NemoClaw v0.0.129 is out!!

[jyaunches]

NemoClaw v0.0.129 improves OpenClaw restart readiness and recovery across local gateways. Restarted gateways load updated native plugins, and recovery follows the sandbox's recorded runtime provider.

Configuration export now preserves effective agent settings and can describe supported local inference services. Hermes and LangChain Deep Agents Code installers can also defer onboarding when NVIDIA hosted inference credentials are absent.

Host-readiness reporting now identifies Linux distribution qualification, and Windows on Arm GPU verification checks every reported device. Documentation and CI updates accompany these changes.

OpenClaw and sandbox recovery

• #11911 reloads native plugins from a fresh module graph during gateway restart, so plugin updates take effect without reusing cached modules.
• #12256 preserves the SQLite temporary-directory environment for standalone memory commands. Restart recovery waits for the gateway to accept work before restoring forwards.
• #11945 routes sandbox status and recovery through the owning gateway registry and gives recovery guidance for the recorded runtime provider.
• #12075 reserves retained host-local inference authority before resumed sandbox creation, so conflicting ownership produces a diagnostic before creating a runtime.

Configuration export and inference

• #12161 preserves effective OpenClaw and Hermes settings when omitted v1alpha1 values would change existing behavior.
• #12054 adds local inference service templates for supported OpenClaw exports using attached Ollama with a managed proxy or the fixed managed vLLM profile. Replace each template's image: null with an official, immutable, compatible v1 runtime image before v1 parsing or planning.
• #11864 stops NVIDIA Endpoints onboarding from selecting the retired MiniMax M3 model. Interactive onboarding asks for a replacement; noninteractive onboarding uses an eligible configured model or agent default. OpenRouter selection remains separate.
• #10245 forwards the existing Ollama proxy bind-probe override to the spawned process, so the child honors the operator's selected setting.

Installation and host readiness

• #10341 extends deferred onboarding to Hermes and LangChain Deep Agents Code. With the opt-in flag and no NVIDIA hosted inference credentials, installation finishes without creating a provider, sandbox, or completed onboarding state. Complete setup later through the installed agent's onboarding command.
• #11780 reports the Linux distribution and release and warns when OS qualification is unavailable or inconclusive.
• #12212 verifies every reported Windows on Arm GPU and aggregate capacity. This does not qualify generic multi-GPU hosts for larger automatic Ollama model selection.

Documentation and contributor workflows

• #12247 updates the release entry and guides for configuration export, deferred onboarding, gateway recovery, host readiness, and WSL GPU verification.
• #12211 selects an eligible PR source repository before publication, avoiding fork submissions that cannot run required same-repository validation.
• #12258 renews immutable Podman toolchain artifacts so qualification can use the pinned toolchain.
• #12265 enforces the immutable Podman staging boundary, making CI validate the reviewed toolchain inputs.
• #12240 updates the trusted bootstrap diagnostics template so CI recognizes the reviewed diagnostic structure.

Compare v0.0.128 with v0.0.129.