NemoClaw v0.0.57 is out #4691
cv
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
NemoClaw v0.0.57 focuses on making sandbox state less surprising when hosts reboot, containers churn, policies change, and gateways drift. Status and connect paths are more careful about preserving registry state, Docker and gateway failure layers are more explicit, and shields-up hardening now keeps runtime paths usable while still locking down sensitive state.
This release also sharpens the operator loop for OpenClaw, Hermes, Slack, WhatsApp, Jira, and local inference. There are new host-side
sessionsandagentscommand groups, better validation before messaging channels are enabled, more visible WhatsApp and policy diagnostics, and a DGX Spark managed-vLLM profile that moves to NVIDIA's Qwen3.6-35B-A3B-NVFP4 setup with clearer download and readiness progress.The maintenance work is broad too: per-agent documentation variants are now in place for OpenClaw and Hermes, newcomer contribution guidance is clearer, nightly E2E results can surface in Slack, and the test suite has more focused probes for strict tool calls, Docker recovery, sessions and agents, link checking, and other release-sensitive paths.
Sandbox, Security, and Network Stability
claude-codepreset, so broad permissive mode no longer opens those endpoints unless the operator opts in.agents/*/sessionswritable, protecting secret-bearing state without breaking first launch.sandbox status, so text and JSON status report the lower-level cause before probing inference.statusnon-destructive when a registered sandbox is temporarily missing from the live gateway, preserving post-reboot registry state for later reconciliation.nemoclaw-startbehavior under CAP_DAC_OVERRIDE drops, keeping hardened startup paths working.Onboarding and Recovery
NEMOCLAW_AUTO_FIX_FIREWALL=1path that can apply the UFW sandbox-bridge rule after the same reachability failure the preflight already explains.exitandquitat numbered onboarding menus, so the wizard cancels instead of silently accepting the default choice.Messaging and OpenClaw Runtime Controls
channels addatomic when a policy preset is missing, so failed channel setup does not leave partial policy or provider state behind.channels addenables Slack, preventing invalid credentials from being saved as an active channel.tmuxfor the OpenClaw bundled tmux-session flow, keeping the bundled session workflow available inside the sandbox.nemoclaw <name> sessionsandnemoclaw <name> agentssubcommand groups, giving operators sandbox-first controls for listing, resetting, deleting sessions, and adding or deleting OpenClaw agents.NEMOCLAW_EXTRA_AGENTS_JSON, allowing operators to bake secondary OpenClaw agents intoopenclaw.jsonat onboard time while keepingmainas the default agent.Inference, Policies, and Provider Behavior
uvin the PyPI policy preset, matching the tooling operators use for Python dependency workflows./v1/modelsreadiness with bounded probes and failure log tails.Skills, Docs, and Contributor Experience
CONTRIBUTING.md, including contribution types, where to start, review expectations, and DCO sign-off guidance.internal:*command family, making internal command behavior discoverable for maintainers without promoting it as normal user workflow.CI and Release Confidence
Thank you
Thank you to external contributors @1PoPTRoN for #3785 and #4075, @latenighthackathon for #4169, and @arnavnagzirkar for #4620.
Full compare: v0.0.56...v0.0.57
Beta Was this translation helpful? Give feedback.
All reactions