Meta releases open-weight Muse Spark Open for coding

#1 𝕏

AI at Meta released Muse Spark Open, an open-weight model built for coding and agentic work, with 816B total parameters, 42B active parameters, and a 1M-token context window. It is natively multimodal across text, image, video, and audio, and is described as excelling at long-horizon tasks, complex software engineering, and multi-step reasoning.

Also covered by: @AI at Meta, @clem 🤗, @clem 🤗

#2 𝕏

NVIDIA AI announced Nemotron 3.5 Bolt 24GB, which fits into 24GB of memory.

#3 𝕏

Cognition announced Devin Oncall, which automatically investigates alerts and bugs to help teams diagnose and resolve production incidents. Cognition says it already handles thousands of alerts every day for some of the world’s largest companies.

Also covered by: @Cognition

#4 𝕏

Aravind Srinivas announced that SPACE, the sandbox platform he says powers all of Perplexity Computer, is being open-sourced to support the ecosystem. He noted that serving millions of sandboxes under production-grade traffic requires substantial detail to scale agents for businesses and consumers.

#5 𝕏

OpenAI released GPT-Live-1.1 with new voices, better accents, and less “checking...” for Voice in ChatGPT, Codex, and dots, including voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese. It began rolling out to all users on the day of the post, with API availability planned for a later, unspecified date.

#6 📝 OpenAI News

Introducing the Rosalind Academic Plan - OpenAI announced the Rosalind Academic Plan on October 15, 2026, offering eligible academic research labs sponsored access to GPT‑Rosalind and tools for life sciences discovery.

Also covered by: @OpenAI

#7 🐙 anthropics/claude-plugins-official

anthropics/claude-plugins-official - **anthropics/claude-plugins-official** is Anthropic’s official directory for Claude Code plugins, covering both Anthropic-maintained and third-party options. Plugins can include MCP configurations, slash commands, agents, reusable skills, and documentation, and are installable through Claude Code’s plugin system. Worth exploring if you need packaged Claude Code extensions, but review each plugin carefully because Anthropic does not verify included software.

#8 🐙 obra/superpowers

obra/superpowers - **obra/superpowers**, shared by obra, is an MIT-licensed development methodology for coding agents. It includes composable, automatically triggered skills that guide requirements, specification approval, implementation planning, red/green TDD, and subagent-based task execution and review. Worth trying if you need a structured workflow for agent-led software development.

#9 🐙 pbakaus/impeccable

pbakaus/impeccable - **pbakaus/impeccable** provides design guidance for AI coding agents through 1 reusable skill, 24 commands, live browser iteration, and 59 deterministic detector rules. pbakaus released Impeccable CLI v4.6.0 on October 15, 2026, adding Node 18.19-and-later installer support, engine 0.1.15 detection fixes, and Oh My Pi targeting. Worth trying if you need structured design workflows and LLM-free frontend checks. Release notes

#10 🐙 browser-use/browser-use

browser-use/browser-use - **browser-use/browser-use** is an MIT-licensed open-source Python library for AI browser agents, with examples, tests, reusable skills, Browser Use Cloud, and a hosted agent API. Browser-use released version 0.13.13 on October 15, 2026, adding an OpenAI browser-tool driver with 18 default tools and six optional actions; it requires the `openai-tools` extra and OpenAI Python 3.31+. Worth trying if you need agents to control local, cloud, or borrowed browser sessions. Release notes

#11 🐙 DietrichGebert/ponytail

DietrichGebert/ponytail - **DietrichGebert/ponytail** is an MIT-licensed coding-agent prompt with reusable skills, agent adapters, docs and examples focused on code reuse and smaller changes. DietrichGebert released Ponytail v5.3.0 on 2026-10-16 at 07:52:12Z, adding a Claude Code status band and local cross-project task board; the band requires Claude Code 2.1.290 or later. Worth trying if you need local visibility into agent cost, tokens, time, code reuse and test status. Release notes

#12 🐙 JuliusBrussee/caveman

JuliusBrussee/caveman - JuliusBrussee/caveman is JuliusBrussee’s open-source skill and proxy for making coding agents read and produce fewer tokens while keeping code unchanged. Caveman 4.0.1 was released on October 13, 2026, with reusable components for concise replies, input compression, model selection, repeatable scripts, waste reduction, and compressed browsing. Worth trying if you need token controls across multiple coding agents and can use Node.js 22.13+. Release notes

#13 🐙 xerj-org/xerj

xerj-org/xerj - xerj-org/xerj is an Apache-2.0 local AI search project that automatically indexes folders containing code, docs, logs, PDFs, SQLite databases, and CSVs for agent retrieval. The repository includes its engine, functions, deployment materials, benchmarks, recipes, demos, tools, and documentation, with Elasticsearch client compatibility. Worth trying if you need local search across mixed project data for coding agents, RAG, audits, or memory.

#14 𝕏

Teresa Torres shared how evals built from personas based on failed calls were used to regression-test model and provider upgrades. She also highlighted automated flagging of sentiment, customer value, and product quality instead of routine transcript review.

#15 🐙 langfuse/langfuse

langfuse/langfuse - **langfuse/langfuse** is an open-source platform for tracing, evaluating, monitoring, and debugging LLM applications, with prompt management, datasets, a playground, APIs, integrations, and self-hosting options. Langfuse released v4.61.0 on October 16, 2026, adding gateway routing controls, skills onboarding and version traces, integration status, and OTEL-path measurement. Worth trying if you need one platform to inspect and improve AI applications. Release notes

#16 🐙 confident-ai/deepeval

confident-ai/deepeval - confident-ai/deepeval is an open-source framework for unit-testing LLM applications end to end, across agent trajectories and individual steps. The Apache-2.0 repository had 18,901 GitHub stars in the supplied snapshot and includes Python and TypeScript code, documentation, examples, tests, and metrics for agents, RAG, conversations, MCP, multimodal systems, and output quality. Worth trying if you need broad, configurable LLM evaluations.

#17 🐙 Arize-ai/phoenix

Arize-ai/phoenix - Arize-ai/phoenix is Arize-ai’s open-source AI observability platform, with tracing, evaluations, datasets, experiments, prompt management, a playground, Phoenix Intelligence, and a remote MCP server. Arize-ai released version 20.23.0 on October 16, 2026, adding OTLP server-log export and fixing Bedrock tool-result handling; the repository has 11870 stars. Worth trying if you need self-hosted tools for evaluating and troubleshooting AI applications. Release notes

#18 𝕏

NVIDIA AI recapped NVIDIA’s release of Nemotron 3.5 Super VL, detailing its post-training, multi-token prediction, and inference optimizations. The model is described as open and customizable, with weights, data, and recipes available on Hugging Face.

#19 🐙 ollama/ollama

ollama/ollama - **ollama/ollama** runs and manages open models locally or in the cloud, with app, REST API, command-line, authentication, and documentation code. Ollama released v0.40.5 on 2026-10-14T19:45:22Z, adding NVIDIA Nemotron 3.5 Super support—120B parameters, 12B active per token, and up to 1M tokens—plus cloud decision models and a custom-tool-call fix. Worth trying if you need a local or cloud model runtime with Nemotron support. Release notes

#20 🐙 heygen-com/hyperframes

heygen-com/hyperframes - **heygen-com/hyperframes** is an Apache-2.0 framework that turns HTML, CSS, media, and seekable animations into deterministic MP4 videos, with a CLI and 21 published agent skills. heygen-com released v0.8.164 on October 16, 2026, with 3 fixes; the repo has 64,887 stars and 5,318 weekly trending gains. Worth trying if you need agent-assisted video rendering. Release notes