Kubo v0.43.0-rc2 is out
Second release candidate for v0.43.0, with what came out of rc1 testing. If you ran rc1, the browser transports and delegated routing changed the most.
Get it:
- Binaries: Release v0.43.0-rc2 · ipfs/kubo · GitHub
- Docker:
docker pull ipfs/kubo:v0.43.0-rc2 - Full changelog: kubo/docs/changelogs/v0.43.md at master · ipfs/kubo · GitHub
Highlights:
Native ipfs:// and ipns:// URIs now work as command and RPC input
webrtc-direct gains v2 handshake support, so browsers can keep dialing your node once Chrome drops the old handshake; your /certhash address now also survives restarts
ipfs files no longer hangs when garbage collection runs at the same time
IPNS and DNSLink cache lifetimes now follow each record’s own TTL and expiration
Web UI updated to v4.13
Several memory-exhaustion and crash fixes; update recommended
New since rc1:
WebTransport now speaks draft-15, so Firefox keeps working once it switches, and Safari 26.4+ can open sessions it previously could not.
Provider records sent to an HTTP router now include your AutoTLS /tls/wsandwebrtc-directaddresses. Before, a browser that found your node that way had nothing it could dial.
Behind NAT: losing a relay no longer benches that relay for up to an hour, so a relayed address comes back right away. Two shutdown races are fixed as well, one hanging ipfs daemonon exit, one crashing it mid-reachability-check.
A truncated CAR response from the gateway now ends with an explicit error marker instead of looking complete.- Two more memory-exhaustion CVEs closed, both reachable through the default WebTransport listener.
go-libp2p is on the tagged v0.49.0 instead of a pre-release commit. boxo v0.42.1 and p2p-forge v0.10.1 fix IPv6 HTTP providers, over-long /routing/v1cache windows, and AutoTLS certificates that expired while a node was offline.
Please report anything you hit, especially if browsers dial your node directly, you provide to a delegated HTTP router, you serve CAR responses, or your node sits behind NAT.