> This page is for version Latest (default). > For other versions, use one of these documentation indexes: > - Latest (default): https://docs.nvidia.com/nemoclaw/latest/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/nemoclaw/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemoclaw/_mcp/server. ## v0.0.106 NemoClaw v0.0.106 converges readiness across lifecycle commands, upgrades the managed OpenShell runtime to v0.0.101, and adds bounded DGX Spark vLLM choices. It strengthens sandbox and Hermes recovery, local inference diagnostics, endpoint policy authority, and scoped uninstall. It also improves update safety, image defaults, and installation failure reporting. * `nemoclaw host probe --json`, onboarding, resume, and rebuild now use one credential-free readiness report and admission policy. Blocking findings stop lifecycle effects before NemoClaw changes credentials, images, gateways, policies, or sandboxes. Gateway readiness accepts OpenShell v0.0.101 `Server:` endpoint output and target-bound process tags only when trusted listener evidence matches the configured gateway. It preserves the selected gateway's stale state when a gateway-scoped OpenShell status check cannot connect, so onboarding can reconcile the registered gateway. For more information, refer to [System Readiness](/user-guide/openclaw/reference/system-readiness) and the [NemoClaw CLI Commands Reference](/user-guide/openclaw/reference/commands). * The managed OpenShell runtime now uses v0.0.101 across installation, gateways, sandboxes, supervisors, and managed agent images. Existing installations retain the v0.0.99 fallback only for migration, and incompatible sandbox names stop the update before destructive work. For more information, refer to [Update Sandboxes](/user-guide/openclaw/manage-sandboxes/operate-sandboxes/update-sandboxes) and [Configure Runtime Identity](/user-guide/openclaw/reference/configure-runtime-identity). * DGX Spark Express now offers a fixed managed vLLM option, while `nvidia/Qwen3.6-35B-A3B-NVFP4` remains the default profile for one DGX Spark. The additional Muse Glimmer profile remains Experimental, runs on one DGX Spark, and excludes vision and DFlash. Dual DGX Spark route discovery accepts the `iproute2` `from` field, and Model Router allows a longer bounded cold start. For more information, refer to [Set Up vLLM](/user-guide/openclaw/inference/local-inference/set-up-vllm) and [Set Up vLLM on Two DGX Sparks](/user-guide/openclaw/inference/local-inference/set-up-vllm-on-two-dgx-sparks). * Managed llama.cpp onboarding now cleans up rejected fixed-port state and distinguishes blocked Docker bridge access from a healthy host and runtime. By default, the local Ollama proxy refuses a backend that listens beyond loopback, and onboarding retries while a selected model becomes tool-call responsive. For more information, refer to [Set Up llama.cpp](/user-guide/openclaw/inference/local-inference/set-up-llama-cpp) and [Use Ollama](/user-guide/openclaw/inference/local-inference/set-up-ollama). * Gemini 3.x managed routes now load compatibility metadata so OpenClaw preserves thought signatures across tool calls. `NEMOCLAW_AGENT_TIMEOUT` now controls both the agent and provider request deadlines, while `status` waits for route convergence after automatic gateway recovery. For more information, refer to [Use Google Gemini](/user-guide/openclaw/inference/hosted-inference/use-google-gemini), [Configure Inference Timeouts](/user-guide/openclaw/inference/manage-inference/configure-inference-timeouts), and the [NemoClaw CLI Commands Reference](/user-guide/openclaw/reference/commands). * Trusted-private endpoints now keep the normalized hostname and address pins bound from admission through policy replay. Managed MCP rejects mixed public and private address answers before mutation, and managed startup rejects a symlinked corporate CA before reading it. The OpenClaw image also defaults to the non-root `sandbox` user, while OpenClaw v2026.7.1 gateway self-dialback stays on loopback inside OpenShell. For more information, refer to [Meet Custom Endpoint Security Requirements](/user-guide/openclaw/inference/custom-endpoints/custom-endpoint-security), [Configure Corporate CA Trust](/user-guide/openclaw/security/configure-corporate-ca-trust), and [Security Best Practices](/user-guide/openclaw/security/best-practices). * Hermes rebuilds restart the gateway after workspace restore and before managed MCP restoration. The cron gate remains active until the replacement process passes MCP and health verification. Rebuild also settles an expired Shields timer before success, while macOS profile migration uses a native atomic no-clobber rename. For more information, refer to [Recover and Rebuild Sandboxes](/user-guide/hermes/manage-sandboxes/operate-sandboxes/recover-and-rebuild-sandboxes) and [Security Best Practices](/user-guide/hermes/security/best-practices). * Sandbox recreation and live creation now wait for OpenShell to release ownership before Docker cutover. Starting a stopped OpenClaw or Hermes sandbox waits through transient supervisor absence and managed-container restart transitions before it repeats recovery, readiness, and port-forward checks. A Shields up restart also keeps `/sandbox` at `root:sandbox` mode `1775` while the explicit process policy selects the sandbox workload identity. Legacy gateway updates reuse validated prepared backups, while rebuilds restore reviewed build contexts and retire only proven obsolete images. For more information, refer to [Recover and Rebuild Sandboxes](/user-guide/openclaw/manage-sandboxes/operate-sandboxes/recover-and-rebuild-sandboxes), [Update Sandboxes](/user-guide/openclaw/manage-sandboxes/operate-sandboxes/update-sandboxes), and [Troubleshooting](/user-guide/openclaw/reference/troubleshooting). * `nemoclaw update --fresh` now refuses older or incomparable maintained versions unless the operator supplies `--allow-downgrade`. Scoped uninstall isolates gateway shutdown by namespace and preserves checkpointed state for retry after a partial failure. Installer preflight preserves managed `fuse-overlayfs` storage remediation and accepts valid two- or three-component NVIDIA driver versions without changing the `580.65.06` minimum. Installer linking skips npm lifecycle scripts, launcher initialization failures produce redacted single-line errors, and LangChain Deep Agents Code images include `dos2unix`. For more information, refer to [Prerequisites](/user-guide/openclaw/get-started/prerequisites), [Set Up llama.cpp](/user-guide/openclaw/inference/local-inference/set-up-llama-cpp), the [NemoClaw CLI Commands Reference](/user-guide/openclaw/reference/commands), and [Uninstall NemoClaw](/user-guide/openclaw/manage-sandboxes/operate-sandboxes/uninstall-nemoclaw).