State of the Radicle network
Radicle offers git-native code collaboration with no central server and no accounts. Nodes replicate whichever repositories they choose to seed. This page tracks the resulting network. The numbers are measured nightly from radicle.at, one of the oldest and largest public seed nodes. No trackers, no cookies, no third parties.
- Most repos are under a year old. Of the 16,762 repositories radicle.at has kept track of, 11,393 (68%) were created in the twelve months through October 2026, 3.9 times the 2,909 created the year before. July 2026 alone accounts for 5,067.
- The whole public network fits on a laptop. Mirroring every public repository on Radicle takes ~157 GiB. The largest 1% of repositories, among them full-history mirrors of linux, HardenedBSD, nixpkgs, godot and guix, hold 50% of the bytes, and 73% of repositories are under 1 MiB.
- Almost every public repo is seeded by several nodes at once. Half have 6 or more seeds, the average is 6.0, and only 199 of 17,467 have a single seed.
- Our seed mirrors most of the network. It stores 15,439 repositories, about 88% of the 17,467 public repos it can see. Everything below is measured from there.
01Replication & network reach
A repo can live on many nodes at once, and the more nodes seed it, the harder it is to lose. Here is how that looks across the 17,467 public repositories our seed can see and the 1,897 nodes seeding them.
Every repository our seed knows of has its own page, and most are listed at /repos.
Our policy is to mirror the whole network. The ~12% of public repos it does not store are a mix of stale, abandoned and test repos that radicle-seed-prune removed and repos not yet fetched or unreachable. Most of the 1,897 nodes are small personal seeds; 28 seed 1,000 repos or more, radicle.at among them. For 93 repositories, radicle.at is the only seed.
How seed counts have moved
The first chart tracks seed counts back to July 2026.
The median seed count of the 16,257 repos in the bars went from 7 nodes in the week to 19 October to 6 this past week, a move this period cannot tell apart from its noise. 1% of repos ended higher and 19% lower. A seed count is how many nodes announced the repo to radicle.at in the past 14 days.
The wider network our seed sees
Their pages: erikli, cloudhead, radicle-mirror, fosskar z6Mkqumz, fosskar z6Mkuikg, maninak, shirahara z6Mkg4WM, kristoff, efstajas z6MkeVA8, josh, shirahara z6MkvVZk, dirk1980-nb, lorenz, julianp, efstajas z6MkqG3P.
Their pages: Mistera-alpha, community.computer, smdev.
Our seed can see 3,630 nodes, but only 1,897 of them seeded any repository in the last 14 days, and just 28 seed 1,000 or more. The largest group, 1,623 nodes, has no known address; 1,323 give a domain name, 420 an IP address, 230 only a Tor address and 34 a Tailscale one. Of the 327 nodes that can be placed on a map, 116 are in Germany and 82 in the US.
Collaboration: issues & patches
Counts are per Radicle identity, one key per device, so one person can appear as several.
In the last year people opened or worked on 8,544 issues and 5,940 patches (about 1.4 issues per patch) in only 796 of the repositories radicle.at stores (5.2%). The ten most active hold 41% of that activity, led by AnyPS5 with 1,714. Most of what radicle.at stores saw no issue or patch all year. Over all time, 83.1% of the issues and patches on the repos radicle.at stores involve only automated identities, and 99% of those sit in 3 repositories.
Automation arrived in November 2023 and did not crowd people out: their issues and patches went from about 150 a month then to about 1,500 a month over the last six months. Read Share for automation's part of each month, and Volume for people's own line.
Collaboration has a committed core and a long tail of drop-ins: only 21% of the 1,224 identities that have ever opened or commented on an issue or patch here (1,259 counting automation) were active in the last three months.
Across the public history radicle.at has replicated, 48,874 patches have merged and 64,788 are still open; issues stand at 12,588 closed against 4,665 open. Automation opened 41,178 of the merged patches. One repository accounts for 58,050 of the open patches. The record starts in December 2022, well before this site began measuring anything.
Beyond issues and patches
Anyone can define a new kind of collaborative object (COB). These are the COB types radicle.at has seen, dated by when each COB was created.
02The repositories we seed
Our seed mirrors public repositories from across the Radicle network so they stay available even when their authors are offline.
Half of all repositories are under 0.25 MiB, little more than metadata and a few commits, while the largest 1% hold 50% of the bytes.
03Traffic & visitors
The seed answers two kinds of web traffic: the explorer at app.radicle.at (the pages people open) and the API at seed.radicle.at (what the explorer and other tools call). Over the last year, automated clients made 99% of explorer requests and 97% of API requests.
Strip out the crawlers and scanners, and people made 242 explorer requests in a typical month over the last year. November 2026 so far counts 1,261, 5.2 times that. Automated clients that send an ordinary browser User-Agent are counted as people, so a jump this size can be automation that was not caught.
04What gets browsed
Requests to the explorer and the API since April 2024 named 16,019 distinct repositories, crawlers included.
Over the last year the top organic repository, webspace_app, drew more requests than any pinned one: 49,359, against 30,291 for radicle-jetbrains-plugin.
05Who visits
Human visits only. Crawlers are told apart by each request's User-Agent and network origin.
06Speed & reliability
Only the API is timed; the explorer is a static app. Blocked requests are not timed. On a typical day in the last month the API answered half its requests within 7.9 ms and 99% within 1.0 s.
Connecting to this always-on seed helps your repos and their updates spread across the network faster. Any public repo you publish is mirrored here too, so others can fetch it while your own node is offline. It also speeds up fetching the repos our seed already stores.
z6MkjDYUKMUeY58Vtr8dGJrHRvnTfjKWVGCBYJDVTHXsXzm5@seed.radicle.at:8776Connect once from your terminal:
rad node connect z6MkjDYUKMUeY58Vtr8dGJrHRvnTfjKWVGCBYJDVTHXsXzm5@seed.radicle.at:8776
Or make it permanent by adding it to preferredSeeds in ~/.radicle/config.json:
"preferredSeeds": [
"z6MkjDYUKMUeY58Vtr8dGJrHRvnTfjKWVGCBYJDVTHXsXzm5@seed.radicle.at:8776"
]
Support radicle.at
This is one of the oldest and largest public seeds on the Radicle network, online since March 2024 and storing about 88% of the network's known public repositories. Around the clock and for free, it keeps projects reachable and runs the explorer at app.radicle.at and the API at seed.radicle.at that other tools call.
Running it costs real disk, bandwidth, and time. If our seed is useful to you or to the network, a small contribution keeps it healthy and growing.
Donate to keep it flourishingThe button goes to Liberapay, which handles cards and recurring donations.
DE63 1001 1001 2624 5948 04Account holder Konstantinos Maninakis. Inside the euro area an instant transfer arrives the same minute and costs nothing. From outside EU use BIC NTSBDEB1XXX.
bc1pemmvu4325yuqrn8y44f6rllwp4hqp7znrv3qsr8yv5vx8la3etvqylwnleEvery bit helps cover disk, bandwidth, and upkeep. Thank you for keeping the Radicle network resilient.
- Everything here is aggregated from the seed's own web server access logs and periodic scans of its Radicle storage and node database. Nothing else.
- Zero tracking. No analytics scripts, no cookies, no pixels, no third-party services, no fingerprinting. The only data is what any web server records to answer a request.
- Logs are processed under GDPR: visitor IP addresses are used to group requests, map approximate country and look up their network during aggregation, and are never published here. Only counts, sizes, and coarse breakdowns are shown.
How each number is worked out, and where it could be incorrect
- What is missing. Repos fetched or synced directly between nodes, including over Tor or other private transports, leave no trace here. "Known to the network" (17,467) exceeds "seeded here" (15,439) because our seed can see repos it does not store. The API's request log from 16:59 UTC on 28 June to 17:38 UTC on 28 July 2026 was lost, so every figure drawn from it (web requests answered, API traffic by client type, API responses by status, what gets browsed, bandwidth and API response time) leaves out the API's requests in that window. The explorer's log is complete, so explorer and visitor figures are too.
- People vs. robots is inferred from request patterns and User-Agents, including crawlers that spread across many addresses and imitate a browser. Some automation still passes as people and the odd real visitor is counted as automated, so human counts can be off in both directions.
- Human vs. automated collaboration is inferred per identity from timing: hundreds of issues or patches minutes apart, a steady schedule kept up across days, or patches in bulk and never an issue. A burst that is mostly issues, on a few days, counts as a person moving a tracker in. Automation that acts only when somebody pushes shares their timing and is missed. Only identities checked by hand are ever named as automation. An issue or patch counts as automated when every participant on it is.
- The monthly people-vs-automation series places each issue or patch in the month it was last active, because the seed keeps only the current state of each thread. Recent months are the most complete; earlier ones undercount whatever is still active today. It keeps repos the seed has dropped since 23 August 2026, which the per-repo lists do not.
- Issues opened and closed, time to close and time in draft replay each issue's and patch's own signed history, op by op, for the instant of every close, reopen and draft. Each instant is its author's own clock. Who may act is judged against each repo's delegates today, so a former delegate's action can be misjudged, and only histories that replay to the state the node itself reports are counted. Automated identities are left out, and so are repos pruned before the replay first read them.
- Repos created per month dates each repo by the commit at the root of its Radicle identity, which is the creator's own clock and sometimes wrong, so dates before 2023 or in the future are dropped. It counts every repo our seed has held since 23 August 2026, including ones since pruned. It misses repos pruned before then (over 2,000) and repos our seed never fetched, which skew old, so it undercounts, most of all in the early months. Some recent repos are spam that was pruned within days.
- CI jobs, releases and other COB types count every collaborative object (COB) beyond issues and patches in the public repos our seed stores, each dated by the commit that created it, which is its author's own clock. A repo or node counts from its first dated COB. Repos our seed does not store are missing, so every figure there is a floor for the network.
- The "Freshly created repos" list shows the newest repos that reached our seed, placed by creation date, so one that arrives late lands below the top. Unlike the births chart, it drops repos no longer stored here.
- Seed counts and node counts come from our seed's routing table and shift daily. A seed count is how many nodes announce a repo: a declared intent to serve it, not a verified copy on disk. Nodes with restrictive seeding policies announce fewer repos than they actually serve, so node sizes here are a lower bound. The repo counts and the count of nodes seeding repos cover the last 14 days: a node quiet that long drops out until it returns. The "nodes known to our seed" count has no such limit, so it does not fall when a node goes quiet.
- Estimated node sizes. A node lists at most 2,973 repos in one announcement. On starting up, a node lists its 2,973 lowest repo ids, and since ids are spread evenly, how far up the id range that list reaches gives the node's size. Each later announcement lists a fresh random 2,973, and how much two such lists overlap gives the size too, the way a wildlife census counts tagged animals caught a second time. The estimates use the announcements our seed received over the last 14 days, and a node is shown only once its range is narrow enough to be useful. Ranges are 95% intervals for a node whose size held steady over that time. A node still growing has a true count near the top of its range or just past it, and one that dropped repos in the last few days still shows its earlier size. Our seed is not among them, because its repos are counted from its own storage. "Nodes by size" places each node by what our seed's routing table lists for it, which for these nodes is about one announcement's 2,973 plus any repos it has since announced changes to or our seed has fetched from it, so that chart counts these nodes at fewer repos than estimated here.
- "Official Radicle" repos are a hand-kept list of Radicle-team projects (core protocol, apps and CI), not inferred from delegates, since a maintainer's personal repos share the same delegate.
- The headline ~157 GiB "to mirror every one of them" is an estimate: the exact size of the repos our seed holds, plus the rest priced at the median repo size (0.25 MiB). The largest repos dominate the total (the top 1% hold 50% of the bytes) and are already stored here, so it should land close, but one large repo our seed lacks would move it.
- Visitors and traffic. Country is approximate, from an offline IP database, by distinct IP; one network address is not one person. Repository request counts include every explorer and API request that names the repository, crawlers included. Response times are measured at the seed itself, so requests answered from a browser or upstream cache never reach it and are neither counted nor timed.
- How far back it goes. The node has run since 15 March 2024. Repository stats begin 16 March 2024, web logs April 2024, and the storage-size series late May 2024.
Take the data. Every headline figure above as JSON at /summary.json, and as plain prose at /llms.txt. Both are rebuilt nightly and are CC BY 4.0: take them, chart them, quote them.
radicle.at · an independent Radicle seed · last updated · built with open data from the node itself · Privacy.
seed.radicle.at and stats.radicle.at are independent community projects and not affiliated with the Better Internet Foundation or the Radicle Team.