NemoClaw v0.0.116 is out!

[apurvvkumaria]

NemoClaw v0.0.116 strengthens the boundaries where sandbox identity, lifecycle authority, and cleanup meet. Retained sandboxes keep verified authority through retries and process restarts, uninstall protects recoverable state, and sandbox destruction can recover from demonstrably abandoned Shields timers.

Inference paths also recover more deliberately. OpenClaw validates compatible endpoints inside the sandbox without depending on messaging, Portable Hermes recovery resumes only its receipt-owned Ollama runtime, and Windows Subsystem for Linux (WSL) public pulls no longer depend on an available Docker Desktop credential helper.

This release also refreshes security-critical managed-image packages and preserves more user intent during onboarding. Experimental Hermes channel credentials are revision-scoped, web-search policy intent survives re-onboarding, and the documentation now covers each changed workflow in its owning task page.

Inference and Hermes Recovery

• OpenClaw compatible endpoint onboarding now validates inference.local from inside the sandbox even when no messaging channel is selected. inference set can reuse a recorded loopback no-auth endpoint, and reasoning-only responses receive one bounded larger-budget content attempt. These changes stop unreachable routes before final verification without rejecting a working reasoning model too early. #10458, #10531, and #10540.
• nemohermes <name> recover and nemohermes <name> connect --probe-only now recover only the published receipt-owned Portable Ollama runner. They verify an already-running runner without restarting it, resume the recorded stopped runner once, and require a final in-sandbox route proof. Failed recovery attempts to return the runner and sandbox registry to their prior state. The command reports when it cannot prove either restoration, which avoids replacing or taking authority over another runtime. #10505.

Sandbox Lifecycle and Data Safety

• Sandbox creation and reuse retain verified identity, policy, provider, credential, and lifecycle authority across retries and process restarts. Same-name replacements and authority drift stop mutation, while a transient non-Ready report does not block registration when durable sandbox identity still matches. This prevents a replacement sandbox from inheriting authority and avoids false onboarding failures during OpenShell re-registration. #10510 and #10512.
• Uninstall creates fresh snapshots for eligible registered non-Portable sandboxes before deletion unless you select --destroy-user-data. It also honors NEMOCLAW_OPENSHELL_GATEWAY_STATE_DIR and preserves abandoned Portable configuration that has no lifecycle receipt. These checks protect recoverable state and unrelated gateway state. Use --destroy-user-data only when you intend to skip snapshots and delete user data. #10231, #10550, and #10562.
• Sandbox destruction can recover an expired Shields auto-restore timer only when its process is gone or its recorded start identity no longer matches. Live or unverifiable timer authority remains fail-closed, and Docker must prove the sandbox container is absent before destruction skips pre-delete lockdown. This shortens recovery from abandoned timers without weakening active Shields protection. #10225.

Onboarding, Policy, and Messaging

• On WSL, public managed-image pulls, generated image builds, GPU probes, and local inference probe pulls use a temporary credential-free Docker configuration when the default Docker Desktop credential helper is unavailable. Managed local inference also waits for disposable Podman probes to disappear after one removal command, and llama.cpp pull failures return bounded diagnostics without raw pull output. These changes let headless onboarding continue while protecting host credentials and improving local-inference failure reports. #10470, #10554, and #10561.
• Re-onboarding or rebuilding an OpenClaw sandbox with a recorded Balanced or Open policy tier keeps its existing brave policy preset when web search is disabled or switched to Tavily Search. This preserves the recorded tier's Brave Search egress instead of silently narrowing policy during reuse. Restricted, unknown, unsupported, and custom-policy cases retain their existing reconciliation behavior. #10457.
• Experimental Hermes WeChat and Microsoft Teams channels now receive revision-scoped OpenShell credential placeholders under the environment names Hermes reads. Raw or identityless values are ignored, while malformed or channel-unbound aliases stop startup. This restores credential delivery without weakening channel identity checks. Both channels remain experimental. #10563.

Managed Images and Security Packages

• Hermes Portable images now select the reviewed multi-platform base that matches their OpenSSL 3.5.7 inventory. OpenClaw images install checksum-pinned libevent 2.1.13 on AMD64 and ARM64. These updates align public installs with reviewed package inventories and deliver the current libevent security fixes. #10532 and #10526.

Documentation and Release Surface

• User documentation now covers the release's inference validation, Portable Hermes recovery, policy reuse, WSL Docker fallback, messaging, and uninstall behavior. The release entry and owning task pages give operators one current procedure for each changed workflow. #10593 and #10530.