https://reporadar.ai/ Evidence-linked AI discoveries ranked for practical usefulness, with news kept separate. Thu, 13 Aug 2026 19:31:03 +0000 https://github.com/infiniflow/infinity https://github.com/infiniflow/infinity infiniflow/infinity is a 4,669-star, Apache-2.0 AI-native database built for LLM applications that delivers incredibly fast hybrid search of dense vector, sparse vector, tensor (multi-vector), and full-text in a single C++20 system from the same team that built RAGFlow; HNSW indexing, BM25 + dense + sparse + multi-vector + tensor + native full-text combined; Python + C++ + Rust + TypeScript RepoRadar tier: Gold. Thu, 13 Aug 2026 18:30:00 +0000 https://github.com/qdrant/fastembed https://github.com/qdrant/fastembed qdrant/fastembed is a 3,140-star, Apache-2.0 lightweight, fast Python library for embedding generation built by Qdrant; uses ONNX Runtime instead of PyTorch so it ships without GBs of PyTorch dependencies, runs on CPU, and is fast enough for serverless runtimes like AWS Lambda; default text embedding model is Flag Embedding (MTEB leaderboard); supports dense, sparse (SPLADE), image (CLIP) RepoRadar tier: Gold. Thu, 13 Aug 2026 18:30:00 +0000 https://github.com/SciSharp/BotSharp https://github.com/SciSharp/BotSharp SciSharp/BotSharp is a 3,093-star, Apache-2.0 'Open Source AI Agent Application Framework' written in C# on .NET Core that connects LLMs to existing business applications with out-of-the-box multi-agent and conversation state management; plug-in and pipeline flow execution design that completely decouples the plug-ins; advanced Agent abstraction layer for managing complex enterprise scenarios; ML RepoRadar tier: Gold. Thu, 13 Aug 2026 18:30:00 +0000 https://github.com/aden-hive/hive https://github.com/aden-hive/hive aden-hive/hive is a 10,904-star, Apache-2.0 'Multi-Agent Harness for Production AI' from Aden (Y Combinator W26); the README tagline: 'the agent harness for production workloads -- state management, failure recovery, observability, and human oversight so your agents actually run'; supports OpenAI / Anthropic / Google Gemini + 102 MCP tools (per the MCP badge in the README); multi-agent systems RepoRadar tier: Gold. Thu, 13 Aug 2026 18:30:00 +0000 https://github.com/compozy/compozy https://github.com/compozy/compozy compozy/compozy is a 2,530-star, MIT 'operating system for AI agents' (CompozyOS) that wraps Claude Code, OpenClaw, Hermes, and Codex into a single Go-runtime agent OS handling loops / triggers / cron / memory / permissions / approvals / observability / glue scripts without hand-rolling each; browser-based supervisor, project-level shared memory, parallel loops; pkg.go.dev published; 144 forks RepoRadar tier: Gold. Thu, 13 Aug 2026 00:02:00 +0000 https://github.com/vshulcz/deja-vu https://github.com/vshulcz/deja-vu vshulcz/deja-vu is a 628-star, MIT agent-session search engine that indexes the months of session history Claude Code / Codex / Pi / Aider / etc. already wrote to disk before the user installed it; 84.9% hit@1 on LongMemEval-S, no LLM, no embeddings, no API key, fully local; single zero-dependency binary exposed over MCP for any agent harness; npm @vshulcz/deja-vu for the npm install path; 43 RepoRadar tier: Silver. Thu, 13 Aug 2026 00:02:00 +0000 https://github.com/MoonshotAI/checkpoint-engine https://github.com/MoonshotAI/checkpoint-engine MoonshotAI/checkpoint-engine is a 995-star, MIT middleware that updates LLM inference engine weights in-place during RL training (a critical step); updating Kimi-K2 (1T params) across thousands of GPUs in ~20s; two implementations (Broadcast -- default, fastest, for synchronous updates; P2P -- via mooncake-transfer-engine, for adding new inference instances without disturbing serving workloads) RepoRadar tier: Gold. Thu, 13 Aug 2026 00:02:00 +0000 https://github.com/intel/auto-round https://github.com/intel/auto-round intel/auto-round is a 1,564-star, Apache-2.0 advanced quantization toolkit for LLMs and VLMs that achieves high accuracy at ultra-low bit widths (2-4 bits) with minimal tuning via sign-gradient descent; full hardware compatibility for CPU / XPU / CUDA with multi-datatype support; full compatibility with vLLM, SGLang, and Transformers; backed by SignRoundV1 (arXiv 2309.05516) and SignRoundV2 RepoRadar tier: Gold. Thu, 13 Aug 2026 00:02:00 +0000 https://github.com/Arize-ai/openinference https://github.com/Arize-ai/openinference Arize-ai/openinference is a 1,140-star, Apache-2.0 OpenTelemetry SDK that instruments every major LLM / agent / vector-store framework (OpenAI, Anthropic, Gemini / Vertex, LangChain, LangGraph, LlamaIndex, Haystack, OpenAI Agents, Vercel AI, Pydantic AI, smolagents, MCP) and emits OTLP-compatible traces to any OTEL-compatible collector; native OTel semconv gen_ai.* span attributes RepoRadar tier: Gold. Wed, 12 Aug 2026 19:30:00 +0000 https://github.com/AI-Hypercomputer/maxtext https://github.com/AI-Hypercomputer/maxtext AI-Hypercomputer/maxtext is a 2,387-star, Apache-2.0 Google Cloud reference JAX LLM training stack -- the canonical TPU-native implementation that ships reference training + inference recipes for every major open model (Llama 2/3/4, DeepSeek V3.2 / R1 0528 / V3 0324, Qwen 2.5 / 3 MoE 30B/235B/480B / 3 Dense, Kimi K2 / K2-Thinking / K2.5 / K2.6, Mixtral, Mistral, Gemma 2/3, GPT-OSS 20B/120B) RepoRadar tier: Gold. Wed, 12 Aug 2026 19:30:00 +0000 https://github.com/omnara-ai/omnara https://github.com/omnara-ai/omnara omnara-ai/omnara is a 2,715-star, Apache-2.0 unified push / pause / steer / observe API for production-grade agents; a single control plane that lets any agent framework (Claude Code, Codex, LangGraph, CrewAI, AutoGen, custom) surface state to a human operator via a dashboard + push notifications + Slack / Discord / Telegram / SMS / email with durable-agent semantics RepoRadar tier: Gold. Wed, 12 Aug 2026 19:30:00 +0000 https://github.com/pbakaus/impeccable https://github.com/pbakaus/impeccable pbakaus/impeccable is a 58,534-star, Apache-2.0 design-system enforcement layer for AI coding agents; ships a CLI (`npx impeccable`) + plugin bundle + per-agent skill packs for 14 supported tools (Cursor, Claude Code, GitHub Copilot, Gemini CLI, Codex CLI, Grok Build, OpenCode, Pi, Kiro, Trae, Rovo Dev, Qoder, Mistral Vibe, Google Antigravity); the skill packs ship a ruleset RepoRadar tier: Gold. Wed, 12 Aug 2026 19:30:00 +0000 https://github.com/Untrivial-ai/agent-orchestrator https://github.com/Untrivial-ai/agent-orchestrator Untrivial-ai/agent-orchestrator is a 9,428-star, Apache-2.0 agentic IDE that supervises parallel AI coding agents (Claude Code, Codex, Aider, OpenCode, Grok, Droid, Amp, Auggie, Cursor, Qwen, Copilot, Goose, Continue) in isolated workspaces with automatic CI-fix / merge-conflict / code-review feedback loops; macOS / Windows / Linux desktop builds (DMG / exe / AppImage / deb / rpm), a persistent RepoRadar tier: Gold. Wed, 12 Aug 2026 17:30:00 +0000 https://github.com/radixark/miles https://github.com/radixark/miles radixark/miles is a 1,969-star, Apache-2.0 enterprise-grade reinforcement learning framework for large-scale model post-training that pairs SGLang for high-throughput rollout with Megatron-LM for scalable training and ships precision, stability, and observability features for trillion-parameter RL runs; day-0 supports for Kimi K3 (2026-07-27), Inkling (2026-07-15), DeepSeek-V4 Flash on AMD RepoRadar tier: Gold. Wed, 12 Aug 2026 17:30:00 +0000 https://github.com/maka-agent/maka-agent https://github.com/maka-agent/maka-agent maka-agent/maka-agent is a 1,331-star, Apache-2.0 local-first AI desktop assistant built for real work with a 'Log is the Runtime' architecture — model messages, Tool Calls, Tool Results, and termination facts enter a Runtime Event Log; sessions, UI, model context, and recovery are projections over that log; local-first (sessions, settings, run records stay on machine by default), choose cloud RepoRadar tier: Gold. Wed, 12 Aug 2026 17:30:00 +0000 https://github.com/sybil-solutions/local-studio https://github.com/sybil-solutions/local-studio sybil-solutions/local-studio is a 1,659-star, Apache-2.0 local-first workstation for running, managing, and using self-hosted LLM backends; v2.0 unifies day-to-day operation around Status / Workbench / Configure / Usage; one machine can launch models, watch GPU/runtime state, chat with OpenAI-compatible endpoints, run agent sessions against local or remote controllers; Bun/Hono controller + RepoRadar tier: Gold. Wed, 12 Aug 2026 17:30:00 +0000 https://github.com/kirodotdev/KiroCrew https://github.com/kirodotdev/KiroCrew kirodotdev/KiroCrew is a 2,750-star, Apache-2.0 persistent workspace for development work that self-improves and continues beyond one session; persistent sessions, memory, schedules, and task checkpoints survive Gateway restarts; self-learning (corrections and task failures become durable lessons); self-evolving (repeated patterns become reusable skills, all visible and editable); runs on Mac / RepoRadar tier: Gold. Wed, 12 Aug 2026 17:30:00 +0000 https://github.com/Tencent-Hunyuan/Hunyuan3D-WorldClaw https://github.com/Tencent-Hunyuan/Hunyuan3D-WorldClaw Tencent-Hunyuan/Hunyuan3D-WorldClaw is a 577-star open-source agentic framework (released 2026-08-10) that turns one open-ended prompt into an explicit, explorable, and editable open-world 3D scene; one of the first reproducible agentic 3D world generation systems that ships weight + project page + working code path on GitHub, building on the existing Tencent-Hunyuan/Hy3 family. RepoRadar tier: Gold. Wed, 12 Aug 2026 13:00:00 +0000 https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 NVIDIA-Nemotron-3.5-Lightning-30B-A3B is a 31.6B total / 3.6B active hybrid Mamba-Transformer open-weight model under NVIDIA Open Model License (with NVFP4 and BF16 variants), released 2026-08-04 on Hugging Face; at the time of the 2026-08-11 NVIDIA blog post it had 19,250 downloads and 192 likes. RepoRadar tier: Gold. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/mitkox/stolen-thoughts https://github.com/mitkox/stolen-thoughts mitkox/stolen-thoughts is the open-source artifact for 'Stealing Reasoning Traces from Proprietary LLM APIs', the 676pt #1 HN research piece for 2026-08-11; it demonstrates a concrete attack class that recovers proprietary chain-of-thought content via an inference-time side channel and ships the reference implementation under MIT. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/lyramakesmusic/neuralese-leaker https://github.com/lyramakesmusic/neuralese-leaker lyramakesmusic/neuralese-leaker is a 70-star Python web app for chatting with LLMs on OpenRouter while exposing leaked unabridged reasoning traces that providers try to redact; the 2026-08-11 release pairs the standalone Hugging Face Space with a one-command local server. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/activeing123/mcptoon https://github.com/activeing123/mcptoon activeing123/mcptoon is a 125-star Apache-2.0 Python Show-HN-launched (71p) token-efficient MCP CLI client that claims 97% fewer tokens on tool discovery and 40-60% fewer on results, with zero dependencies and cross-platform support that works with every AI agent. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/TheRealYT/git-knife https://github.com/TheRealYT/git-knife TheRealYT/git-knife is a 223-star TypeScript Show-HN-launched (160p, 100 comments) tool that lets you edit commit messages, authors, and dates across an entire repo in a spreadsheet-style UI rather than one commit at a time; license field is null on the GH API snapshot. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/run-llama/ExtractBench https://github.com/run-llama/ExtractBench run-llama/ExtractBench is a 35-star Apache-2.0 Show-HN-launched open-source benchmark for schema-guided enterprise document extraction; it ships with reproducible datasets, a leaderboard, and a working LlamaIndex extraction pipeline under Apache-2.0. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/karanb192/hindcast https://github.com/karanb192/hindcast karanb192/hindcast is a 7-star MIT-licensed JavaScript tool that lets you browse, search, and replay every Claude Code session on your Mac; it indexes past session JSONL files locally and presents them as a searchable UI, with replay-to-resume supported. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/Riccardo8888/agent-link https://github.com/Riccardo8888/agent-link Riccardo8888/agent-link is a 41-star MIT-licensed Python end-to-end encrypted channel that connects coding agents running anywhere (laptop, CI, server) without an intermediary broker; surfaces as a search-discovered (not Show HN) tool in the AI/LLM GitHub Search band on 2026-08-11. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/helloyulife/CoderBar https://github.com/helloyulife/CoderBar helloyulife/CoderBar is a 52-star native macOS menu-bar app (Swift) that monitors and approves Codex / Claude coding-agent runs from the menu bar, with one-click jump-to-session and a per-run approval gate; license field null on the GH API snapshot. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/visnia-ai/browser-agent https://github.com/visnia-ai/browser-agent visnia-ai/browser-agent is a 10-star MIT-licensed TypeScript Show-HN-launched browser-automation agent focused on cost-efficient LLM-driven browser actions; ships with a working local runner and an OpenAI-compatible API. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/marvin-socialista/claude-design-mcp https://github.com/marvin-socialista/claude-design-mcp marvin-socialista/claude-design-mcp is a 32-star MIT-licensed JavaScript MCP server that lets Claude Code talk to Anthropic Claude Design product (send design instructions and read its replies) over the MCP protocol; appears in the 2026-08-11 MCP-search band. RepoRadar tier: Silver. Wed, 12 Aug 2026 13:00:00 +0000 https://github.com/microsoft/markitdown https://github.com/microsoft/markitdown microsoft/markitdown is a 173,143-star, MIT-licensed Python 3.10+ lightweight utility from the Microsoft AutoGen Team that converts diverse files (PDF, PowerPoint, Word, Excel, Images with EXIF + OCR, Audio with EXIF + speech transcription, HTML, CSV/JSON/XML, ZIP archives, YouTube URLs, EPubs) into Markdown — the format mainstream LLMs natively 'speak' — for downstream LLM and text-analysis RepoRadar tier: Gold. Tue, 11 Aug 2026 18:05:00 +0000 https://github.com/muratcankoylan/Agent-Skills-for-Context-Engineering https://github.com/muratcankoylan/Agent-Skills-for-Context-Engineering muratcankoylan/Agent-Skills-for-Context-Engineering is a 17,687-star, MIT-licensed comprehensive open collection of Agent Skills focused on context engineering and harness engineering principles for building production-grade AI agent systems, cited as foundational work in arXiv 2601.21557 'Meta Context Engineering via Agentic Skill Evolution' RepoRadar tier: Gold. Tue, 11 Aug 2026 12:02:00 +0000 https://github.com/herdrdev/herdr https://github.com/herdrdev/herdr herdrdev/herdr is a 27,352-star, Apache-2.0 single-Rust-binary runtime that coding agents live on — an always-running background server that owns the terminals of claude code / codex / cursor / opencode / grok and the rest; close the lid, drop the network, or restart the machine and agents keep working and sessions come back; reattach from any terminal or over ssh; every pane is marked working / RepoRadar tier: Gold. Tue, 11 Aug 2026 12:02:00 +0000 https://github.com/ace-agent/ace https://github.com/ace-agent/ace ace-agent/ace is a 1,255-star, Apache-2.0 framework for Agentic Context Engineering that enables large language models to self-improve by treating contexts as evolving playbooks, backed by arXiv 2510.04618 'Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models' (2025), with a three-role agentic architecture (Generator, Reflector, Curator) that performs incremental RepoRadar tier: Gold. Tue, 11 Aug 2026 12:02:00 +0000 https://github.com/MigoXLab/webqa-agent https://github.com/MigoXLab/webqa-agent MigoXLab/webqa-agent is a 225-star, Apache-2.0 fully-automated web testing agent with multimodal page understanding and three testing forms — WebQA Flash (default · recommended, Chrome DevTools MCP-backed, second-scale execution, zero selector maintenance, multimodal AI identifies page elements directly), Standard Generate (AI discovery → dynamic generation → precise execution), and Run Mode RepoRadar tier: Silver. Tue, 11 Aug 2026 12:02:00 +0000 https://github.com/apple/embedding-atlas https://github.com/apple/embedding-atlas apple/embedding-atlas is a 4902-star, MIT-licensed interactive visualization tool for large embeddings and their metadata from Apple, backed by arXiv 2505.06386 'Embedding Atlas: Low-Friction, Interactive Embedding Visualization' (Donghao Ren, Fred Hohman, Halden Lin, Dominik Moritz); ships as `pip install embedding-atlas` (PyPI), `npm install embedding-atlas` (npm with React / Svelte bindings) RepoRadar tier: Gold. Tue, 11 Aug 2026 08:01:00 +0000 https://github.com/stakpak/agent https://github.com/stakpak/agent stakpak/agent is an Apache-2.0, 1726-star Rust DevOps autopilot that lives on your machines 24/7 and only pings a human when it needs one; install via `curl -sSL https://stakpak.dev/install.sh | sh` then `stakpak autopilot up`, runs as a systemd user service with secret substitution (the LLM never sees your real credentials), Warden network-level guardrails that block destructive operations RepoRadar tier: Gold. Tue, 11 Aug 2026 08:01:00 +0000 https://github.com/SuanmoSuanyangTechnology/MemoryBear https://github.com/SuanmoSuanyangTechnology/MemoryBear SuanmoSuanyangTechnology/MemoryBear is a 4968-star, Apache-2.0 next-generation AI memory system from RedBear AI that models the full memory lifecycle of 'perception → extraction → association → forgetting' inspired by biological cognitive mechanisms (hippocampus encoding, neocortex consolidation, synaptic pruning), backed by arXiv 2603.22306 RepoRadar tier: Silver. Tue, 11 Aug 2026 08:01:00 +0000 https://github.com/pinchbench/skill https://github.com/pinchbench/skill pinchbench/skill is a 1309-star, MIT-licensed real-world benchmarking system for AI coding agents from PinchBench (Organization created 2026-02-10, 5 public repos, 125 followers), with latest release v2.0.0 (2026-05-06) and a 53-task suite that measures how well LLM models perform as the brain of an OpenClaw agent on real tasks RepoRadar tier: Silver. Tue, 11 Aug 2026 08:01:00 +0000 https://huggingface.co/meta-models/Muse-Glimmer-30B https://huggingface.co/meta-models/Muse-Glimmer-30B meta-models/Muse-Glimmer-30B is a 30B-parameter open-weight model from Meta Superintelligence Labs, released 2026-08-10 under Apache-2.0; designed for local agent workflows on a single consumer GPU with strong results on DeepSearch QA, MCP-Atlas, tau-Bench, and SWE-Bench; optimized llama.cpp / MLX / ExecuTorch integrations land in the days after release. RepoRadar tier: Gold. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/antirez/h3.c https://github.com/antirez/h3.c antirez/h3.c is an MIT-licensed, 742-star C implementation of a native Apple Silicon inference engine for the MiniMax-H3 video-and-audio model; runs prompt-to-video / prompt-to-audio / first-last-frame / ordered Ref2VA references end-to-end from a single binary with a Metal backend, with --show preview decode for Kitty / Ghostty / iTerm2 / WezTerm / Konsole and an Iris-style interactive session. RepoRadar tier: Gold. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/cactus-compute/needle https://github.com/cactus-compute/needle cactus-compute/needle is an MIT-licensed, 3,541-star project from Cactus Compute; Needle 2 is a 45M-parameter open 14MB binary that runs full agentic sessions in 28MB RAM, hits 500 tok/s decode on a Raspberry Pi 5, 300-700 on sub-$200 phones, and runs on ESP32-S3 microcontrollers; trades wins with FunctionGemma 270M, LFM2.5 230M, and Apple FM at 5x-70x smaller and 2 bits against their f16. RepoRadar tier: Gold. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/AntigmaLabs/ante https://github.com/AntigmaLabs/ante AntigmaLabs/ante is an Apache-2.0, 1,051-star Rust coding-agent harness that ships as a single binary, works offline, has no model or vendor lock-in, and runs as 'Ghost in your shell' with a highly optimized core. RepoRadar tier: Gold. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/lexmount/moli https://github.com/lexmount/moli lexmount/moli is a dual Apache-2.0 / MIT-licensed, 60-star, 17.4 MB Rust headless browser engine built from scratch for AI agents; runs real V8 JavaScript with native DOM/CSS state, exposes pages through structured data first, and only renders pixels on demand; CLI, CDP, WebDriver Classic, and WebDriver BiDi support. RepoRadar tier: Silver. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/0xnyn/airship https://github.com/0xnyn/airship 0xnyn/airship is an MIT-licensed, 60-star TypeScript visual editor that puts an infinite design canvas in front of a running dev server; select an element, describe a change, and watch Claude Code, Codex, or OpenCode update the source without rebuilding the UI in a separate design tool. Distributed via `npx @airshiplabs/cli`. RepoRadar tier: Silver. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/surya-koritala/loomfeed https://github.com/surya-koritala/loomfeed surya-koritala/loomfeed is an MIT-licensed, 157-star Go (with Next.js 15 frontend) self-hostable Reddit alternative that adds agent-as-first-class-participant, claim provenance, epistemic status labels, earned reputation, and structured agent-vs-agent debates; one `docker compose up` to deploy. RepoRadar tier: Silver. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/Kane-Wan/oss-maintainer-kit https://github.com/Kane-Wan/oss-maintainer-kit Kane-Wan/oss-maintainer-kit is an MIT-licensed, 34-star TypeScript CLI and GitHub Action that helps maintainers review pull-request metadata and diffs, triage issues, draft maintainer replies, and turn change lists into structured release notes via the OpenAI Responses API; treats repository content as untrusted data and never executes code from a pull request. RepoRadar tier: Silver. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/MkThingsHQ/mkagent https://github.com/MkThingsHQ/mkagent MkThingsHQ/mkagent is an Apache-2.0, 40-star TypeScript local-first AI agent workspace that runs on Desktop, WebUI, and CLI; powered by the badlogic/pi-mono agent runtime with persistent local workspaces, model flexibility, browser tools, and document tools; application state stays under a local data directory and credentials are stored through the OS keychain. RepoRadar tier: Silver. Tue, 11 Aug 2026 03:00:00 +0000 https://github.com/aiming-lab/SimpleMem https://github.com/aiming-lab/SimpleMem aiming-lab/SimpleMem is an MIT-licensed, 3682-star Python library for efficient lifelong memory in LLM agents (text + multimodal), backed by an arXiv paper (2601.02553, cs.AI); the unified `simplemem` package auto-routes between SimpleMem (text) and Omni-SimpleMem (multimodal: text, image, audio, video), with EvolveMem (v3.0) adding LLM-driven self-evolving retrieval RepoRadar tier: Silver. Mon, 10 Aug 2026 22:01:00 +0000 https://github.com/openonion/connectonion https://github.com/openonion/connectonion openonion/connectonion is an Apache-2.0, 1316-star Python framework for production-ready AI agents: a thin agent core (model + prompt + tools) wrapped by 12 lifecycle hooks, a plugin system (re_act / auto_compact / subagents / ulw / eval), Claude-Code-compatible skills auto-discovery from .co/skills, multi-agent trust via fast rules (open / careful / strict), `host()` for HTTP + P2P agent relay RepoRadar tier: Silver. Mon, 10 Aug 2026 22:01:00 +0000 https://github.com/ldclabs/anda https://github.com/ldclabs/anda ldclabs/anda is a dual Apache-2.0 + MIT-licensed, 437-star Rust framework for building composable AI agent runtimes; agents and tools register through stable traits and function definitions so components compose into larger workflows without hard-coded application shapes, model requests route by capability labels (primary / pro / flash / lite) with provider adapters behind a common request/output RepoRadar tier: Silver. Mon, 10 Aug 2026 22:01:00 +0000