NemoClaw v0.0.40 #3466
ericksoa
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 v0.0.40 is a release about the quiet machinery that lets a user trust the first minutes of an install. We spent this cycle making the small signals more honest: a gateway that is reported healthy should actually be reachable, a failed dashboard should say so with a non-zero status, long URLs should fit inside the box that contains them, and setup should explain the tools and permissions it needs before the user is left to infer them. These are modest changes individually, but together they move NemoClaw closer to a system whose behavior can be observed, reproduced, and repaired.
The largest body of work is around messaging and the OpenShell runtime boundary. v0.0.40 lands the native WebSocket credential rewrite path, preserves disabled channel state across rebuilds, improves Hermes next-step output, and adds macOS VM rootfs compatibility. This is meaningful progress toward a cleaner messaging architecture, but it is not the end of the voyage. Messaging still needs more stabilization and real-world testing across Slack, Discord, Telegram, and the OpenShell gateway paths before we should treat it as settled.
We also made the release more practical for Spark, UAT, and local-inference validation. The release adds compatibility metadata for OpenClaw, a running-vLLM option in onboarding, a Qwen3.6 35B starter model for large-memory Ollama hosts, documentation for local tool-call reliability, and fixes around custom gateway ports, GPU sandbox policy, Ollama loopback repair, and OpenClaw baseline recovery. In the large space of machines, networks, and user workflows, v0.0.40 is one more calibrated instrument: better at telling us what is working, clearer when it is not, and better prepared for the next round of UAT evidence.
Messaging, Hermes, and OpenShell
Spark, UAT, and Local Inference
nemoclaw inference getcommand.qwen3.6:35bto Ollama starter models on large-memory hosts./procwildcard in GPU sandbox policy.Onboarding, Gateway, Docker, and Install Reliability
openclaw.jsonon sandbox restart.zstdbefore the Ollama Linux installer and clarify sudo prompts.Security, Policy, and Shields
/usr/bin/ghfrom GitHub preset binaries.CLI, UX, and Observability
Testing, CI, Refactors, and Release Docs
spawnSynctest invocations.Beta Was this translation helpful? Give feedback.
All reactions