Frequently asked questions
Including the ones with answers we would rather not have to give.
Compatibility
How do I send events to Parsemend if I have never used Sentry?
Install the open-source Sentry SDK for your language from your normal package manager (composer require sentry/sentry-laravel, npm install @sentry/node, and so on) and set its DSN to the one your Parsemend project issues. That is the whole integration. The SDK is a client library for the ingest protocol Parsemend implements; the DSN decides which server receives the events, and yours points at Parsemend.
Do I need a Sentry account or subscription?
No. The Sentry SDKs are MIT-licensed open-source packages that work against any server implementing the protocol. Installing one creates no account, needs no key from Sentry, and sends nothing to sentry.io. The DSN you configure is the only destination the SDK uses.
I already send events to Sentry. What changes?
Only the DSN value. Parsemend implements Sentry's envelope endpoint, the legacy store endpoint, and all three DSN authentication forms, so the SDK stays installed and its initialisation code stays where it is. Change the DSN, redeploy. We have verified this end to end with sentry/sentry 4.29.0 for PHP and @sentry/node 10.64.0.
Which Sentry SDKs have you actually tested?
The official PHP SDK and the official Node SDK. Other Sentry SDKs speak the same wire protocol and should work, but we only claim what we have run. If you test another one, tell us and we will list it.
What happens to transactions, logs and cron check-ins?
They are stored. Transactions become traces with their spans, logs are kept and searchable, and check-ins drive cron monitors that raise an issue when a job is missed, times out or fails. What Parsemend still drops, with a reason code, are session, sessions and client_report items, so an SDK that sends those never starts erroring.
Which request compressions are supported?
gzip, deflate and identity. Brotli and zstd request bodies are rejected with a recorded invalid outcome rather than a 500.
The agent
Will the agent merge code into my repository?
No. The agent opens a draft pull request with the root cause in the description. A human reviews, runs the tests, and merges. There is no setting that lets it merge, and there will not be one.
Does the agent run my tests against its patch?
Not today. Sandboxed fix validation, meaning your suite runs against the patch in a container, is the next item on the roadmap. Until it ships the agent reasons about the code rather than verifying it, which is why the pull request arrives as a draft.
What stops it from opening nonsense pull requests?
Two gates. A fixability score decides whether a run starts at all, and a run only begins above the project's threshold. Then every step emits a confidence score; below the threshold the run pauses for a human instead of proceeding from a bad root cause.
Can I edit the plan before it writes code?
Yes. The change plan is a separate step from the patch precisely so you can correct or reject it before a line of code exists. Correcting a plan is cheaper than reviewing a wrong diff.
Does Parsemend review pull requests?
Yes, on repositories that opt in. Three agents run per review: one explores the diff, one writes the findings, and one fact-checks each finding before anything is posted. Unverified findings are dropped or marked uncertain. Open production issues on the touched files are injected as context. Reviews also diff your lockfiles against OSV and Packagist advisories and scan added lines for secrets with local rules. It comments; it never approves, requests changes or merges.
Does Parsemend run browser tests?
Yes. QA flows are user journeys written as plain-language steps. A model resolves each step to concrete selectors once, then every replay is a deterministic Chromium run with zero model calls, and the cached selectors heal when your markup drifts. Assertions are never healed: a failing assertion fails the run and files an issue through the same pipeline as an error. On a pull request, the parsemend/qa check reports pass or fail and only gates the merge on repositories that mark it a required check; a triage pass annotates failures but never gates by itself. Running qa:verify-fix re-runs the failing flow against a fix's pull request to confirm the patch actually works.
Privacy and control
Does Parsemend see my source code?
Only in managed mode, where it reads the repository through a GitHub App installation. In bring-your-own-agent and CLI modes your source never leaves your infrastructure. In your-keys mode Parsemend assembles the context bundle and transits it to your model provider without retaining it. Parsemend stores events, not code.
Can I use my own model?
Yes, in three of the four modes: Anthropic, OpenAI, Gemini, or any model you serve yourself through Ollama. Managed mode is the exception, because there Parsemend is paying for the tokens.
Is Parsemend self-hosted, or can I run the fix agent on my own hardware?
Parsemend is a hosted service; there is no self-hosted distribution. What you can run on your own hardware is the fix agent: in CLI mode the fix run executes on your machine against a local Ollama endpoint, so your source code and inference never leave your infrastructure. Nothing about that path requires a network call to us.
Pricing
Is error tracking really free forever?
Yes. 5,000 events a month on one project, with no time limit and no credit card. It is not a trial. What it does not include is the fix agent.
Why is every AI feature paid?
Because running the agent costs real money, and a free version of it would either be rate-limited into uselessness or have to fund itself some other way. The free tier is the full tracker; the agent is what the paid plans buy.
How do AI credits work?
Every paid plan includes a monthly allowance of AI credits, and the agent's work draws from that allowance. When a month outgrows it you buy a top-up; when it does not, there is nothing extra to pay. If you bring your own API keys or serve a model yourself through Ollama, your provider bills you directly and your credits are never touched.
The company
Who builds Parsemend?
One engineer. The ingest pipeline, the fix-run state machine and the agent framework described on this site are all his work.
What does Parsemend not do?
No session replay, no profiling, no native symbolication, no release health or crash-free rates (so no crash-free-rate alerts), and no ad-hoc analytics dashboards or arbitrary-tag query layer. It does have tracing, logs, uptime and cron monitoring, source maps, issue alert rules plus metric and anomaly alerts over error rate, failure rate and p95 (rolling statistics, not ML), pull request review with dependency and secret scanning, and autonomous QA flows. Where Sentry is still ahead, the comparison page says so.