Rayfish
xyz.rayfish.android
Zapstore _@zapstore.dev Republished from GitHub / F-Droid by the Zapstore main account.
First release: Sep 29, 2026, 11 total releases.
Most recent release: Nov 3, 2026.
Appears in 0 app stacks.
0 sats / 0 zaps received in the past year.
Sats Received
Underlying data available via MCP: app_zaps, app_releases.
Zap Count
Underlying data available via MCP: app_zaps, app_releases.
Releases
- Nov 3, 2026 0.5.15### Bug Fixes - **mesh**: Disable Linux UDP GSO until fallback recovers ### Documentation - **changelog**: Prepare 0.5.15 release ### Miscellaneous
More…
- Release **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.14...v0.5.15 - Nov 2, 2026 0.5.14### Features - **exit-node**: Use Mullvad's 10.64.0.0/10 range for IPv4 exits - **files**: Resume accepted receives after a daemon restart - **files**: Resume interrupted receives in the background ### Bug Fixes - **ci**: Regenerate the Apple and iOS Swift bindings
More…
- **ci**: Regenerate the Kotlin bindings and resolve the macOS temp dir in a test - **macos**: Build the ssh child-process helper on every desktop unix - Restore exit node forwarding and DNS routes - **files**: Keep Windows device names and long names saveable - **mesh**: Stop handing out a record older than the coordinator's floor - **dns**: Drop a peer's old name when it comes back under a new one - **cli**: Escape control characters in peer-supplied names - **files**: Rerun an outbox flush when an offer is queued mid-pass - **files**: Clear a send's interrupted note once it completes - **files**: Find a file saved before a restart instead of saving it twice - **files**: Give each offer one temporary name so a resume replaces its leftover - **files**: Satisfy clippy in the give-up span - **files**: Report a short give-up window in minutes - **files**: End a receive whose blob is not the offered size - **files**: Cut a received name that is too long for one directory entry - **files**: Sync a received file before giving it its name - **mesh**: Hand reconnecting members the record last published - **mesh**: Keep the record floor across restarts - Abort network tasks that outlive teardown - Refuse a network already live under another name - **management**: Admit delegated joins only through the controller - **dns**: Warn once about a roster hostname Magic DNS cannot serve - **listener**: Skip backoff and warning for clients that leave before accept - **dns**: Clamp the EDNS size in place instead of re-encoding the query - **firewall**: Hand off the async worker while an edit saves - **exit-node**: Keep DNS in the tunnel when upstreams name our own resolver - **mesh**: Ignore a replayed current record on the signed-record fast path - **mesh**: Confirm a kick only with a record newer than the last applied - **mesh**: Read the device cert on each member reconverge - **mesh**: Start a network's tasks only once its install wins the name - **files**: Offer an interrupted send again once the recipient stops resuming it - **android**: Receive each file into its own folder and bound the accept - **files**: Count a receive's give-up by time spent trying, not elapsed - **windows**: Keep a LocalSystem receive's caller across a restart - **files**: Check the destination before downloading an accepted file - **files**: Stop auto-accept following a link out of the configured directory - **files**: Write a receive under a temporary name and rename it into place - **files**: Record a delivery before marking the receive done - **cli**: Say not authorized once in ray status --json - **management**: Send a delegated leave under the network's exact name - **macos**: Build ray stop again - **ipc**: Refuse staged file frames from clients and path sends on the macOS tunnel - **files**: Build the startup resume only where it runs - **macos**: Return from file accept once the receive starts - **files**: Show an interrupted send as in progress on the sender - **windows**: Save accepted files only where the caller can write - **ssh**: Accept ssh off and refuse ssh on wherever mesh SSH is unavailable - **daemon**: Poll the signed record on every coordinator - **management**: Limit delegated join and leave to the controller's networks - **mesh**: Ignore unsigned hostname claims on plain members - Send ray files accept's output directory as an absolute path - Drop the PTR of a peer renamed to an invalid name - Write received files only into the caller's own directory - Require root to set download-dir and download-user - Untangle split doc comments and drop a sleep-poll from a test - Refuse a fresh join of a network live under another name - Keep coordinator replies consistent with live state when a save fails - Release the restore claim before a warm resume's version check - Keep a queued join's transport across a restart - Record an admin grant even when the snapshot cannot be saved - Keep the path-send fallback and launchctl stop as before - **apply**: Reject nested groups instead of publishing them as hostnames - **pair**: Reject a ticket combined with a pair subcommand - **complete**: Read the offline config without creating or migrating it - **firewall**: Open the pending picker only on a terminal - **pair**: Print the ticket and pairing result as JSON under --json - **files**: Honor --json for auto-accept list, download-dir and download-user - **invite**: Print a minted invite as JSON under --json - **cli**: Print daemon Ok messages as JSON under --json - **uninstall**: Require root before tearing down the service - **stop**: Fail when the service manager cannot stop the service - **cli**: Report a refused daemon socket as not authorized - **cli**: Name the ray binary in pairing and files help text - **send**: Fail on an unexpected reply on Windows - **status**: Fail under --json when the daemon is down or refuses - **cli**: Reject durations that overflow instead of wrapping - **ephemeral**: Fail on an invalid duration, before connecting - **apply**: Keep live suggestions when they cannot be read - **apply**: Exit non-zero when a network, suggestion or invite step fails - **open**: Join links auto-accept own-device files like ray join - **gui**: Drain command output while the child runs - **ssh**: Cache the sftp lookup and bound sshd -T - **stats**: Stop exporting metrics for peers that left - **daemon**: Run the host firewall probe off the async workers - **mobile**: Drop a packet the Android TUN only partly wrote - **mobile**: Do not mint an identity when reading stopped status - **ssh**: Start sessions in / when the home directory is missing - **rootless**: Keep proxy listeners up after a failed accept - **v4bridge**: Back off on accept errors and cap connections per port - **ssh**: Keep the macOS helper listening after a failed connection - **exit-node**: Re-enable pf when the saved token is stale - **dns**: Capture IPv6 upstreams on macOS - **dns**: Truncate forwarded replies that do not fit the TUN - **dns**: Keep a peer's PTR while another network still holds it - Free the registry when the daemon is dropped - Drop idle invite delivery locks - Prune finished transfers on insert - Answer a connect retry from the saved direct network - Run one connect retry loop per peer - Reject ambiguous and empty short ids across networks - Cap the ping interval - Build ray report off the async workers - Report a hostname that could not be saved - Read the current device cert after pairing or unpairing - Bound inline steps in the mesh control loop - Flag a warm-resumed network on a mesh version mismatch - Give cold-registered member networks a reconverge worker - Claim the name on fresh join and never replace a live network - Run one pending-join retry per network and stop once live - Start the stale-member pruner on promotion - Match resumed pending joins to live networks by key - Report an error when a coordinator change cannot be saved - Serialize firewall edits so concurrent changes are not lost - Skip invalid saved destruction proofs instead of dropping all of them - Keep publishing when a saved destruction proof is unreadable - Apply every signed record through one routine - Validate hostnames peers claim - Honor --tor on create and join - Refuse path-based file sends from non-root callers - Reject file offers whose name escapes the download directory ### Performance - **forward**: Derive the peer address once per inbound reader - **membership**: Derive the mesh IPv6 without allocating ### Refactor - **daemon**: Keep a network's group contents as one GroupBlob - **membership**: Carry rosters as MemberList and ApprovedList - **membership**: Keep the roster in identity order - **firewall**: Parse excluded-peer selectors without nom - **daemon**: Build the pkarr relay client once per daemon - **config**: Read and write settings.toml through AppConfig - **cli**: Move the clap definitions and process setup out of main.rs - **config**: Split the schema, paths and Windows operator file out of config.rs - Move forward, firewall, config and ssh tests into tests.rs - **daemon**: Move the test modules out of daemon/mod.rs - **daemon**: Build NetworkRegistry from named handles - **daemon**: Read shared handles from the registry, not copies on Daemon - **daemon**: Pass the registry instead of the MeshCtx handle bundle - **cli**: Send single-reply requests through ipc_request - **daemon**: Typed results for status, pairing and managed machines - **daemon**: Typed results for firewall and settings operations - **daemon**: Typed results for file sends, receives and the file list - **daemon**: Typed results for join requests, connects and invitations - **daemon**: Typed results for create, join, invite, leave and rename - **daemon**: Register every live network through one install step ### Documentation - **changelog**: Prepare 0.5.14 release - **changelog**: Say which Windows receives now resume - **firewall**: Keep normalize_ssh_users' doc on its function - **daemon**: Coordinators now poll the signed record - Changelog for the audit fixes ### Testing - **files**: Run the app-accept tests off Windows only - **e2e**: Refuse a boat fleet record in the droplet scripts - **e2e**: Add a Boat VM backend - **e2e**: Stop forwarding LC_ALL to the nodes - Property tests for decoders of untrusted input - **e2e**: Add files-authz scenario for file send and save authority - **e2e**: Add coordinators scenario for multi-coordinator edits, restarts and racing joins - **e2e**: Add manage scenario for controller scope - **e2e**: Add file-resume scenario for interrupted receives - **e2e**: Check mixed-version rosters by membership and mark old-build defects - **e2e**: Add a mixed-version mesh scenario - **e2e**: Read mesh addresses from ray status --ips - **mesh**: Give the delegated join test a key claim - **apple**: Restore expect_ok response handling test - **mesh**: Cover the floor a resumed member takes from its current record - **mesh**: Run the member apply tests on a warm-resumed plain member - **files**: Wait for the requeued offer before reading the saved queue ### Miscellaneous - Release - Merge master into fix/review-round - Put the evict_oldest_file doc back on its function **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.13...v0.5.14 - Oct 31, 2026 0.5.13### Features - **windows**: Manage SSH access and reduce connect setup - **windows**: Add firewall page and fix SSH and ping setup - **windows**: Support embedded SSH sessions for local accounts - **windows**: Match macOS design and add desktop shortcut - **exit-node**: Route IPv4 and add macOS menu selection (#210) - **macos**: Select exit nodes from menu bar - **exit-node**: Tunnel IPv4 traffic and block direct fallback
More…
- **windows**: Replace WebView2 with native desktop controls ### Bug Fixes - **windows**: Configure mesh routes natively and report setup failures - **windows**: Keep confirmation dialogs compact - **windows**: Keep SSH terminal output on the pseudoconsole - **windows**: Suppress noninteractive PowerShell progress output - **windows**: Complete SSH profile creation and auth shutdown - **windows**: Stream SSH text as UTF-8 without a console - **mesh**: Synchronize coordinator rosters before renewal - **windows**: Handle fresh SSH accounts and rejected connections - **mesh**: Preserve admitted networks across stale roster updates - **windows**: Reserve SSH endpoints and bound session cleanup - **windows**: Show when network admission needs approval - **windows**: Finish quitting while the window is hidden - **windows**: Identify IPC callers after reading requests - **windows**: Keep the daemon alive during standby - **windows**: Use daemon IPC for desktop actions and status - **windows**: Place direct shortcuts in their own component group - **windows**: Correct shortcuts and desktop lifecycle - **dns**: Retain encrypted fallbacks for underlay lookups - **dns**: Preserve transport resolver options across resets - **exit-node**: Merge master and resolve integration conflicts - **exit-node**: Drop unsolicited inbound to clients and route both families - **exit-node**: Merge master and address review findings - **exit-node**: Fall through to a usable exit selection - **exit-node**: Stop rebuilding pf for every macOS DNS socket - **exit-node**: Skip LAN resolvers while an exit is selected - **exit-node**: Keep IPv6 transit when the IPv4 pool overlaps a route - **exit-node**: Let macOS underlay sockets past the exit filter - **exit-node**: Keep inbound replies off the Linux tunnel - **macos**: Match generated Swift file ending - **exit-node**: Satisfy macOS Clippy condition lint - **ping**: Report per-probe selected-path observations (#271) - **ping**: Detect replacement after weak connection handle expires - Expire and bound cached endpoint hints - Preserve endpoint hint recovery across write failures - Cache the illumos MTU only after it is applied - Keep the macOS DNS toggle and Windows standby off short names - Keep dns off away from the OS and derive the overlay MSS - Size the overlay route for the TUN and require .ray by default - Batch and throttle endpoint hint writes - Expire and bound cached endpoint hints - **ping**: Report per-probe selected-path observations ### Refactor - Keep endpoint hint management internal ### Documentation - **changelog**: Prepare 0.5.13 release ### Testing - **windows**: Exercise installed tunnel packets through Magic DNS - **e2e**: Wait for legacy invite discovery before asserting - **macos**: Cover exit-node menu state - **ping**: Wait for retired QUIC connection teardown before checking weak handle ### CI - **windows**: Retain installer build dependencies - **macos**: Match extension build architecture - **macos**: Build packet tunnel extension ### Miscellaneous - Release - Release - Merge upstream changes - Stop tracking the macOS Xcode build output - Merge master into exit-node branch - Merge master into IPv4 exit branch - Sync upstream and reconcile changelog - Merge master into endpoint hint expiry - Merge upstream master into ping probe fix ### Other - Bump tun-rs (cache the MTU only after it was applied) **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.12...v0.5.13 - Oct 28, 2026 0.5.12### Bug Fixes - Repair reused SSH sessions and pending request counts ### Documentation - **changelog**: Record v0.5.11 release ### Miscellaneous
More…
- Release **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.11...v0.5.12 - Oct 27, 2026 0.5.11### Features - **linux**: Add rootless mesh proxy mode (#262) - **linux**: Discover rootless IPC and print startup instructions - **linux**: Support rootless HTTP proxy and CONNECT - **linux**: Add rootless mesh proxy mode - **macos**: Confirm shell command installation in settings - **ios**: Show the Rayfish logo in the Control Center toggle - **ios**: Add a Control Center toggle for the VPN
More…
### Bug Fixes - **mesh**: Forget a direct hint that reaches a different endpoint - **dns**: Preserve IPv6 control-plane resolvers - **ssh**: Close channels the client closes, via russh 0.64 - **ssh**: Cap connections accepted on reverse and agent sockets - **ssh**: Resolve login groups once at login - **ssh**: Cap macOS helper connections per source - **ssh**: Limit socket helpers per account and bound the wait - **ssh**: Close the daemon's descriptors in socket helpers - **ssh**: Refuse self-SSH from sockets the daemon holds - **ssh**: Bound socket helpers and count connections per peer - **ssh**: Close review gaps in socket forwarding - **ssh**: Run socket forwarding as the login account - Acknowledge peer requests and stop refused retries - **ci**: Handle missing daemons and platform guards - **firewall**: Keep return traffic through TCP half-close - **linux**: Correct rootless isolation and proxy handling - Fail fast for denied mesh SSH connections - Retry approved connections after join failures - **illumos**: Clean up stale tunnel interfaces (#260) - **pair**: Keep networks from stopping while a revocation is lifted - **pair**: Refuse to lift a revocation while a network is not running - **pair**: Re-authorize a device before it gets its new certificate - **pair**: Count a re-pair only once the device saved its certificate - **pair**: Keep a new pairing from a late unpair of the earlier one - **pair**: Keep each unpair until the other side confirms it - **pair**: Retry a device's unpair request until its primary receives it ### Refactor - **linux**: Simplify rootless state and URI handling ### Documentation - **changelog**: Move the unpair entries from 0.5.10 to Unreleased - **changelog**: Move the unpair entry back under Unreleased ### Testing - **ssh**: Check the helper's peer pid only on Linux - **ssh**: Fix macOS dead-code lints in self-SSH tests - **e2e**: Check SSH forwards run with the login account's rights - **linux**: Run rootless mesh without container privileges - **linux**: Cover rootless mesh proxies end to end - **pair**: Wait for the restore to end before stopping the network again - **pair**: Keep the re-pair test's primary off the device's networks - **dns**: Give the queued DNS transaction time to finish - **pair**: Keep the device offline while the test speaks with its key ### Miscellaneous - Release - Release - Merge master before SSH fix push - Merge master connection retry fix - Merge master tunnel dependency updates - Merge master for Windows dependency repair - Merge master into rootless proxy branch - Merge master into illumos followups - **macos**: Log connection startup timings ### Other - Illumos, unit tests for the resolv.conf helpers - Illumos, keep the backup when resolv.conf is unreadable; keep host search domains on revert - Illumos backend taking over /etc/resolv.conf, with crash recovery - Bump tun-rs (fatal message read mode, MTU on both families) - Bump tun-rs (lock enabled(), drop order, probe errors) - Drop the swarm-discovery patch: illumos support is in 0.6.5 - Bump tun-rs (unplumb IPv6 on drop, remove stale tunN) - Bump tun-rs (unplumb IPv6 on drop, review fixes) - Pin netwatch to the illumos-0.19 branch (upstream main is 0.20.0) - Bump netdev (lifreq fix), netwatch and tun-rs review fixes - Bump tun-rs (quiet enabled() on IPv6-only devices) - Bump tun-rs for the illumos review fixes - Implement set_link_state for illumos - Push ptem before the initial pty resize on illumos - Rustfmt the illumos cfg - Point swarm-discovery at the illumos branch - Fix swarm-discovery branch - Skip login(1) handoff on illumos - Push ptem/ldterm onto the pty slave on illumos - Port to illumos/OmniOS **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.10...v0.5.11 - Oct 19, 2026 0.5.10### Features - **ios**: System body font, QUIC engine switch, automatic Keychain backup - **ios**: Report to the same Sentry project as Android - **mobile**: Name a new device after its Settings name - **ios**: Report crashes and diagnostics to Sentry - **cli**: Add local network rename (#240) - **cli**: Add ray rename to rename a network on this machine - **macos**: Join or decline an invitation from its notification
More…
- **apps**: Invitations, invite by identity, blocking and inbound policy - **apply**: Invite aliases privately - **invite**: Deliver invites privately and retry - **invite**: Private invitation inbox - **connect**: Block list, inbound policy and deny - **config**: Add inbound policy and block list - **connect**: Connect by identity, drop contact keys - **ios**: Save the identity to iCloud Keychain without Face ID - **ios**: Deploy directly to a connected iPhone - **ios**: Add just ios-run to build and launch on an iPhone - **ios**: Back up the identity to iCloud Keychain and tidy the app - **ios**: Add the iOS app - **daemon**: Clean up after a dead daemon from the service manager - Support settings on ray up ### Bug Fixes - **ios**: Bump the app version with each release - **mobile**: Make idle phones outgoing-only to stop overnight wakeups - **ios**: Apply the crash reporting toggle on the main thread and before each report - **config**: Move a network's invites with its rename - **config**: Keep exit and SSH state on rename, finish it after a crash - **mesh**: Keep a restore sweep off a network still being created - **mesh**: Order handle announcements and close races around join lanes - **mesh**: Admit joins for different networks side by side - **connect**: Keep valid invitations, bound idle dialers - **connect**: Own ALPN for invite offers, tighter blocking - **connect**: Bound invite codes, block a user's devices, keep app rejects - **android**: Survive a refused foreground start, group exit reports by reason - **mesh**: Release the dial lock once a join registers the peer - **daemon**: Raise the open-file hard limit in the daemon, not the service files - Gate FatalExit::arm off Android, correct the fwmark del comment - **daemon**: Do not re-apply exit or TUN state that ray down is removing - **exit**: Leave accept_ra alone where an interface already forwarded - **exit**: Do not read an nft without JSON as a missing table - **dns**: Reload NetworkManager after removing a stale drop-in - **dns**: Keep a stale backup until its restore succeeds - **ssh**: Probe the whole scope wrapper, off the runtime - **sysv**: Only ever raise the hard open-file limit, note no respawn - **v4bridge**: Hold no ports on a host with only loopback IPv4 - **dns**: Withdraw the resolvconf stanza after a crash - **ipc**: Cap log streams and time out stalled log sends - **exit**: Clamp TUN MSS to a fixed 1220 - **forward**: Refuse tiny first fragments and unrewritable SSH packets - **dns**: Create the resolv.conf temp file fresh, never through a link - **exit**: Put back the exit route and mesh SSH accept without waiting - **config**: Copy the blob store when /var is a separate filesystem - **exit**: Replace forwarding exceptions left for another TUN - **dns**: Reload NetworkManager and respect newer files on emergency restore - **ssh**: Fall back to a direct spawn when systemd scopes fail - **daemon**: Keep the Linux blob store in /var/lib/rayfish - **daemon**: Name the daemon lock's file field - **daemon**: Lock out a second daemon before probing the socket - **daemon**: Rebind after a suspend on Linux - **ipc**: Cap concurrent clients and time out stalled requests - **ssh**: Run direct sessions in their own systemd scope - **ssh**: Resolve supplementary groups before forking the session - **config**: Create staged config files with their final mode - **update**: Stage the new binary privately beside the executable - **v4bridge**: Stop presenting mesh peers as loopback clients - **forward**: Carry IPv6 fragments instead of refusing them - **forward**: Stop counting the kernel's link-local chatter as drops - **tun**: Restore the Linux TUN after an external link down - **tun**: Set up the Linux TUN's IPv6 before the address goes on - **forward**: Restart the daemon when the TUN read loop dies - **exit**: Re-apply exit-node kernel state another tool removed - **exit**: Keep the Linux gateway's uplink and forward path working - **exit**: Stop clobbering other tools' marks in the client tunnel - **hostfw**: Read both firewall ladders before calling a port clear - **dns**: Gate the resolved ownership helpers to Linux - **dns**: Reset TCP connections to the magic resolver - **dns**: Keep IPv6 nameservers in direct mode - **dns**: Pick systemd-resolved only when it owns resolv.conf and is running - **dns**: Reload NetworkManager over D-Bus, read tab-separated search lines - **dns**: Write resolv.conf atomically and never through a symlink - **dns**: Revert the held DNS backend before configuring again - **dns**: Only restore a stale resolv.conf backup over our own file - **dns**: Stop the resolved link becoming a default DNS route ### Performance - **forward**: Read the clock only for fragments on the inbound path - **exit**: Read the kernel check from one ruleset dump, unlocked - **v4bridge**: Pick the host address per scan, not per connection - **tun**: Create the Linux TUN at a 1280-byte MTU ### Refactor - **mobile**: Drop the iOS engine mirror and keep dots in device names - **mobile**: Sentence-case error text in the core, tidy iOS telemetry - **apple**: Share the provider reply helper between macOS and iOS - **mobile**: Let the core own error text, device names and records - **mesh**: Share the signed-record publish path, trim test setup - **mesh**: One publish-and-mark helper, bounded join lanes - Simplify identity invites - **daemon**: Share the own-identity lookup - **daemon**: One restore_host_state for the panic hook and cleanup - **config**: Share the owner group choice, simplify the store copy check - **dns**: One upstream list for both families - Share the bounded child wait - **tun**: Share the netlink link lookup and sysctl helpers - **dns**: Refuse DNS over TCP with the reject module's reset ### Documentation - Update changelog for 0.5.10 release - Release the apps only when a version is cut - **changelog**: Say what survives a daemon restart under mesh SSH - **changelog**: Note mesh SSH accept reassert - **exit**: Reword the MSS clamp comment for a 1280 TUN - **exit**: The Linux TUN MTU is 1280 in the MSS clamp comment - **changelog**: Linux hardening fixes - Update changelog for 0.5.9 release ### Styling - Import PoisonError only where it is used - Use imports and poison-tolerant locks in new code ### Testing - **e2e**: Cover revoke, restarts and an offline minter; run invites in CI - **e2e**: Invites scenario for the inbound gate and private delivery - **e2e**: Deliver an alias's invite privately and accept it - **config**: Run the blob store migration tests on Unix only ### CI - Keep nightly CLI and daemon builds, drop the APK from them - Stop publishing nightly builds ### Miscellaneous - Release - **android**: Regenerate bindings for connect deny **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.9...v0.5.10 - Oct 15, 2026 0.5.9### Features - **firewall**: Add ray firewall test to explain a packet's verdict ### Bug Fixes - **tun**: Keep forwarding after an unreadable TUN packet (#238) - **deps**: Update iroh fork to 1.3 - **dns**: Disable DNS over TLS on the TUN link (#234)
More…
- **mesh**: Stop lost fragments from blocking later packets (#236) - **mesh**: Stop lost fragments from blocking later packets - Allow unjoined aliases in apply target exclusions - **macos**: Confirm file replacement and show download progress - **config**: Prevent saved network name collisions (#235) - **config**: Keep suffixed network names within 63 characters - **config**: Stop create and join from overwriting a saved network - **dns**: Disable DNS over TLS on the TUN link - **firewall**: Set the probe packet's transport offset - **membership**: Deliver ray leave to coordinators with no live link (#230) - Reuse macOS VPN sessions during polling - Restore the root systemd unit so the daemon starts and manages DNS - Drop unreachable code in check_root on Windows - Use the parsed transport offset past IPv6 extension headers - Report SSH exit signals by their RFC 4254 names - Show transfer progress and state in ray files - Match fresh-install defaults and lock pending-join writes - Remove a peer's network and route in one locked pass - Close the mDNS window opened during a stop - Use the same 5-minute mDNS window on Android - Align mDNS windows to the clock so nodes listen together - Run mDNS discovery in short windows to save battery - Reduce standalone QUIC send queue - **membership**: Deliver ray leave to coordinators with no live link - Fix - Log the full error chain for failed mesh dials - Handle receive credentials on Windows - Allow retrying interrupted file downloads - Allow SSH to the local mesh address - Use published release notes for macOS appcasts - Sync macOS app versions during cargo release - Remove release version check and retry CI image builds ### Refactor - Derive enum strings with strum - Simplify code across modules ### Documentation - Correct the ray files changelog entry ### Testing - **forward**: Cover the mock-TUN data path (#232) - **forward**: Cover the mock-TUN data path ### Miscellaneous - Release - Release - Merge master into DNS-over-TLS fix - Sync upstream CI fixes for Tier 1 tests - Sync upstream for Tier 1 review ### Other - Use safe - Allow non-root - Harden systemd service **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.8...v0.5.9 - Oct 4, 2026 0.5.8### Features - Add QUIC engine selectors to apps - Add FQ-CoDel flow scheduling flavor - Default to loss-tolerant QUIC congestion control - Add per-peer file auto-accept and macOS controls ### Bug Fixes
More…
- Match firewall rules across shared peer networks - Normalize generated Apple binding newlines - Finalize 0.5.7 release metadata ### Refactor - Simplify QUIC forwarding and fix queue metrics - Separate QUIC engines from congestion control - Simplify FQ-CoDel sender startup - Use Tokio mpsc for FQ-CoDel ingress - Use async-channel for FQ-CoDel ingress - Enqueue FQ-CoDel packets directly ### CI - **e2e**: Run Docker end-to-end tests (#215) ### Miscellaneous - Release - Release **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.7...v0.5.8 - Oct 3, 2026 0.5.7### Features - Add a quic-congestion toggle to the macOS app - Add the quic-congestion setting (#222) - Resolve contact identities with ray identityof - **android**: Clarify connection state and add in-app file transfers - **android**: Add in-app file sending and transfer results - **android**: Improve connection clarity and hide IPv6 addresses - Keep Android file relay reachable while idle
More…
- Add top-level mesh SSH commands - Make mesh SSH port configurable - Show macOS receive progress and reveal saved files ### Bug Fixes - Bridge IPv4 services outside the macOS tunnel extension - Keep bridge helper checks platform-specific - Run macOS IPv4 bridging outside the tunnel extension - Enable IPv4 bridge by default and start it in macOS app mode - Notify peers when the last coordinator nukes a network - Publish coordinator departure and accept signed destruction - Reserve network destruction for the last coordinator - Destroy networks across coordinators - **android**: Preserve firewall peers and limit transfer cancellation - **android**: Preserve foreground state and keep network sheets usable - Register macOS Bash completions in current shell - Kick removed hosts during apply - Copy peer domains and prevent ping wakeups on Android - Preserve paired status offline and enable macOS completion - Restore Android mesh notification after dismissal - Reconcile apply membership with machine state ### Performance - Make the QUIC congestion controller selectable (#221) - Remove packet forwarding bottlenecks (#220) - Increase QUIC datagram send buffer ### Refactor - **apple**: Prepare status fields before construction - **apple**: Group node status fields - Share macOS helper retry handling - Simplify network destruction handling - **android**: Simplify lifecycle and UI code ### Documentation - Keep macOS mesh listeners outside NetworkExtension - Keep Android release notes clear of concurrent additions - Record 0.5.6 release notes ### Miscellaneous - Release - Release - Merge master into Android UI branch - Merge latest master into network nuke branch - Merge latest master into network nuke branch - Merge master into network nuke branch - Merge master into network nuke branch **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.6...v0.5.7 - Oct 1, 2026 0.5.6### Features - Show macOS version and offer update restart ### Bug Fixes - Reuse sshd host key when config inspection fails - **macos**: Update app version to 0.5.5 - Replace existing macOS release assets on rerun
More…
- Bound serialized mesh dials - Sign Sparkle helpers for notarized macOS releases - Highlight RTT relative to peer history (#214) - Highlight RTT relative to peer history ### Documentation - Record v0.5.5 release ### Miscellaneous - Release - Release - Merge upstream master **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.5...v0.5.6 - Sep 29, 2026 0.5.5### Features - Sync paired networks after reconnect - Show network mode in status - **apply**: Support target exclusion selectors - Add automatic updates to macOS app - **cli**: Resolve peer identities across all networks ### Bug Fixes
More…
- **macos**: Retry VPN after network returns - **macos**: Remove redundant Sparkle embed phase - **apply**: Skip local machine when applying SSH grants - **firewall**: Allow mesh SSH through host firewalls (#207) - **firewall**: Allow mesh SSH through host firewalls - **android**: Sign release APKs with a permanent key ### Refactor - Use pairing message variants for network sync - Reuse pairing protocol for network sync - Use typed output across CLI - Use typed output for CLI lists - Centralize CLI JSON output ### Documentation - Update changelog through v0.5.4 ### Miscellaneous - Release - Release - Track Sparkle package resolution - Merge master updates - Merge master into macOS app updates **Full Changelog**: https://github.com/rayfish/rayfish/compare/v0.5.4...v0.5.5 **macOS DMGs:** Built from master commit b3675d50 after the v0.5.5 tag to fix Sparkle helper signing. Apple accepted the app and DMG notarization submissions for both architectures.