Track errors free. Pay when they start fixing themselves.
The tracker is free forever, because a tracker is table stakes. The agent is the reason to pay, so the agent is never free.
todo— the paid tier prices below are placeholders pending sign-off. They are deliberately excluded from this page's structured data until they are final.
Free
Free error tracking. Forever.
- events
- 5,000 events / month
- projects
- 1 project
- retention
- 30-day retention
- Sentry-compatible ingest
- Grouping, tags, breadcrumbs
- Regression detection
- Rate limits and quotas
This button exists on your issues. Upgrading unlocks it. Free accounts never consume Parsemend's model tokens, which is why it is not free.
Developer
agentThe agent, on your compute.
- events
- 100,000 events / month
- projects
- 5 projects
- retention
- 90-day retention
- Everything in Free
- Bring your own agent, keys, or CLI
- Context bundles and confidence gating
- Draft PR post-back
Team
agentManaged runs, or your own.
- events
- 1,000,000 events / month
- projects
- Unlimited projects
- retention
- 90-day retention
- Everything in Developer
- Managed mode — our infrastructure and tokens
- GitHub App integration
- Editable change plans
A closer look at each plan
Every capability, one row each, with the tier it lands in. The last group is what Parsemend does not do — stated plainly, with a path to the roadmap.
| Capability | Free | Developer | Team |
|---|---|---|---|
| Error tracking | |||
| Events per month | 5,000 | 100,000 | 1,000,000 |
| Projects | 1 | 5 | Unlimited |
| Retention | 30 days | 90 days | 90 days |
| Sentry-compatible ingest (envelope + store) | Included | Included | Included |
| Works with unmodified Sentry SDKs (verified: PHP, Node) | Included | Included | Included |
| Grouping cascade + custom fingerprints | Included | Included | Included |
| Tags, breadcrumbs, environments | Included | Included | Included |
| Regression detection | Included | Included | Included |
| Rate limits and quotas (429 + Retry-After) | Included | Included | Included |
| The fix agent | |||
| Fix runs (RootCause → Solution → Coding → ChangeDescription) | Not included | Included | Included |
| Fixability score and confidence gating | Not included | Included | Included |
| Context bundles | Not included | Included | Included |
| Draft PR post-back | Not included | Included | Included |
| Bring-your-own-agent, your-keys and CLI modes | Not included | Included | Included |
| Model providers: Anthropic, OpenAI, Gemini, Ollama | Not included | Included | Included |
| Managed mode (our infrastructure and tokens) | Not included | Not included | Included |
| GitHub App integration | Not included | Not included | Included |
| Editable change plans | Not included | Not included | Included |
| Benchmark reports | Not included | Not included | Included |
| Not in ParsemendThese are on the roadmap, not in the product. The comparison with Sentry is at /vs/sentry; where each one stands is at /roadmap. | |||
| Uptime monitoring | Not included | Not included | Not included |
| Cron monitoring1 | Not included | Not included | Not included |
| Metric monitors and alerts | Not included | Not included | Not included |
| Tracing and performance1 | Not included | Not included | Not included |
| Session replay | Not included | Not included | Not included |
| Logs1 | Not included | Not included | Not included |
| Profiling | Not included | Not included | Not included |
1transaction, log and check_in envelope items are accepted and dropped with a reason code, so an SDK that sends them never errors. That is tolerance, not support.
How billing works
- Parsemend, flat.
- A fixed monthly framework fee: context bundles, the fix-run state machine, confidence gating, draft PR post-back. It does not move when your error volume spikes or when you switch models.
- Your provider, metered.
- Inference is billed to your Anthropic, OpenAI or Google account at their public rates, or costs nothing if you serve the model yourself with Ollama. Parsemend never resells it.
- Managed mode, one bill.
- The exception: there the tokens are ours, so there is a single bill. You trade model choice and repository access for the simplicity.
Cost per fix, worked out
A calculator that turns your issue volume into a token bill is only useful if its numbers come from real runs. It ships with the benchmark suite, together with the per-run token counts it is built on. Until then there is nothing here worth reading.
Paying your model provider directly means a second account, an API key to hold, and a second invoice to watch. We think owning your inference is worth that friction. If you do not, managed mode exists and there is no judgement in it.
Frequently asked questions
Is the free tier a trial?
No. Error tracking is free forever, with 5,000 events a month on one project. It never expires. What it does not include is the fix agent.
Why is every AI feature paid?
Because the agent costs real money to run, and a free tier that consumed our model tokens would either be rate-limited into uselessness or fund itself by selling your data. Free signups never consume Parsemend's tokens. The differentiator is never free.
Who bills me for model usage?
Your model provider does, directly, at their published rates. Parsemend charges a flat monthly framework fee — context bundles, the fix-run state machine, confidence gating, draft PR post-back — and never resells inference, so it has no incentive to route you to a bigger model than the job needs. Managed mode is the exception: there the tokens are ours and there is one bill.
What does a fix run actually cost in tokens?
It depends on the size of the context bundle and the model you choose, so any single number here would be misleading. The cost calculator ships alongside the benchmark suite, where the token counts come from real runs rather than estimates.
Do I need to give Parsemend access to my repository?
Only in managed mode. Bring-your-own-agent, your-keys and CLI modes all work without a GitHub App installation, and in the first and third your source never leaves your infrastructure.
Still deciding where the agent should run? Compare the four fix modes.