Exploring the roots and future of Internet Relay Chat

EFnet vs Undernet: The Fork That Shaped IRC Forever

Every fork in open-source software is a statement. It says: "We think there's a better way to do this, and we're willing to do the work to prove it." The Undernet's separation from EFnet in 1992 was exactly that — a group of people who looked at how IRC was being run and said, "We can do better."

Whether they were right depends entirely on what you think "better" means.

Two Philosophies of Online Communication

The EFnet/Undernet split represents something deeper than a technical disagreement. It's a fundamental philosophical divide about how online communities should be governed:

EFnet's position: The network provides the infrastructure. What users do with it is their business. Minimal rules, minimal intervention, maximum freedom. If your channel gets taken over, that's your problem — run a better bot.

Undernet's position: The network has a responsibility to provide a stable, fair environment. Channel registration, abuse prevention, and user services are network responsibilities, not things that should be left to individual users to figure out on their own.

You can see echoes of this debate everywhere in modern platform design. Discord and Slack lean heavily toward the Undernet model — centralized services, registered identities, structured moderation tools. Matrix and some Fediverse platforms lean more toward EFnet's decentralized ethos. Neither approach is wrong. They serve different values.

The EFnet Way: Freedom and Chaos

EFnet in the early 1990s was the Wild West of the internet. The name itself — "Eris-Free network" — came from a specific event in 1990 when the server eris.berkeley.edu was delinked from the network after its administrator refused to implement server-level bans (K-lines) against abusive users. The administrator's position was that his server should be a free speech zone. The rest of the network disagreed. Eris was cut off, and the remaining network adopted the name EFnet to distinguish itself.

The irony is that EFnet, despite ejecting the most extreme free-speech absolutist server, remained deeply committed to a hands-off philosophy. Server administrators (opers) had power over their own servers but generally avoided interfering with channel-level affairs. If two groups were fighting over a channel, that was between them. The network didn't pick sides.

This created an environment that was simultaneously thrilling and frustrating. Channels could be vibrant, creative spaces where brilliant people gathered to share knowledge and build things. They could also be war zones where bot armies clashed for control, and ordinary users were caught in the crossfire.

Channel takeovers on EFnet were an art form — and I mean that without irony. The techniques involved were sophisticated: exploiting netsplits, social engineering server administrators, compromising channel bots, coordinating timing across multiple connections. It was a cat-and-mouse game that drove the development of increasingly sophisticated protection tools.

Birth of the Undernet

The Undernet was founded in October 1992. The key figures included Laurent Demailly (dl), Danny Mitchell, and others who felt that EFnet's approach to network management was unsustainable. Their core belief was simple: a network should serve its users, and that means providing tools and services that make the experience better for everyone, not just the technically sophisticated.

The early Undernet was small — a handful of servers, a few hundred users. But it grew steadily because it offered something EFnet didn't: stability. Channels on the Undernet didn't get taken over every other week. Nicknames could be registered and protected. There were actual rules, and they were enforced.

The Undernet also innovated at the technical level. Their ircu server software — a fork of the original ircd — implemented features that EFnet's server code didn't have:

  • Timestamped channels: Each channel had a creation timestamp. During netsplit recovery, the older channel took precedence, preventing split-riding takeovers.
  • Juped nicknames/channels: The ability to reserve nicknames and channel names at the network level.
  • Improved flood protection: Better throttling of excessive messages, reducing the effectiveness of flood attacks.
  • Connection classes: Different server-level settings for different types of users, allowing better resource management.

The Services Debate

The single biggest difference between EFnet and the Undernet was — and remains — services. The Undernet's channel service (CService, later known as "X") was a centralized system for channel registration and management. If you registered your channel through CService, you had:

  • Persistent channel ownership that survived server restarts and netsplits
  • An access list system with multiple levels of privilege
  • Automatic channel mode enforcement (topic lock, invite-only, moderation)
  • A ban system that worked across the entire network, not just on individual servers
  • Protection against takeover attempts by unauthorized users

EFnet opers viewed this with suspicion — and still do. Their argument: centralized services create a single point of failure and a single point of control. If the services go down, every registered channel is vulnerable. If the services team becomes corrupt or biased, they can effectively censor any channel they want. Better to let users manage their own channels with bots, where no central authority can interfere.

Both sides had valid points. EFnet channels that were well-run by dedicated operators with good bots were extremely resilient. But channels without that level of dedication were easy prey. The Undernet's services made good-enough security available to everyone, at the cost of centralization and the bureaucracy of the registration process.

Cultural Differences

Over time, the different governance models created distinctly different cultures:

EFnet: Technical, often abrasive, darwinistic. If you couldn't handle yourself, that was your problem. Help channels existed but expected you to have done your homework first. The average EFnet user was more technically sophisticated than on other networks, partly because you had to be to survive. "Read the RFC" was a common — and serious — response to questions.

Undernet: More welcoming, more diverse, more... normal. The services safety net meant that non-technical users could participate without being instantly victimized. Channels covered a wider range of topics — not just technology but hobbies, geography, language practice, social chat. The average user was less likely to know what a K-line was and more likely to just want to talk to people.

Neither culture was inherently superior. EFnet's harshness filtered for competence, creating some of the most knowledgeable technical communities on the early internet. The Undernet's accessibility created some of the most vibrant and diverse communities. Both had their share of trolls, drama, and petty politics — because both were populated by humans.

The Peak Years

Both networks peaked in the early 2000s. EFnet regularly had 40,000-50,000 simultaneous users. The Undernet peaked at around 120,000 simultaneous users — its services and user-friendly approach attracting a larger but less technical user base.

For a few years, it genuinely felt like IRC would be the default communication medium of the internet. Every open-source project had an IRC channel. Every gaming clan coordinated on IRC. Tech companies used IRC internally. The question wasn't whether you used IRC, but which network you were on.

Then came the DDoS attacks of the early 2000s, the rise of MSN Messenger and AIM, the eventual dominance of social media, and later, Slack and Discord. Both networks declined — but they declined differently. EFnet's decline was gradual, its users slowly migrating to newer platforms over years. The Undernet's decline was punctuated by severe DDoS attacks that disrupted service for extended periods and drove users away in waves.

Where They Stand Today

In 2026, both networks are still operational. EFnet runs about 15-20 servers and hosts roughly 10,000-15,000 users at peak times. The Undernet has a similar server count and user base. Both are shadows of their former selves in terms of raw numbers, but the communities that remain are dedicated and active.

EFnet still has no services. Channels are still managed by bots. The culture is still technical and unforgiving. It's a living museum of the original IRC ethos — and I mean that as a compliment.

The Undernet still runs CService (now called "X"). Channels are still registered through an application process. The culture is still more accessible than EFnet's. It's evolved, but its core values haven't changed.

The fork that happened in 1992 produced two networks that each proved their model could work. Neither killed the other. Neither was proven wrong. They simply demonstrated that there's more than one valid way to run an online community — a lesson that's worth remembering every time someone argues that their favorite platform's approach is the only possible one.

Frequently Asked Questions

Can you use the same IRC client for both EFnet and Undernet?

Yes. IRC clients like irssi, WeeChat, and HexChat work with any IRC network. You simply configure different server connections for each network. The protocol differences between EFnet and the Undernet are at the server level — from the client's perspective, they're both standard IRC.

Why doesn't EFnet add services?

It's a philosophical choice that EFnet has maintained for over 30 years. The EFnet community values decentralization and user autonomy over convenience. Adding services would fundamentally change the network's character and culture. There's also a practical concern: retroactively adding services to a network that has operated without them for decades would create enormous political conflicts over channel ownership.

Which network should a new user choose?

For most new users in 2026, neither EFnet nor the Undernet is the obvious first choice. Libera.Chat is more active for open-source and technical communities. If you specifically want the EFnet or Undernet experience, the Undernet is more beginner-friendly due to its services. EFnet is better if you want the raw, unmediated IRC experience and are comfortable managing your own channel security.

Did the Undernet fork affect other networks?

Significantly. The Undernet's services model — NickServ, ChanServ, and related services — became the template that DALnet, QuakeNet, and virtually every subsequent IRC network adopted. The services packages used today (Anope, Atheme) are direct descendants of the concepts the Undernet pioneered. In a sense, the Undernet won the services debate — not by defeating EFnet, but by establishing the model that everyone else followed.

FAQ

What is the difference between EFnet and Undernet?

EFnet (Eris-Free Network) is the oldest IRC network, favoring minimal rules and no channel registration services. Undernet introduced CService and channel registration to provide stability and reduce takeovers.

Which network is bigger?

Both have declined from their peaks. In the early 2000s, EFnet had around 100,000 concurrent users and Undernet around 150,000. Today both are much smaller but still active.

Does EFnet have services like NickServ?

No. EFnet deliberately chose not to implement NickServ or ChanServ. Nick and channel ownership is enforced socially and through individual server policies, not centralized services.