v0.76.1
browser-use/video-usev0.76.1Sep 12, 2026by github-actions[bot]
Upgrading?
AI Summary
This release focuses on diagnostics and stability improvements for the Mesh-LLM system. It adds visibility into GPU memory capacity and capture methods, introduces diagnostics for direct-connect paths, and fixes several issues regarding tool definitions and public address advertising.
Key Highlights
- Added `gpu_name_source` to display how GPU names were captured.
- Added diagnostics for direct-connect advertisement and peer paths.
- Fixed rejection of malformed tool definitions.
- Fixed advertising externally observed public address.
- Fixed keeping direct GGUF split loads local.
New Features
- Itemized the advertised capacity behind `vram_bytes`.
- Surface the advertised memory breakdown.
- Diagnose direct-connect advertisement and peer paths.
Full Release Notes
## [0.76.1] - 2026-09-12 ### Added * Itemize the advertised capacity behind vram_bytes by @Virgile-pct in https://github.com/Mesh-LLM/mesh-llm/pull/1673 * Gpu_name_source beside gpu_name — how the value was captured, display unchanged by @StevenMih in https://github.com/Mesh-LLM/mesh-llm/pull/1679 * Diagnose direct-connect advertisement and peer paths by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1803 * Surface the advertised memory breakdown by @Virgile-pct in https://github.com/Mesh-LLM/mesh-llm/pull/1754 ### Fixed * Roll up the 0.76.1 bugfix branch into main by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1789 * Reject malformed tool definitions by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1761 * Link Apple OpenMP runtime by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1766 * Advertise externally observed public address by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1774 * Reject self-identity joins and add node key override by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1775 * Keep direct GGUF split loads local by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1776 * Report client mode from --client --auto console state by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1787 * Match NVIDIA driver devices by identity instead of index by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1760 * Honor native runtime selection across startup paths by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1759 * Keep the OpenAI listener alive across accept errors by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1762 * Support Linux CUDA on driver-only hosts by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1793 * Serialize materialized cache writes on Windows by @Bortlesboat in https://github.com/Mesh-LLM/mesh-llm/pull/1685 * Use served model IDs in ready commands by @Bortlesboat in https://github.com/Mesh-LLM/mesh-llm/pull/1688 * Retain device placement in native log summaries by @Bortlesboat in https://github.com/Mesh-LLM/mesh-llm/pull/1689 ### Internal <details> <summary>CI, build, test, and repository work with no user-facing behavior change (16 changes)</summary> #### CI and release engineering * Declare and enforce the Linux glibc floor for shipped artifacts by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1768 * Stamp the real source revision into release attestations by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1765 * Guard Linux product smoke selector drift by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1733 * Preflight packaging dispatch credentials by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1734 * Publish SDK smoke cache policy by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1735 * Simplify llama canary state machine by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1727 * Stop duplicate release tag runs by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1731 * Bound llama canary repair stalls by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1764 * Harden llama canary runner and stage loading by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1781 * Fly console image build: pin pnpm and copy the whole crates tree by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1779 * Reject backward llama.cpp pin moves by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1783 * Simplify llama canary agent harness by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1813 * Regroup release notes from conventional commits by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1758 * Let llama canary verification reach certification by @i386 in https://github.com/Mesh-LLM/mesh-llm/pull/1819 #### Build and dependencies * Probe lld and fall back to the platform linker when it cannot link by @ndizazzo in https://github.com/Mesh-LLM/mesh-llm/pull/1771 #### Tests * Make the read-only filesystem warning test portable by @Virgile-pct in https://github.com/Mesh-LLM/mesh-llm/pull/1658 </details> **Full Changelog**: https://github.com/Mesh-LLM/mesh-llm/compare/v0.76.0...v0.76.1
Related Repositories
public-apis/public-apis★ 491,126
A collective list of free APIs
NousResearch/hermes-agent★ 255,895
The agent that grows with you
yt-dlp/yt-dlp★ 199,990
A feature-rich command-line audio/video downloader
microsoft/markitdown★ 191,478
Python tool for converting files and office documents to Markdown.
Significant-Gravitas/AutoGPT★ 187,699
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.