lifted from Discord
:
: Thursday Update ::
Hello @everyone — weekly update time. We’ve been watching support channels closely — our AI agents now scan them every night — so this update is shaped by what you’ve reported, what has improved, and what the team is working through next.
Uploads and payments A lot of recent feedback has been around uploads and payments, and several improvements are landing in the next planned release. For large uploads that complete payment but don’t return an address, we’ve added a download-style deferred retry for Merkle uploads, so a stalled final step gets another chance instead of failing — plus diagnostics that log where a transfer got stuck. We’ve also tightened the payment path: token approval is now held back until actual payment, quote freshness is judged on a node’s own stored state rather than its neighbours’, and a disk-space pre-check helps avoid a node taking payment for a chunk it can’t store. Merkle payments now accept flexible single-node proofs and a more lenient closeness threshold, reducing avoidable rejections and retries. Net effect: uploads should be more reliable, with less wasted retrying and fewer cases where cost climbs.
Keeping your data safe A few of you have reported downloads slowing or stalling near the end, sometimes with chunks reported as lost. We’re using telemetry from those cases to separate temporary unavailability from slower repair paths, and focus replication work where it matters most. Storage auditing is also being strengthened. The next step is a deeper check where nodes prove they genuinely hold data by fetching a real chunk, not just presenting a claimed hash. In controlled testing this has shown promise against bad actors while leaving honest nodes untouched, but it is being tested carefully before release. Replication work continues too: the aim is to make repair faster when a node leaves, while avoiding false punishment of honest nodes that are briefly lagging.
Nodes, upgrades and full drives Some node operators have reported restart issues after upgrades, and others have asked for safer behaviour as drives fill. We’ve made another fix around node state transitions and are continuing to investigate the remaining restart reports, while approaching-full handling remains an active improvement area. The good news you’ve been sharing: V2 uploads are feeling much better than V1, and many upgrades have landed cleanly this cycle — thank you.
The Reading Room Real proof, growing weekly: over 150 academic papers from the Foundations collection are now stored on Autonomi and served peer-to-peer, readable and downloadable straight from the network. The four-book collection is live to explore, links now carry proper filenames, and the in-browser reader has had an overhaul. Permanent data is doing the job it was built for: preserving and opening access to important knowledge without relying only on centralised, contract-dependent infrastructure.
Browser access A recurring question, so worth being clear: direct raw browser access is not the right route for us, because it would weaken the post-quantum protections that make the network what it is. The path we’re building instead is a small, secure local helper that renders autonomi:// links with that protection intact. More on this soon.
x0x and agents The app layer keeps taking shape — agent-to-agent messaging, and late joiners now receiving full state rather than only what changes next. It’s great to see the use-case ideas coming from you — agent swarms, messaging patterns, sovereign-cloud framings — keep them coming. Behind the scenes, our AI agent team runs daily network-health and signal scans, helping pinpoint the issues you report, while ADRs help humans stay in control as AI writes more code. The next release is planned for tomorrow and includes the bulk of the upload and payment improvements above. Please keep the reports coming :
:
EDIT: neo: fixed some formatting
