These notes reconcile compatibility reports and their follow-up comments as of 15 November 2026. They supplement the tested-vacuums table, which retains earlier results. Observations are attributed to their reports; they are not independent hardware tests by the documentation author.
Keep firmware, certificate chain, client, and server build together when comparing results. A successful MQTT connection does not establish working controls, maps, camera, or every Home Assistant entity. A later success with another chain or firmware does not invalidate an earlier failure. An open PR's hardware report does not mean its proposed implementation is in a release.
On server v1.2.1, Docker Compose with the embedded broker and no reverse proxy, Actalis RSA-2048 worked using TLS 1.0. The reported chain was Actalis Domain Validated TLS Server RSA CA 2025 → Actalis TLS Server RSA Root CA 2025 → Actalis Authentication Root CA. Four GUI pairing cycles recovered the key and connected. HA status, controls, both floor maps and room targeting worked. Water-box-attached and mop-mode entities available through the cloud were absent locally; mop intensity remained available. The official app was not tested after onboarding. #191
A second unit on the same firmware rejected a Let's Encrypt ECDSA chain YE1 → Root YE → ISRG Root X2 → ISRG Root X1, but accepted a provided Actalis RSA-2048 certificate issued through Certbot. Both reporters found that their Actalis 90-day EAB credentials had issuance quota while their 1-year credentials did not. One reporter changed a manually entered Bucharest POSIX timezone to auto-filled Berlin settings before traffic started; the report does not isolate the timezone from a malformed CST string as the cause. Follow-up
On v1.2.1, 02.33.88 worked with Actalis after Let's Encrypt and ZeroSSL
attempts failed. Docker ran in a Proxmox LXC with Caddy managing certificates.
HA and LocalRock controls/maps worked after three or four pairing cycles. Camera
setup remained unresolved and an authenticated external Mosquitto broker could
not be used; the reporter used the embedded broker instead.
#231
The newer 02.62.46 accepted ZeroSSL through the v1.2.1 HA add-on's ACME
flow. LocalRock onboarding and HA/LocalRock controls worked with Pi-hole DNS.
The official app did not work in that setup. This is firmware-specific evidence,
not a reason to overwrite the older certificate failures.
#236
Actalis RSA-2048, issued by Actalis Domain Validated TLS Server RSA CA 2025,
worked on a custom scheduling build in Docker on Linux Mint. Nginx terminated
HTTPS on 443 and MQTT TLS on 8881, forwarding to external_tls listeners. After
a firewall correction, three LocalRock pairing attempts completed onboarding.
Later tests confirmed official-iOS-app video, Call and remote driving with an
existing remote-view PIN. TURN had to be enabled, but the observed bulk media
was direct across the LAN/VLANs; that test did not prove forced-relay operation.
#215
A separate HA add-on/HAProxy setup on the same firmware used automatically provisioned ZeroSSL base-domain/wildcard certificates. With WAN blocked, the reporter observed camera bootstrap, TURN allocation and relayed SD video using published UDP 3478 and 49160–49179. The precise server version was not given in that follow-up. See the camera results for the separate Let's Encrypt and forced-relay test from PR #185. ZeroSSL follow-up
The S7 Pro Ultra (a62) Actalis RSA-2048 result and Let's Encrypt R3/E1 rejection are already in the table. Its firmware was not reported. The reporter observed the connection closing after the TLS handshake but before HTTP/MQTT application traffic with the rejected certificate. That symptom warrants checking the exact chain and hostname; it does not prove a universally incompatible CA. PR #65
QX Revo Ultra 2 (a298), 02.15.44, was reported on v1.2.1, unmodified Docker
Compose, with Actalis: maps and cleaning/docking acknowledgements worked, and
the reporter observed five days of local operation. Recovery used header samples
while query samples stayed at zero. Idle reconnects were reported on this setup;
do not use that observation to dismiss command failures or stalled TLS listeners
on other models. The table keeps Actalis as the only established issuer for this
report. Proposed camera tooling in the PR is separate from that compatibility result.
PR #194
The reported working setup used v1.2.0, Podman, LocalRock cloud-import
onboarding and HA controls, with both Let's Encrypt and ZeroSSL. Caddy supplied
certificate files but did not proxy the server. Earlier v1.1.0 attempts failed.
Increasing the container's constrained CPU/memory allocation helped; public-key
recovery can be expensive. The robot attempted DNS directly to 8.8.8.8 until
WAN access was blocked, then used DHCP-provided local DNS. See
network troubleshooting.
#114
02.27.04 worked on server build 94bffa7, Docker/Proxmox behind Traefik with
Let's Encrypt YR2 → Root YR → ISRG Root X1. The 02.29.56 follow-ups used
v1.2.1, both direct TLS with an ECDSA P-384 chain
YE1 → Root YE → ISRG Root X2 → ISRG Root X1 and Traefik with a YR2 chain.
HA maps, status and controls worked. One follow-up recovered the key from header
samples in about two minutes; WAN blocking before onboarding was needed in those
follow-ups. EU cloud import incorrectly selected the US region for one reporter;
explicit base_url = https://euiot.roborock.com returned the inventory.
#139 and follow-ups
Remaining model/client limitations: HA displays Vacuum then Mop as unknown;
set_wash_towel_mode was rejected in one setup. Another reporter found
set_custom_mode accepted only 101, while combined set_clean_motor_mode
accepted 102/103/104/105/106/110; mop-only used fan power 105. Mop modes 300 and
303 worked, 301 did not. get_recover_maps returned unknown_method. HA map
editing worked inside start_edit_map / end_edit_map. These are observations
on 02.29.56, not universal command recipes for other firmware.
Command details
On the v1.2.1 HA add-on with ZeroSSL, the initial report had successful onboarding and telemetry but no replies to commands. The reporter later confirmed working operation after deleting/recreating server state and performing three fresh pairing cycles. The suspected link to an intervening firmware update was not established. Record this as a resolved setup report, not proof of a general firmware-update fix or a recommendation to delete server data as a first step. #217 resolution
On v1.2.0, Docker behind Nginx/SWAG with Let's Encrypt, HA worked. The reporter manually backed up maps to the robot before migration; LocalRock did not display carpet surfaces or custom room names. An admin-password UI change was also requested. Recording the compatibility result does not implement those UI improvements. #110
QV 35A 02.49.24 worked on the v1.2.0 HA add-on, using ZeroSSL via Cloudflare
DNS-01, local dnsmasq DNS, and three script pairing cycles. HA and LocalRock
commands were confirmed. The requested external_tls option remains unavailable
in the HA add-on; that mode is for standalone/Docker deployments.
#119
QX Revo Plus 02.27.40, reported model identifier a168, worked on v1.2.1
Docker with a ZeroSSL/Sectigo/USERTrust chain. HA connected and a cleaning
routine ran; ACME timeouts were reported during setup. Keep this retail name and
firmware distinct from other QX Revo Plus entries.
#248
S5V 02.33.88 onboarded after four to six cycles on v1.2.0 Docker with
Let's Encrypt, but map saving was disabled. A follow-up enabled lab_status
through HA's vacuum.send_command with set_lab_status and
[{"lab_status": 1}]. A later LocalRock zone-cleaning coordinate problem remains
unresolved; the workaround does not establish that zone cleaning is fixed.
#148
QRevo 5AE 02.33.88 was reported working through HA and the Roborock app using
ZeroSSL, with the same map-saving workaround. The server version was entered
as v1.1.2 in the report and has not been verified; the model identifier was not
given. Do not infer an a170 identifier from the shared workaround.
#218
On v1.2.1-1-gc177a76, Let's Encrypt RSA-2048 with YR1 → ISRG Root YR → ISRG Root X1 worked. HA retrieved status and maps after re-authentication. Pi-hole initially ignored queries from the routed IoT subnet; allowing that subnet's DNS requests resolved it. The onboarding PC also needed a hosts-file entry and routing adjustment while attached to the robot hotspot. HA's local re-authentication asks for the configured PIN under the label "verification code"; it sends no email. Stale MQTT credentials can be logged until this step completes. See HA setup. #156
The v1.2.1 Let's Encrypt/Caddy layer-4 setup had working HA but misleading missing Region/NC Prepare messages in the admin page. A separate direct-TLS Docker setup used a ZeroSSL ECDSA P-256 wildcard and completed NC Prepare after three GUI cycles, with HA/LocalRock maps and controls working. Both blocked WAN traffic. These are separate profiles, not evidence that the first status problem was fixed. #171, ZeroSSL result
The v1.2.1 HA add-on with provided Let's Encrypt certificates (HA Let's
Encrypt add-on, Porkbun DNS-01) supported cloud-free initial onboarding, maps,
controls and room cleaning. HA reported firmware 02.33.88, while device OTA
telemetry reported 02.37.80; the report does not resolve that discrepancy.
On server 1.2.0, control failed with rrcheck access denied after about 48 hours;
1.2.1 worked after a fresh /f/c exchange. LocalRock room splitting worked but
room renaming did not. The first map saved after a completed run returned to dock.
#192
On 02.40.54, a ProX carrying a294 worked with v1.2.1, rootless
Podman Compose and Let's Encrypt certificates copied from Nginx Proxy Manager.
HA and LocalRock controls were confirmed. The report's public-port workaround is
not a prerequisite for compatibility; use local DNS and reachable LAN endpoints
as described in Installation.
#242
v1.2.1 Docker, with Let's Encrypt at a reverse proxy, worked without reported functional limitations. Three attempts collected two samples and a fourth completed pairing; repeated cycles alone were not evidence of incompatibility. #219
Edge 2 Pro a226, 02.33.88 worked on the v1.2.1 HA add-on with ZeroSSL.
HA and LocalRock connected after fixing VLAN MQTT access and inconsistent DNS
answers from a Technitium forwarding/wildcard setup. The unpatched official app
could not log in on GrapheneOS; the reporter used fresh local onboarding instead.
Restarting LocalRock made the new vacuum appear. Resuming interrupted onboarding
was requested as a usability improvement.
#237
The a226, 02.43.62 reports are explicitly
custom-build results.
In #223, the cloud called the product Curv 2 Pro (numeric product ID 199).
Let's Encrypt YR1 → Root YR → ISRG Root X1 worked on Kubernetes with a server at
6030080 plus product/plugin patches. Initial blind onboarding and HA control
worked. Camera activation subsequently used the official cloud app; re-onboarding
after cloud import hit the stale identity/credentials problem in #222. The
official iOS app needed the #221 plugin changes and the MITM/WireGuard tunnel
disabled after login. For go2rtc two-way audio, the microphone-enabled stream
had to establish the producer first; switching from an existing listen-only
producer did not enable the microphone.
#223
In #239, a custom ARM64 build at 02a1770 on a Pi 5 used an apex-host opt-in,
camera-bootstrap patches, a custom L01 bridge and modified go2rtc at 990f4de.
Let's Encrypt YE2 → Root YE → ISRG Root X2 → ISRG Root X1 covered the apex.
Video/audio worked with WAN blocked and resumed after a host reboot (about
8.5 minutes to reconnect). Camera access/PIN had been enabled in the cloud app.
No cleaning, maps, HA or post-migration official-app control was tested.
#239
Curv 2 Pro 02.45.78 also has a separate v1.2.1 success report with a
provided Let's Encrypt RSA-4096 wildcard, YR1 → ISRG Root YR, exported from
Traefik. Two onboarding cycles were needed, with about 20 minutes waiting after
the first. The report does not identify the model code or explicitly validate
camera/audio; do not carry the custom a226 camera results forward to this firmware.
#233
Edge 3 Pro, a296 as reported, 02.32.54, worked through LocalRock on the
v1.2.1 HA add-on. Three or four cycles recovered the key and connected even
though query samples stayed at zero. The issuer was entered as unknown;
cloudflare_acme and zerossl settings alone do not verify the served chain.
#246
Edge 2 Flow, a328 as reported, 02.33.88, worked on the v1.2.1 HA add-on
with Let's Encrypt issued by the separate HA add-on using DNS-01. It had never
been connected to the official app/cloud, and mopping worked. Camera remains
unverified: the reporter could not set the PIN and received a parameter error.
Do not copy camera success from the differently named Curv 2 Flow.
#256 and follow-ups
Firmware 02.07.60 onboarded using the HA add-on and ZeroSSL via Cloudflare
DNS-01. Four signature samples across repeated cycles recovered the key; the
official HA integration connected using Manual region. pfSense supplied local
DNS, there was no reverse proxy, and no Internet port forwarding was used.
The entered server version 18.2 is unverified and must not be treated as a
local-server release number. The report confirms onboarding/HA connection rather
than an exhaustive feature test.
#216
The China G30 a146, 02.43.88, on v1.2.1-3-g5b99e93, onboards using
ZeroSSL but rejects control with rrcheck access denied; a China VPN did not
resolve that setup. Separate a143 and a96 failures are listed in
Known failures.
#161
For China P20 a184, factory 02.22.56, #228 reports a startup /f/c call to
a hardcoded cloud hostname as well as renewal through /devices/{did}/sh.
A TLS failure before HTTP can therefore leave no /f/c entry in the local
server's logs. #229 reports camera hostname validation after the handshake.
#230 reports commands, map editing, schedules, camera and daily renewal checks
working in a custom intercepted setup using compromised-intermediate certificates.
That setup was never online and never updated. It establishes neither a standard
owned-domain certificate path nor a fix for already region-locked units.
#228,
#229,
#230
Q7 TF+ onboarding failure in #48 remains distinct from newer experimental sc05 results. PR #96 reports B01 bootstrap/inbound decoding on Q7 M5+, while its TF+ follow-up uses cloud pairing and custom interception. Neither makes all Q7/Q10 models supported. The PR discussion also contains a Q7 L5 bootloop report during OTA probing, not successful local onboarding of that model. #48, PR #96
Recording a result does not implement the requested fixes. In particular, the admin-password UI (#110), HA external TLS (#119), Saros 20 Neo import/mode issues (#139), zone coordinates (#148), onboarding status (#171), missing HA entities (#191), room renaming (#192), and Edge 2 Flow camera/PIN setup (#256) remain separate from the compatibility rows. Custom plugin and identity workarounds (#221/#222 referenced by #223) are not stock-release guarantees.