arrived after the deploy, and no label or activity line called it a regression. Since that group also held the missing-book crash, the coupon bug's return couldn't be told apart from the other one.
Release comparison
Capability
PostHog
Airbrake
Resolve tied to a version
✘: no
✘: no
Returning bug reopened
✔: just that issue
✔: along with every other group
Regression label
✘: shown as Active
✘: count since the last deploy
Errors by release
✔: Breakdown panel
✔: Version aggregation
Release records
From source map uploads
✔: Deploys list
Email affected users
✔: offered on resolve
✘: not offered
Verdict: PostHog. It reopened the bug that came back and nothing else. Airbrake's default deploy setting made the whole list look new, which buries the one regression you care about.
The coupon bug looked like any open issue in PostHog, and Airbrake had reopened the lot. Better Stack marked just that bug when 1.1.0 shipped:
A red "Reoccurred" sits on the coupon row, and no other row changed state. The regression stands out from the list at a glance.
Both tools time server requests. Only one ran the slow release.
Airbrake: Routes, percentiles and Apdex for free
Airbrake's Performance page filled itself with no setup:
After one run it reported 270 requests, Apdex 0.92, an average of 239 ms and roughly a tenth of requests failing. Ordering routes by p95 put my two slowed routes on top:
Search sat at 1.90 s and the price lookup at 1.70 s, both marked orange. Below the route, the breakdown bar was blank and the Queries tab empty, so Airbrake can tell you a route is slow, not what's slow inside it. The 1.2.0 checkout slowdown wasn't run through it.
PostHog: The slow query, filtered by release
PostHog's Operations view ranks operations by percentile and error rate. With the version facet on 1.1.0, checkout's p95 read 4.21 ms:
Across all releases, the same table told a different story:
Checkout's p95 climbed to 4.76 s, and the book lookup span showed 2.93 s at p95 with a median of 0.15 ms. Ordering by p95 and switching the version facet found the query and the release that slowed it. GET /search was a false lead at 1.23 s, slow on every release in code the spans didn't cover.
Performance comparison
Capability
PostHog
Airbrake
Route latency
✔: p50, p95, p99 per operation
✔: p50, p95, Apdex
Setup needed
OpenTelemetry
✔: none
Breakdown inside a request
✔: spans
✘: empty in my test
Latency by release
✔: version facet
Not tested
The 1.2.0 slowdown
✔: query found
Not run
Verdict: PostHog. It pinned the slowdown on one query in one release. Airbrake's route timing is a generous extra that comes free with the notifier, and it stops at the route.
Airbrake ranked slow routes without saying why, and PostHog found the query from percentiles. One 1.2.0 checkout in my Better Stack test carried both problems:
Of that request's 2.19 s, the lookup took 1.53 s and the rejected charge 660 ms, side by side.
PostHog stores logs. Airbrake keeps nothing beyond what you attach to an error.
PostHog: A logs product tied to traces
PostHog took my app's logs over OpenTelemetry into its own viewer:
The hour held 1,068 logs, 968 Info and 100 Error, with Patterns, Group, Live tail and a SQL tab. From any trace, the span panel's Logs tab lists that request's lines.
Airbrake: Parameters only
Airbrake sells error and performance monitoring. Logs exist there only as parameters you send with an error, as my breadcrumbs did.
Capability
PostHog
Airbrake
Stored and searchable logs
✔: with a SQL tab
✘: not offered
A request's own logs
✔: one click from its trace
✘: not offered
Streaming view
✔: Live tail, untried
✘: not offered
Verdict: PostHog. It keeps logs and links them to the request. With Airbrake, you'd add a log tool.
PostHog links logs from a trace, and Airbrake keeps none. In my Better Stack test, the failing 1.2.0 checkout's trace was the log filter:
Only that request's lines appear, with the slow-query warning right before the failure, and they remain searchable afterwards.
PostHog can notify you about issues that are new, reopened or spiking, through Slack, Teams, PagerDuty and others, and it nags with a banner until a destination is set. Spike detection and severity rules sit in Configuration.
Airbrake's alerts are mostly per-user email settings:
Its Jira integration reopens a linked ticket when a resolved error happens again, a box that comes ticked. Uptime checks aren't part of either product.
Capability
PostHog
Airbrake
New errors
✔: new issues
✔: email
Returning errors
✔: reopened issues
Jira reopen, not tested
Spikes
✔: spiking issues
Not tested
Uptime checks
✘: not offered
✘: not offered
Where alerts go
Slack, Teams, PagerDuty and more
Email, plus integrations
Verdict: Not compared. I received no alert from either. On paper, PostHog covers reopened bugs in-product, and Airbrake leans on Jira for that.
AI-assisted debugging
PostHog ships an assistant. Airbrake, as far as its product and pricing pages show, has none.
PostHog AI: Right mechanism, invented details
Every PostHog issue has "Open Fix prompt", which hands the error to PostHog AI or to an outside coding agent. On the coupon crash it found the mechanism, a coupon that's checked for presence but never looked up. Then it guessed:
The coupon codes it lists don't exist in the app. Asked about the 1.2.0 payment failure, it pinned the split on the charge amounts, pointed me at the gateway's status page, and drafted a "plausible reconstruction" of unseen code. The release and the slow query never came up. With no repository connected, it read stack frames only.
Capability
PostHog
Airbrake
Assistant in the product
✔: PostHog AI
✘: none found
Hand-off to coding agents
✔: Fix prompt
✘: none found
Stayed within its evidence
✘: invented coupons and code
Not applicable
Verdict: PostHog only. It's the only one with AI, and without your code it filled gaps with fiction. Connect a repository or check its claims.
PostHog AI invented code it couldn't read, and Airbrake has no assistant. Better Stack's AI SRE took on release 1.2.0 with no repository either:
Its answer named the checkout query as the cause of the slowdown and put the failures at 10 of 48 charges, every one from 1.2.0. That run cost $0.49 and about four minutes.
PostHog bills each product by usage past a monthly free tier, with no base fee. Airbrake sells a plan with an error allowance and meters performance events on top.
PostHog's pricing page was read on November 4, 2026, Airbrake's on November 16. Confirm both before you sign up.
Free plans and trials
Starting out
PostHog
Airbrake
Free plan
✔: free tier, no card
✘: none
Trial
Not applicable
30 days, card required
Errors
100,000 exceptions a month
25,000 in my trial
Recordings
5,000 monthly
✘: not offered
Logs
10 GB monthly
✘: not offered
AI credits
500 monthly
✘: none found
PostHog's plan step showed one free tier on both plans:
On Free, events past the allowance are dropped. Pay-as-you-go has no base fee, allows six projects and keeps recordings for 90 days in place of one month.
Entry-level paid plans
Cheapest tiers
PostHog Pay-as-you-go
Airbrake Dev
Airbrake Basic
Monthly
$0 base
From $19
From $38
Yearly
$0 base
From $17 a month
From $34 a month
Errors included
100,000 a month free
25,000
25,000
Users
Unlimited
1
Unlimited, 3 teams
Retention
Recordings 90 days
30 days
30 days
Performance
Tracing billed separately
Metered separately
Metered separately
Past the free tier, PostHog charged $0.000370 per exception from 100,001 to 325,000 in my October reading. On Airbrake, 100,000 errors costs $59 monthly, and errors past your tier start at $0.00065 apiece.
Comparing the actual cost
For a small team, PostHog costs nothing until you pass 100,000 exceptions a month, and that free tier also includes replay, logs and AI credits. Airbrake starts at $17 to $19 monthly for a single developer and $34 to $38 for a team, route timing included, card required.
Both bills move with errors. PostHog charges per exception past the free tier, and you can cap each product. Airbrake counts every error against its tier and bills on demand past it. Our PostHog vs TrackJS comparison weighs PostHog's metered bill against one priced by page views.
Verdict: PostHog. It's free at small volume and covers more. Airbrake is cheap for a paid tool, but it's never free.
Full feature comparison
Feature
PostHog
Airbrake
Server errors
✔: source lines and trace
✔: file and line, JSON context
Unhandled rejections
✘: missed in my setup
✔: captured
Browser errors
✔: after turning off the headless filter
Notifier exists, not tested
Source maps
✔: private upload
Not tested
Session replay
✔: off by default
✘: not offered
Share of the route failing
✘: not shown
✔: on each error
Grouping
Stack frames, Node internals included
Message with placeholders
Split or merge issues
✔: rules and merge
✘: none found
Search with per-user counts
✘: not offered
✔: 27 keys
Returning bug
✔: reopens just that issue
Reopens every group after a deploy
Notify affected users
✔: "Tell them it's fixed"
✘: not offered
Route performance
✔: via OpenTelemetry
✔: built in
Query-level timing
✔: spans
✘: empty in my test
Logs
✔: own product
✘: not offered
AI
✔: PostHog AI
✘: none found
Free plan
✔: 100,000 exceptions a month
✘: trial with a card
Which should you choose?
These two come from opposite directions. PostHog puts errors inside a product platform. Airbrake keeps errors and route timing in one small package.
Choose PostHog if you want the error and everything around it
You debug from the user's side. Browser errors come with source-mapped code and the session recording.
You need to know why a release got slower. Its Operations view found the slow query by release.
You want one bill that starts at zero. The free tier covers errors, replay, logs and AI credits.
The trade-offs: a long list of switches at setup, a missed unhandled rejection, Node timing changes that split issues, no regression label, and an AI that invents what it can't see. Our Sentry vs PostHog comparison weighs it against a dedicated error platform.
Choose Airbrake if your API is the product and you want it cheap
You want each error's cost in traffic. The route error rate shows the share of requests a bug breaks.
You triage by user, release and route. Its filter chips and per-user counts beat PostHog's search here.
You want route timing with no extra setup. Percentiles and Apdex come with the error notifier.
The trade-offs: payment details up front, error pages built from JSON, two separate crashes folded into one row for good, deploys that reset every error by default, and no logs or AI. Our TrackJS vs Airbrake comparison puts it against a browser-first tool.
FAQ
Does PostHog have a separate error tracking product?
Yes. Error Tracking is one PostHog product, sharing a project with analytics, replay, logs and tracing. It isn't switched on by the "Find and fix issues" onboarding, so add it yourself.
Does Airbrake monitor performance?
For server routes, yes. The same notifier times each request out of the box and reports p50, p95, Apdex and error rate by route. It showed no spans or query detail in my test.
Why did Airbrake reopen every error after a release?
A project setting, on from the start, clears every error at each deploy. Whatever fires again afterwards reopens, whether it's a regression or not.
Which is cheaper?
PostHog for most small teams, since it's free up to 100,000 exceptions a month. Airbrake has no free plan; Dev starts at $17 a month on a yearly plan.
Final thoughts
Airbrake was the more precise tool where it chose to look. It caught the rejection PostHog missed, told me what share of checkout traffic the coupon bug broke, and let me count one user's share of every error. PostHog was the more useful tool around the error: the code, the trace, the recording, the slow query, and the one bug that actually came back.
So the choice is between a sharp, narrow error tracker and a broad platform that needs tuning. Airbrake's grouping and deploy defaults will cost you the moments when a release goes wrong, and PostHog's setup will cost you an afternoon.
If you only fix what your error tracker tells you is new, Airbrake will tell you everything is.