Plausible does less, deliberately
It has shipped since 2018, it is paid for by the people who use it, and on the human half of analytics it is the more finished product. This page is about the half it does not measure, and about when you should stay exactly where you are.
Where Plausible is ahead
Not throat-clearing. These are the reasons a sensible person picks them over this project today.
- It is finished, and this is not
- Micaforge has no tagged release, no production install and no benchmark. Plausible has years of other people's edge cases already found and fixed. That is worth more than any row in the table below.
- Somebody is accountable for it
- A company, an inbox, a managed service on EU infrastructure, and support from the people who wrote the code. Micaforge offers an issue tracker.
- The dashboard is genuinely simpler
- One page, no menus, no training. Simplicity is their product decision rather than a gap in it. Micaforge is a denser instrument, and density costs something.
- Search Console
- Keyword data in the dashboard. Micaforge has not built it, and there is no route for it in the repository.
- Their bot filtering is better than ours at being a filter
- The cloud build excludes roughly 32,000 data-centre ranges on top of user-agent and referrer-spam rules. If all you want is a clean human number, that is the shorter road.
What is not a difference
Two claims a comparison page would love to make, which happen to be false.
Plausible already treats AI assistants as a channel. Referrals from AI assistants are reported as an AI Assistants channel, next to Organic Search rather than buried in Referral. Micaforge does the same thing under a different name.
Plausible can already recognise an AI crawler. GPTBot and its peers are known to it. What happens next is the difference: a recognised crawler's hit is dropped, so nothing about it survives to be reported.
What is a difference
One thing, from which everything else on this page follows. Both products hold the returned half of the exchange, the reader who arrived from an answer. Only Micaforge holds the taken half, the crawl that produced the answer, in the same place and against the same URL.
That half cannot be collected by a script tag at all. Crawlers do not run JavaScript, so they never reach an endpoint a browser calls; the only record of them is your web server's log. Micaforge reads that log into a table of its own, with an operator, a purpose and a verification result on every row, and then reports the two counts as one ratio per page.
If nothing is fetching your pages to answer questions about them, all of that is weight you do not need. If something is, no amount of filtering will tell you what it took, and no configuration of Plausible will either, because half the arithmetic is not there.
Row by row
| Question | MicaforgeAGPL-3.0, pre-release | Plausible CESelf-hosted community edition | Plausible CloudTheir managed service |
|---|---|---|---|
| Licence | AGPL-3.0Tracker and SDK under MIT, so embedding them does not reach your code. | AGPL-3.0Tracker under MIT, for the same reason. | Same code, hosted |
| Self-hosting | The whole buildOne command, five containers, no feature gate. | Community Edition | Not applicableThe managed service is the product. |
| Features held back from the self-hosted build | NoneWhat a hosted build would run is what docker compose up runs. | FourTheir README lists marketing funnels, ecommerce revenue goals, SSO and the sites API as unavailable in CE. | NoneEverything in their pricing plans. |
| Release cadence | Pre-releaseNo tagged release, no production install, no track record. That is the honest state today. | Twice a yearTheir README calls CE a long term release, so new work reaches it late. | ContinuouslyTheir README says multiple times per week. |
| Cookieless, no personal data | YesDaily-rotating salted hash. No raw IP on a human event row. | Yes | YesGDPR, CCPA and PECR compliance is their headline claim. |
| Answer-engine referrals grouped as their own channel | YesChannel answer_engine, beside search, social and direct. | YesAI assistant referrals are reported as their own AI Assistants channel. This is not a difference between the two products. | YesSame as the cloud service. |
| Where non-human traffic is counted from | The server access logA shipper reads the log your web server already writes. Crawlers never run JavaScript, so this is the only place they can be counted. | The tracking scriptA client that does not execute JavaScript never reaches the endpoint, so it can be neither counted nor filtered there. Their events API accepts server-side hits, but those land as pageviews in the same table. | The tracking script |
| What happens to a crawler once recognised | Kept and namedIts own table, with operator, purpose, robots verdict and a verification result. Excluded from the human report by default, never deleted. | ExcludedRecognised crawlers, GPTBot among them, are filtered out of the stats. The classification is used and then discarded. | ExcludedTheir README: non-human traffic patterns plus roughly 32,000 data-centre IP ranges. |
| Agents grouped by operator and by purpose | YesTraining, retrieval, search index, preview, evaluation. Blocking a corpus crawler and blocking the fetch behind a person's question are different decisions. | No | No |
| Agent verified against published address ranges | Yes, or honestly unverifiedA user agent is an assertion. Where an operator publishes ranges or reverse-DNS suffixes, the row says verified, unverified or failed. | No | No |
| Citation Gap per URL | YesFetches against answer arrivals, per page, with a published threshold behind every verdict. | NoThey hold the returned half. The taken half is not stored, so the ratio cannot be formed. | No |
| robots.txt and llms.txt against observed behaviour | YesYour own policy read back against what actually fetched, and what fetched anyway. | No | No |
| Session replay | YesThirty-day retention by default, on a clock of its own. | Not part of the productPlausible is deliberately one simple page. | Not part of the product |
| Funnels | Yes | Not in CEListed among the premium features in their README. | Yes |
| Google Search Console integration | Not builtThere is no route for it in this repository. Claiming it would be the first lie on the page. | Yes | Yes |
| Managed hosting | NoneThere is no hosted Micaforge. You run it, or you do not use it. | Not applicable | EU onlyTheir README: EU-owned infrastructure, servers in Germany. |
| Support | Issues, and nothing elseNo company, no inbox, no promise. | Community forumTheir README says CE is community supported only. | The people who build it |
| Next | Install Micaforge | Plausible CE | plausible.io |
Moving between them
Micaforge imports Plausible's CSV exports. Both count a session on a thirty-minute inactivity gap and neither has ever stored a personal identifier, so the human numbers should land close together. Two things will move: crawler traffic becomes a second set of numbers beside your own instead of nothing, and the AI Assistants channel you already have gains a second column showing what those assistants took to build the answers.
Running both for a fortnight costs a few kilobytes on the page and settles the question better than this table can.
Plausible's rows restate their README and their public documentation, and describe how their product behaves, as of August 2026. Micaforge's rows describe the code in this repository. Nothing here is a benchmark, because this project has benchmarked neither product, and nothing here is a customer's opinion, because Micaforge has no customers.
Stay with Plausible if
- You want one simple page anyone on the team can read.
- You want somebody to call when it breaks.
- You would rather never think about crawlers again.
- Search Console keywords in the dashboard matter to you.
- You need software that has already proven itself in production.
Try Micaforge if
- Your traffic is documentation, a changelog, a blog, an API reference.
- You want to know what the crawl took, not only what came back.
- You want both halves on one time axis and one ratio per page.
- You can get at your web server's access log.
- You are comfortable running pre-release software you can read.