Middle relays: the workhorse tier of the Tor network
Exit relays get the headlines and guard relays get the security analysis, but between them sits a tier that gets far less attention. Middle relays are the non-exit relays that carry traffic from hop to hop without ever connecting to the public internet. The Tor Project's relay documentation describes the current network as "quite small compared to the number of people who need to use Tor" and makes the case that more relays mean better stability, more robustness against attacks and greater user safety. Here is what the middle tier actually carries, why network health depends on it, and what running one involves.
What a middle relay carries
A standard Tor circuit has three hops: a guard, a middle and an exit. The middle's position is fixed by that design — it sits between the guard, which knows the client's address, and the exit, which knows the destination. The middle sees neither end. Traffic to onion services travels through the network in a similar relayed fashion, with additional relays serving as introduction and rendezvous points, so the middle tier carries that flow as well. Tor's overview of relay types states plainly that "a non-exit relay does not allow exiting in its exit policy" — it only shuttles traffic between other relays. And a relay that is not fast and stable enough to earn the guard flag "will remain a middle relay".Guard and middle relays usually do not receive abuse complaints.That line, from the Tor Project's relay documentation, is the practical reason many operators choose this tier. Because the relay never connects to outside services, there is no exit traffic for anyone to complain about. The documentation adds one caveat: all relays are listed in the public directory, so platforms such as banks and streaming services may still block relay addresses. That is an IP-reputation inconvenience, not an abuse complaint.
Why the middle tier matters for network health
Two things depend on it. The first is capacity. Tor Metrics tracks the number of running relays and publishes bandwidth and relay-flag graphs, and the three-hop design means every circuit uses a relay in the middle position. More middles spread that load across more machines; when the tier thins out, guard-flagged relays end up doing middle duty as well, and the pool of distinct paths shrinks. The second is anonymity itself. Tor keeps a client's entry point stable by assigning guard relays that rotate only every few weeks — the Tor Project's write-up on the lifecycle of a new relay describes guard rotation of four to eight weeks, later extended to twelve. That stability defends against predecessor-style attacks, in which an adversary watches both ends of connections and waits for the observed hops to overlap. A large middle tier adds an unpredictable hop between guard and exit, so an adversary observing a guard still faces a wide field of possible middle relays before traffic reaches an exit it also controls.Running one is deliberately simple
The Tor manual covers everything a middle needs in its server options, and the list is short:- An ORPort, defined in the manual as the port to "listen for connections from Tor clients and servers", plus a nickname and a working contact address.
- No exit policy to configure — with exiting disabled, the relay ignores exit policies entirely and never talks to destinations outside the network.
- Bandwidth controls: BandwidthRate and BandwidthBurst cap throughput, and the project's support documentation notes there are "hibernation options where you can tell Tor to only serve a certain amount of bandwidth per time period", which suits capped connections.
- A MyFamily entry whenever one operator runs several relays. The manual is blunt: "If you run more than one relay, the MyFamily option on each relay must list all other relays", so clients never place two of them in the same circuit.