Ollama adds macOS speech streaming and d3 decision models
Today's top 20 insights for PM Builders.
Ollama adds macOS speech streaming and d3 decision models
#1 🐙 ollama/ollama
ollama/ollama - ollama/ollama is an MIT-licensed project for running and managing open models locally or in the cloud; the repository includes app, API, CLI, authentication, model-creation, discovery, and documentation code. Ollama released Ollama v0.40.11 on 2026-10-24, adding macOS speech streaming through an OpenAI-compatible endpoint and vLLM Semantic Router’s d3 decision models in 2B, 4B, 9B, and 27B sizes. Worth trying if you need local model serving, speech, or decision workflows. Release notes
#2 🐙 aws/agent-toolkit-for-aws
aws/agent-toolkit-for-aws - **aws/agent-toolkit-for-aws** is AWS’s official collection of MCP servers, reusable skills, and plugins for helping AI coding agents build, deploy, and manage AWS applications. It supports Claude Code, Codex, Cursor, Devin, Kiro, and fx, and has 3,249 GitHub stars with 13 daily trending gains. Worth trying if you need agent access to AWS APIs, sandboxed scripts, and documentation search.
#3 🐙 android/skills
android/skills - android/skills provides reusable, AI-optimized instructions that help agents follow Android development guidance across device AI, Compose, media, navigation, security, Wear, XR, and more. android released Android Skills v1.0.17 on October 23, 2026, updating 11 of 33 skills; the Apache-2.0 repository has 9,643 stars and gained 1,317 weekly. Worth trying if you need modular Android guidance for coding agents. Release notes
hi3d-official/Hi3Dv1.0 - **hi3d-official/Hi3Dv1.0** was released by hi3d-official on October 23, 2026, with inference code, model weights, documentation, and examples for refining a coarse mesh into high-resolution geometry from a reference image. Built on Sparc3D, it supports 512, 1024, and 1536 resolutions; CPU inference is unavailable, and 1536³ was validated only on a 96 GB NVIDIA H20. Consider it if you need image-guided 3D mesh refinement and have suitable NVIDIA hardware. Release notes
#5 🐙 NVIDIA-NeMo/Speech
NVIDIA-NeMo/Speech - NVIDIA-NeMo/Speech is NVIDIA-NeMo’s Apache-2.0 framework for building and customizing speech recognition, text-to-speech, and speech language models. The repository includes reusable code, documentation, and examples, and the README links to pretrained checkpoints. Training requires an NVIDIA GPU with CUDA. Worth trying if you need a PyTorch-based toolkit for speech or multimodal AI.
thedotmack/claude-mem - **thedotmack/claude-mem** is an Apache-2.0 persistent-context system that captures agent activity, compresses it with AI, and injects relevant context into future sessions. On 2026-10-25 at 07:44:32Z, thedotmack released Version 15.1.0, replacing the localhost viewer at port 37777 with a live React My Work app and adding per-prompt server injection and opt-in MCP transport; the repo has 102,931 GitHub stars. Worth trying if you need cross-session memory. Release notes
paperclipai/paperclip - paperclipai/paperclip is an MIT-licensed Node.js server and React app for coordinating AI agent teams, with a CLI, integrations, adapters, docs and evaluations. paperclipai released Paperclip v2026.1023.0 on 2026-10-23 with 200 commits promoted from 2026.1012.1-beta.0, adding private work, user-provided model gateways and expanded MCP controls. Worth trying if you need one dashboard to govern multiple agents. Release notes
#8 🐙 mvschwarz/openrig
mvschwarz/openrig - **mvschwarz/openrig** is an Apache-2.0 system for running persistent Claude Code and Codex agent teams, with CLI packages, reusable skills, docs, and a demo. Owner mvschwarz released OpenRig 0.7.1 on October 25, 2026, adding per-seat automation and permission controls, TUI/Slack question routing, expanded MCP support, and managed compaction that survives daemon restarts. Worth trying if you need coordinated coding agents with shared context and owned work; Node.js 22 or 24 is required. Release notes
#9 🐙 Nutlope/hallmark
Nutlope/hallmark - **Nutlope/hallmark** is an MIT-licensed design skill for Claude Code, Codex, Cursor and OpenCode, described as made by Together AI, that aims to prevent AI-generated-looking interfaces. It includes 48 themes, custom design, redesign and audit modes, browser linting at eight widths, and 755 openly licensed photos; Node.js 18+ is required. Worth trying if you need reusable guardrails for AI-assisted product design.
#10 🐙 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; it includes a CLI, seven reusable agent skills, and about 450 editable components. heygen-com released v0.8.197 on October 25, 2026, cutting opaque VP9 cloud-rendering costs by about 40 percent and improving error reporting and memory use. Worth trying if you need agents to create repeatable videos from web content. Release notes
#11 🐙 promptfoo/promptfoo
promptfoo/promptfoo - promptfoo/promptfoo is an MIT-licensed CLI and library for evaluating, comparing, and red-teaming LLM applications, with declarative configurations and CI/CD and code-scanning support. Promptfoo released version 0.125.1 on October 25, 2026, restoring the authentic Discord invite. Worth trying if you need automated LLM evaluations or vulnerability scanning. Release notes
#12 🐙 langfuse/langfuse
langfuse/langfuse - langfuse/langfuse is an open-source platform for tracing, evaluating, and improving LLM apps; the repository contains the app, AI gateway, APIs, and self-hosting configurations. Langfuse released v4.70.0 on October 24, 2026, adding unstable read-only Topics endpoints, a runBatchEvaluation MCP tool, and broader Cmd+K navigation; it has 36,422 GitHub stars. Consider it if you need LLM observability and evaluation. Release notes
#13 🐙 morluto/rea
morluto/rea - **morluto/rea** is an MIT-licensed tool that lets AI agents locally reverse-engineer native binaries, JavaScript/Electron and .NET apps, websites, and runtime behavior, returning evidence and limitations. On October 25, 2026, morluto released rea-agents 6.8.0, adding local agent environments from captures or prompts, more file formats, investigation summaries, and CLI exports; the repository includes native and bridge components, documentation, examples, and experiments. Worth trying if you need evidence-backed analysis of software without source access. Release notes
#14 🐙 stablyai/orca
stablyai/orca - stablyai/orca is an MIT-licensed workspace for coordinating CLI coding agents across desktop, mobile and remote runtimes, and its repository includes source code, skills-related files and examples. stablyai released Orca v1.4.240 on October 25, 2026, adding searchable browser history, better chat and remote recovery, improved Codex task-status reporting, approval cards for certain Codex MCP tool approvals, Lean and Zig highlighting, support for saving more than 40 quick commands and the new `orca ports` command. The repository has 105234 GitHub stars and 9828 weekly trending gains. Worth trying if you manage multiple coding agents. Release notes
#15 🐙 NousResearch/hermes-agent
NousResearch/hermes-agent - **NousResearch/hermes-agent** is Nous Research’s MIT-licensed, self-improving AI agent, with CLI and desktop apps, messaging integrations, scheduled automation, subagent delegation, model switching, and local or remote execution. NousResearch released v0.22.5 on October 22, 2026, rolling 81 merged pull requests into a stable release across its distribution channels. Worth trying if you need a multi-channel agent that retains knowledge across sessions. Release notes
#16 🐙 ruvnet/ruflo
ruvnet/ruflo - **ruvnet/ruflo** is an MIT-licensed agent harness for coordinating AI workflows across Claude Code, Codex and other MCP clients, with CLI tools, plugins, memory, documentation and visual or terminal consoles. On October 25, 2026, ruvnet released v3.63.0 with 4 opt-in MCP tools for Anthropic Claude Managed Agents, covering budget-bounded workflows, gated controls and signed-outcome advisory learning. Worth trying if you need configurable multi-agent orchestration with explicit operational limits. Release notes
#17 🐙 langchain-ai/langchain
langchain-ai/langchain - **langchain-ai/langchain** is an MIT-licensed framework with libraries and integrations for building agents and LLM-powered applications. On October 22, 2026, langchain-ai released langchain-core 1.7.0, changing streamed tool-call arguments from `block-delta` snapshots to `tool-call-delta` slices; `.tool_calls.get()` and `.output` are unchanged. Consider it if you need interoperable components for AI applications, but pin `langchain-core<1.7` if you require snapshots. Release notes
#18 🐙 akitaonrails/ai-memory
akitaonrails/ai-memory - **akitaonrails/ai-memory** v3.2.3 was released by akitaonrails on October 25, 2026; the MIT-licensed repo has 13,361 GitHub stars and 3,541 weekly trending gains. It provides a self-hosted app, hooks, integrations, and docs for sharing git-backed Markdown memory across more than 20 coding-agent harnesses, with zero LLM calls by default. Worth trying if you need portable agent context across tools, machines, or teams. Release notes
#19 🐙 mvanhorn/last30days-skill
mvanhorn/last30days-skill - **mvanhorn/last30days-skill** is an MIT-licensed agent skill that researches Reddit, X, YouTube, Hacker News, Polymarket, and the web, then produces grounded summaries. It includes reusable skill and plugin files, configuration docs, and fixtures; mvanhorn released v3.30.0 on October 22, 2026, fixing the engine to search installed lanes on host plans and require an X handle decision. Worth trying if you need multi-source research in one of 50+ Agent Skills hosts. Release notes
#20 🐙 makecindy/cindy
makecindy/cindy - **makecindy/cindy** is an Apache-2.0 open-source client for a local AI agent that works with real files and logged-in apps, with desktop and mobile apps plus shared packages; its backend is separate. makecindy released Cindy desktop v0.1.115 on 2026-10-23, adding sequential group-chat replies, private-chat topics, Qwen and CosyVoice speech, task prewarming, and direct device connections with relay fallback. Worth trying if you need a customizable agent client across desktop and mobile. Release notes