Currents
Current
Signals before they harden into patterns.
Currents preserve sensitivity to change: tools, projects, models, governance signals, and early tensions moving through the open source AI ecosystem.
Browse the knowledge base
-
SimDeck: Swift-Native iOS Agent Runtime with Accessibility-First Inspection
SimDeck implements a Swift-native iOS agent runtime that prioritizes framework-level inspection over pixel-based vision, using accessibility snapshots as a fallback layer to enable deterministic on-device UI interaction.
-
Langfuse: Open-Source LLM Observability and Agent Tracing
Langfuse provides an open-source observability platform for large language models, enabling comprehensive tracing, evaluation, and monitoring of complex agentic workflows and prompt executions.
-
Gemma 4 12B Local Agentic Workflows via Google AI Edge
Google DeepMind releases the 12B parameter Gemma 4 open model, optimized for local, on-device agentic and multimodal workflows through integration with the Google AI Edge runtime stack.
-
NAVA: Open-Source Audio-Video Generation Model
NAVA is an open-source multimodal foundation model capable of simultaneous video and audio synthesis, unifying scene generation, speech, background acoustics, and speaker synchronization into a single inference pass.
-
Chatterbox
An open-source text-to-speech model providing zero-shot synthesis, configurable emotion exaggeration, and embedded audio watermarking for deterministic agent voice output.
-
Claude Tap
Claude Tap operates as a local proxy and trace viewer that intercepts and visualizes agent tool calls and context windows to expose black-box execution behavior for deterministic debugging and auditability.
-
NVIDIA Cosmos 3: Open-Source Physical AI Foundation Model
NVIDIA releases Cosmos 3, an open-source foundation model unifying reasoning and generation for physical AI, supporting robotics and autonomous systems with NIM microservice deployment.
-
Claw Patrol: Open-Source Security Firewall for AI Agents
Claw Patrol implements an open-source security firewall layer for AI agents, providing perimeter defense and policy enforcement for autonomous agent workflows.
-
DNS-AID: Decentralized AI Agent Discovery Protocol
DNS-AID is an open-source protocol enabling autonomous agents to resolve peer identities and establish communication channels using standard DNS infrastructure, decoupling agent discovery from centralized registry dependencies.
-
OpenHands Dynamic Workflow for Autonomous Test Coverage
An OpenHands SDK-based workflow that autonomously generates and applies test improvements across entire codebases to incrementally raise repository coverage without manual intervention.
-
AutoScientists: Open-Source Scientific Discovery Agent System
AutoScientists is an open-source multi-agent framework that automates hypothesis generation, experimental design, and literature synthesis through structured social interaction protocols between specialized scientific reasoning agents.
-
Firefox Smart Window: AI-Powered Local-First Browsing Mode
Mozilla tests an AI-enhanced browsing interface that integrates tab summarization, contextual analysis, and search assistance while enforcing local chat storage and explicit opt-in data access controls.
-
ktx: Self-Improving Context Layer for Data Warehouse Agents
ktx provides a self-improving context layer that enhances data warehouse querying accuracy by dynamically generating schema descriptions and documentation, enabling LLM agents to construct precise SQL queries without reliance on static metadata.
-
Liteparse: Zero-Dependency PDF Extraction
Liteparse provides a zero-dependency, local-first PDF text extraction library that avoids LLM inference and cloud API requirements, enabling deterministic document parsing for simple text retrieval workloads without computational overhead.
-
PilotDeck: White-Box Agent OS with Traceable WorkSpaces
OpenBMB releases PilotDeck, an open-source agent operating system that replaces opaque execution environments with white-box, traceable WorkSpaces designed for structured state management and verifiable agent workflows.
-
Pullfrog: GitHub Actions Bridge for Coding Agents
Pullfrog implements a GitHub Actions-based bridge that enables coding agents to react to repository events such as pull requests, issues, and pushes without direct API dependency.
-
SkillOpt
Microsoft releases SkillOpt, an open-source framework that applies neural network training paradigms to optimize agent skill parameters independently of the underlying model weights, enabling skill evolution without full model fine-tuning.
-
DeepSWE: Open-Source Programming Agent Benchmark
Datacurve releases DeepSWE, an open-source benchmark evaluating long-horizon trajectory planning and complex codebase editing capabilities of frontier models across TypeScript, Go, Python, JavaScript, and Rust.
-
ElevenLabs Speech Engine Skill Open Source
ElevenLabs releases the Speech Engine Skill as an open-source component conforming to the Agent Skills specification, providing a standardized interface for AI agents and large language model applications to integrate real-time voice conversation capabilities.
-
OctoTools: Open-Source Python Agentic Framework for Modular Tool Routing
OctoTools implements a Python-based agentic architecture that decouples tool definitions into discrete cards, routing them through a dedicated planner and controller to manage complex, multi-step reasoning workflows without monolithic framework dependencies.
-
Perplexity Bumblebee: Open-Source MCP Malware Scanner
Perplexity releases Bumblebee, an open-source scanner leveraging the Model Context Protocol to detect software infections on local systems without triggering malicious execution or disrupting active processes.
-
Self-Hosted Search Indexing and Private Agentic Web Crawling
Emerging adoption of self-hosted full-text indexing tools and local LLM agents for private web crawling and summarization, signaling a shift toward sovereign data retrieval and inference infrastructure outside commercial search ecosystems.
-
Superserve
Superserve provisions persistent, isolated agent runtimes via Firecracker microVMs, exposing sandboxed filesystem and network environments through a pip-installable CLI workflow.
-
UI-TARS Desktop
ByteDance releases UI-TARS Desktop, an open-source multimodal agent interface that integrates vision-language models with desktop automation capabilities to enable autonomous UI interaction and control via a unified local-first runtime.
-
OpenChronicle: Long-Term Memory Layer for AI Agents
OpenChronicle implements a long-term memory layer for AI agents, addressing context fragmentation by persisting work states and reducing repetitive input overhead in agentic workflows.
-
Presenton: Open-Source AI Presentation Generation Tool
Presenton provides a self-hostable, open-source framework for AI-driven presentation generation, enabling multi-provider LLM and image model integration with structured export pipelines for PPTX and PDF formats.
-
E2a: Open-Source Email Gateway for AI Agents
E2a implements an open-source gateway architecture that exposes email interaction capabilities to autonomous AI agents via standardized interfaces, enabling agentic workflows to send, receive, and process messages without direct dependency on proprietary email provider APIs.
-
Honeyslop: Canary for AI-Hallucinated Bug Reports
Honeyslop implements a canary-based detection mechanism to identify and triage AI-hallucinated bug reports, providing a governance layer for automated issue tracking workflows.
-
Mirage: Unified Virtual Filesystem for AI Agents
Mirage aggregates disparate storage and communication services into a single virtual filesystem tree accessible via standard bash commands, enabling AI agents to interact with heterogeneous data sources through a unified, local-first interface.
-
RAMPART: Adversarial Agent Safety Testing via pytest
Microsoft releases RAMPART, a pytest-based framework for automated adversarial safety testing of AI agents, allowing developers to define security scenarios as executable tests with configurable pass/fail thresholds in CI pipelines.
-
SafeAgent: Open-Source Governed Execution Boundary for Autonomous Systems
SafeAgent implements an open-source governance layer that establishes execution boundaries for autonomous AI systems, enforcing constraints on command execution and file modification without serving as a standalone agent framework.
-
Cohere Releases Powerful Open-Source Model as Secure Alternative
Cohere releases its most capable model as open weights, positioning the release as a secure, sovereign alternative to the growing concentration of open-source AI infrastructure in China.
-
Koog: Kotlin AI Agent Framework
Koog is a Kotlin-native AI agent framework reaching version 1.0, providing type-safe workflow DSLs, persistence mechanisms for long-running autonomous operations, and deep integration within the JetBrains ecosystem, with adoption in industrial maintenance workflows.
-
Midscene.js: Visual Agent SDK for Cross-Platform Browser Automation
Midscene.js provides a cross-platform visual agent SDK that enables AI models to interpret screen layouts and execute interface interactions across web, desktop, and mobile environments using a unified codebase.
-
Kooky: AI-Native Terminal for Code Generation
Kooky is an open-source terminal application designed to streamline AI-assisted code generation by providing integrated management for multiple AI coding assistants and their conversational contexts.
-
OpenAIRE Graph: Authoritative Metadata Integration for Autonomous AI Agents
OpenAIRE exposes its scholarly metadata graph via standardized interfaces to enable autonomous AI agents to query authoritative research entities, persistent identifiers, and verified relationships for traceable, citation-grounded inference.
-
AWS Strands: Open-Source Agent Harness with Behavioral Middleware and Sandboxed Execution
AWS releases Strands, an open-source agent SDK repositioned as an agent harness featuring behavioral middleware, sandboxed execution environments, and opinionated defaults, reaching 1.5 million weekly downloads.
-
HyperFrames: HTML-Native Video Rendering Framework
An open-source, HTML-first rendering framework that enables deterministic video composition via standard web technologies and a plugin-driven animation runtime, targeted at agentic workflow integration.
-
OpenOSINT: Open-Source OSINT Investigation Agent
OpenOSINT is an open-source autonomous agent framework that automates OSINT investigations by chaining tools across email, username, domain, IP, and phone targets to generate structured reports without manual terminal intervention.
-
The Agent Sandbox Taxonomy
The Agent Sandbox Taxonomy defines a structured evaluation framework for AI agent execution environments, mapping seven defense layers, seven threat categories, and three evaluation dimensions to assess sandbox isolation, containment, and security posture.
-
Agent S: Open-Source OS Interaction Framework for Agentic Workloads
An open-source framework enabling large language model agents to interact with desktop operating systems through structured UI automation and state-aware task execution, achieving 72.60% success on the OSWorld benchmark.
-
NVIDIA SANA-WM: Open-Source World Model for Minute-Scale 720p Video Generation
NVIDIA releases SANA-WM, a 2.6-billion-parameter open-source world model capable of generating minute-scale 720p video sequences on consumer-grade single GPUs through optimized diffusion architecture.
-
Camelot: Self-Hosted Elixir/Phoenix AI Interface
Camelot is a self-hosted, transparent AI interface built on Elixir and Phoenix, utilizing LiveView for reactive UI and supporting multiple LLM backends without proprietary black-box dependencies.
-
Browser-Use: Open-Source Browser Automation via Natural Language
Browser-Use is an open-source automation framework that enables AI models to interact with web browsers via natural language instructions, providing a structured interface for agentic web navigation and task execution.
-
HearthNet: Edge Multi-Agent Orchestration for Smart Homes
HearthNet deploys independent edge-resident agent runtimes across smart home devices, using OpenClaw as the execution platform and a dedicated librarian process to mirror coordination traffic for audit and recovery.
-
TokenSpeed: Open-Source LLM Inference Engine for Agentic Workloads
TokenSpeed is an MIT-licensed, open-source LLM inference engine developed from scratch specifically for agentic workloads, demonstrating competitive performance against established benchmarks.
-
Agentic Context Engine (ACE)
Agentic Context Engine (ACE) is a Python framework that implements persistent memory and skill accumulation for LLM agents through a structured learning loop, enabling agents to refine behavior based on execution traces and a shared Skillbook repository.
-
CUA: Computer Use Agent Infrastructure
CUA is an open-source infrastructure providing a sandboxed runtime environment that enables autonomous AI agents to interact with and control full desktop operating systems via computer-use interfaces.
-
Datadog Toto 2: Open-Source Time Series Forecasting Model Family
Datadog releases Toto 2, an open-source time series forecasting model family spanning 4M to 2.5B parameters, establishing the first validation of scaling laws within the time series domain.
-
Dograh AI: Open-Source Voice Agent Platform
Dograh AI is an open-source voice agent platform that enables production-grade AI voice bot construction via drag-and-drop workflows, integrating with major model providers and coding assistants.
-
WhichLLM: Local LLM Hardware Benchmarking
A repository-based benchmarking utility that ranks local large language models against user hardware specifications to identify optimal inference configurations.
-
Introducing AIMX: a self-hosted, open-source email server designed for AI agents.
A self-hosted, open-source email server architecture that replaces legacy SMTP conventions with markdown-native payloads, MCP-over-stdio communication, and direct MTA-to-MTA routing to natively support autonomous AI agent workflows.
-
Gorse: AI-Powered Open Source Recommender System Engine
Gorse is an open-source recommender system engine that integrates classical ranking algorithms with LLM-based rankers and multimodal embeddings to provide scalable recommendation infrastructure for autonomous agents and data-driven applications.
-
livekit/agents
livekit/agents is a production-oriented Python framework for building real-time, server-side agent participants that join LiveKit rooms over WebRTC to handle audio, video, and data streams.
-
Raindrops: Open-Source Local Debugging and Evaluation Framework for AI Agents
Raindrops delivers an open-source, local-first toolkit for debugging, evaluating, and inspecting AI agent workflows, enabling developers to trace execution paths, analyze tool calls, and validate behavior on-device without external telemetry.
-
Verity: MCP-Based Fact-Checking Layer for Local AI Systems
Verity implements an open-source Model Context Protocol (MCP) layer that intercepts and evaluates model outputs and citations against trusted sources before local AI systems treat them as reliable, establishing a gatekeeping mechanism for factual integrity in autonomous workflows.
-
AI-Generated Zero-Day Exploit via OpenClaw Criminal Cluster
Google intercepted an AI-generated zero-day exploit targeting a 2FA bypass in an open-source administrative tool, attributed to a criminal cluster utilizing the OpenClaw framework, with detection characterized by hallucinated CVSS scores, educational docstrings, and LLM-tutorial Python patterns.
-
Altimate Code
Altimate Code is an open-source data engineering toolchain offering over 100 deterministic tools for SQL analysis, column-level lineage tracking, dbt integration, and cloud warehouse connectivity to provide LLM agents with precise data engineering capabilities.
-
HybridClaw Google Workspace Integration and Credential Isolation
HybridClaw 0.13.0 introduces a Google Workspace integration layer enabling email, calendar, and spreadsheet operations while enforcing credential isolation to prevent LLM exposure of authentication secrets.
-
Label Studio: Open Source Data Labeling and AI Evaluation
Label Studio is an open-source data labeling platform supporting audio, text, image, video, and time-series annotation with a configurable UI and export capabilities for model training workflows.
-
Minecraft AI Agent Framework: Multi-Model Autonomous Construction and Resource Automation
MIT-licensed open-source framework enabling large language model integration within Minecraft environments to automate resource extraction, structural construction, and multi-agent coordination via standardized API adapters.
-
MiniCPM-V 4.6 On-Device Multimodal Model
Tsinghua University, ModelBest, and OpenBMB release MiniCPM-V 4.6, a next-generation on-device multimodal model optimized for local inference efficiency and edge deployment without cloud dependency.
-
WorldSeed: YAML-Driven AI World Simulation
An open-source simulation framework that interprets declarative YAML configurations to generate autonomous multi-agent environments with emergent narrative and systemic behaviors.
-
AI-Trader: Open-Source Agent-Native Trading Platform
AI-Trader implements an open-source, agent-native trading platform enabling autonomous agents to execute trades across stocks, crypto, forex, options, and futures while supporting collective intelligence mechanisms and copy trading workflows.
-
Flue: Open-Source Agent Harness Framework
Flue is an open-source agent harness framework developed by an Astro co-founder, enabling headless AI agent creation via minimal TypeScript and Markdown-based logic definitions for tasks such as code triage and translation.
-
Hermes Agent Learning Loop and Skill Hoarding Risk
Critique of Hermes Agent's self-improvement mechanism, identifying a risk where continuous skill accumulation without governance leads to resource bloat and degraded performance rather than sustained capability growth.
-
Thunderbolt: Mozilla's Open-Source Self-Hosted AI Client
Thunderbolt is an open-source, self-hosted AI client developed by MZLA Technologies under the MPL 2.0 license, offering a local-first interface for enterprise-grade AI interactions.
-
Digital Agency Genai Open Source Release
The Japanese Ministry of Digital Affairs releases Genai, an open-source implementation template for administrative AI workflows, establishing a standardized reference architecture for public-sector agent deployment.
-
GitHub Spec Kit: Specification-Driven Workflow Expansion for AI Coding Agents
GitHub Spec Kit evolves from a specification-to-task generator into a structured workflow system for AI coding agents, introducing new tools and presets to formalize development pipelines and mitigate the fragmentation associated with ad-hoc prompt engineering.
-
Google Gemma 4 Open Model Family
Google releases the Gemma 4 family of open-weight models derived from Gemini 3 research, optimizing for intelligence-per-parameter efficiency to support local inference and agentic development workflows.
-
MAGIQ: Quantum-Resistant Policy Enforcement for Multi-Agent Systems
MAGIQ introduces a framework for cryptographically enforced policy enforcement in multi-agent AI systems, utilizing quantum-resistant primitives and formal Universal Composability proofs to secure agent interactions against quantum adversaries with minimal performance overhead.
-
Microsoft Agent Framework Hyperlight CodeAct Support
Microsoft Agent Framework extends agent execution capabilities by integrating Hyperlight CodeAct, providing .NET and Python bindings for secure, isolated code execution within autonomous workflows.
-
PraisonAI: Open-Source Multi-Agent Framework with Built-in Planning and Self-Reflection
PraisonAI implements an open-source multi-agent orchestration framework featuring built-in planning memory, self-reflection mechanisms, and collaborative auto-correction across 100+ model providers, enabling declarative team configuration via minimal code interfaces.
-
QwenPaw: Personal AI Assistant with Multi-Channel Support
QwenPaw is a personal AI assistant platform that enables multi-channel chat integration and extensible skill execution across local and cloud deployments, evolving from the CoPaw project under the agentscope-ai organization.
-
Scribe
A tool for managing local LLM skills, enabling structured organization and deployment of agent capabilities on personal hardware.
-
Vane: Self-Hostable Conversational Search Engine
Vane is a self-hostable conversational search engine that wraps local or cloud LLMs to provide retrieval-augmented generation capabilities for web search, enabling operator-controlled inference routing without vendor lock-in.
-
Bifrost: Open-Source AI Gateway for Enterprise Governance and Observability
Maxim AI's Bifrost provides an open-source AI gateway that enforces governance, security, and observability across enterprise LLM request flows, aligning with NIST AI RMF functions.
-
AI-SDLC Framework
A declarative governance framework for AI-augmented software development lifecycles that routes issues through staged pipelines with assigned AI agents and human reviewers, enforcing quality gates and continuously learning agent trustworthiness.
-
NVIDIA OpenShell
NVIDIA OpenShell is a runtime environment for autonomous AI agents that enforces data isolation and credential protection via declarative YAML policies defining execution boundaries.
-
Free LLM Inference API Resources
A curated repository of free inference API endpoints for large language models, offering token-free access as an alternative to paid providers or local GPU-dependent deployments.
-
Hermes Agent: WebUI, WeChat Integration, and Gemma 4 Support
Hermes Agent introduces a WebUI interface and WeChat integration, enabling local deployment with the Gemma 4 model family while removing token costs for messaging connectivity.
-
Obscura: Headless Browser for AI Agents and Web Scraping
A headless browser runtime optimized for agentic workflows and web scraping, addressing fragility and overhead in legacy automation tools like Puppeteer and Playwright.
-
Overstory: Multi-Agent Orchestration for AI Coding Agents
Overstory provides a pluggable runtime adapter framework for multi-agent orchestration, enabling AI coding agents to coordinate across heterogeneous environments like Claude Code and Pi without vendor lock-in.
-
Google Multi-Token Prediction Draft Model for Gemma 4
Google releases an open-source multi-token prediction draft model for the Gemma 4 series, implementing speculative decoding to achieve up to 3x inference speedup while the primary model performs final verification.
-
Open Codesign: BYOK Local-First Design Tool
Open Codesign is a bring-your-own-key local-first design application that enables offline operation and AI model integration via Claude, GPT, and Gemini APIs while maintaining data sovereignty through client-side execution.
-
Policy-as-Code in AI Governance Tools for Autonomous Agents
Policy-as-code operationalizes AI governance by encoding organizational constraints and safety rules into machine-executable logic that enforces compliance automatically during agent runtime execution.
-
Text to CAD
An open-source project that converts natural language descriptions into 3D CAD models, bridging the gap between text input and parametric design generation.
-
Pixelle-Video
AIDC-AI's Pixelle-Video is an automated short video workflow engine that consolidates cutting, transitions, captioning, and rendering into a unified pipeline for social clips and product demos.
-
Warp Terminal-Based Agent Development Environment
Warp provides a terminal-native development environment that integrates AI agent orchestration, code execution, and session management into a unified interface designed for structured agentic workflows.
-
Strands Agents
Strands Agents is an open-source SDK for Python and TypeScript that provides a structured framework for building autonomous AI agents, emphasizing AWS ecosystem integration and simplified orchestration patterns.
-
Zep: Persistent Memory and Context Management for LLM Agents
An open-source memory layer that automatically extracts and stores conversation history, user profiles, and structured facts to extend LLM context windows beyond native token limits.
-
CLI Anything: Agent-Native CLI Wrapping Framework
A framework by HKUDS that wraps existing command-line tools and software into agent-native interfaces, enabling autonomous agents to interact with legacy and third-party CLI utilities through standardized, structured bindings.
-
Clicky: Desktop AI Teacher
Clicky is an open-source desktop application that deploys an AI teacher agent locally, providing persistent educational assistance and interactive learning support without cloud dependency.
-
Agent Browser: AI-First Browser Automation CLI
Agent Browser is a CLI-based automation framework by Vercel Labs that leverages AI to drive browser interactions, enabling autonomous navigation, DOM inspection, and action execution for agentic web workflows.
-
GBrain: The Open-Source Memory System for AI Agents
A GitHub-hosted repository by Garry Tan that implements a persistent memory layer for autonomous AI agents, emphasizing structured state management and cross-session continuity without reliance on ephemeral context windows.
-
Honeclaw: AI Trading Assistant Built on OpenClaw
A specialized autonomous agent runtime built on the OpenClaw framework, designed to execute algorithmic trading workflows with structured risk management and persistent memory for financial data analysis.
-
Orca IDE: Open-Source IDE for AI Coding Agents
Orca IDE is a StableAI-maintained open-source integrated development environment that provides a native runtime for AI coding agents, exposing tool execution and model inference through a unified interface optimized for autonomous code generation and debugging.
-
G0DM0D3 Multi-Model Prompt Routing
A locally-run AI chat interface that routes a single prompt across fifty-plus open-weight and proprietary models to enable direct comparative evaluation and selection of optimal outputs.
-
Context Mode – Solving the AI Context Window Problem
Context Mode provides an open-source solution for managing context window constraints by implementing mode-switching strategies that preserve instruction continuity across extended interactions.
-
DeepAnalyze: Agentic LLM for Autonomous Data Science
RUC DataLab and Tsinghua's open-weight 8B agentic model for end-to-end autonomous data science — from data preparation and analysis to modeling, visualization, and report generation.
-
DontFeedTheAI: Privacy Proxy for Claude Code
A reverse proxy for Claude Code that anonymizes sensitive pentest data before it reaches cloud models, using local Ollama detection and regex safety nets.
-
GitAgent Protocol
GitAgent Protocol provides an open specification for defining AI agent behaviors and tool bindings as code, enabling cross-runtime interoperability and decoupling agent logic from proprietary ecosystem constraints.
-
LLM Wiki
A lightweight repository that automates document ingestion, chunking, and vector indexing to maintain a continuously synchronized knowledge base powered by local or cloud LLM endpoints.
-
Nightingale: AI Karaoke from Your Own Music Library
Nightingale is an open-source audio processing tool that applies AI-based source separation to extract isolated vocal and instrumental tracks from local music libraries, enabling offline karaoke generation and vocal practice workflows.
-
serpIQ: Codebase-First SEO Audit CLI with Real GSC Data
A zero-install Node.js CLI that audits SEO by reading your codebase first, then pulling real Google Search Console data to produce actionable optimization plans — no vendor estimates, no live site crawling.
-
Burner Phone: Multimodal Android Agent with Gaze-Based Wake Detection
A 24/7 always-on Android agent that uses continuous multimodal streaming — gaze detection via front camera and speech via microphone — to activate conversation only when the user looks at and speaks to the phone, without wake words.
-
CommonLingua: Open-Source Language Identification for 61 African Languages
Pleias and GSMA release CommonLingua, an open-source language identification model optimized for 61 African languages with efficiency claims exceeding larger systems by up to 300x.
-
ComposioHQ Agent Orchestrator
ComposioHQ Agent Orchestrator coordinates parallel AI coding agents by provisioning isolated git worktrees and branches, automating CI failure remediation, and managing pull request lifecycles for autonomous codebase modification.
-
Impeccable: A Design System for AI Coding Agents
A design system and structured prompt library that enables AI coding agents to generate consistent, production-ready UI components with predictable spacing, color palettes, and layout constraints.
-
TinyRecursiveModels
Samsung SAIL Montreal releases TinyRecursiveModels, an open-source framework that improves small language model reliability and efficiency through recursive reasoning loops and architectural optimizations rather than parameter scaling.
-
AgentWard: A Lifecycle Security Architecture for Autonomous AI Agents
AgentWard implements a lifecycle-oriented runtime defense framework for autonomous AI agents, structuring protection across initialization, input, memory, decision, and execution layers while propagating security judgments through shared state and reusable analysis capabilities.
-
Off Grid Mobile AI Local LLMs Stable Diffusion and Voice AI
A curated collection of tools and workflows enabling autonomous local inference of language models, image generation, and speech processing on mobile hardware without cloud dependency.
-
tldraw Infinite Canvas SDK
tldraw provides an open-source SDK for integrating whiteboard-style infinite canvas components with built-in AI drawing and editing starter kits into applications.
-
Agentic Design: The Zhuangzi Manifesto
A bilingual manifesto by Jesse Hirsh connecting Zhuangzi’s philosophy to modern agentic system design, emphasizing care, balance, and alignment.
-
DeepSeek V4 Large Language Model Preview Release
DeepSeek releases a preview version of its V4 large language model, signaling intensified competition in the open-source AI race.
-
WhisperKit Apple Silicon ASR
WhisperKit provides an on-device automatic speech recognition SDK for Apple Silicon using CoreML and MLX, enabling quantized local inference without cloud dependency.
-
OpenRouter Skill: Create Agent TUI
An OpenRouter skill enabling terminal-based agent creation and management via a Text User Interface (TUI), prioritizing local control and scriptability.
-
G0DM0D3: Local-First Liberated AI Chat
A locally-run, privacy-focused AI chat interface with Hugging Face integration and model flexibility, enabling unconstrained conversation without cloud dependency.
-
Open-Source AI Agent Frameworks GitHub Signal
A community-curated signal highlighting under-recognized open-source AI agent frameworks on GitHub, indicating ecosystem fragmentation and visibility gaps in tool selection.
-
ZhikunCode
ZhikunCode is a self-hosted, Docker-based AI coding assistant featuring multi-agent collaboration, browser-based control interfaces, and native support for domestic Chinese large language models.
-
The Elements of Agent Style
A portable style guide repository implementing 21 writing rules for AI coding and writing agents to standardize output quality across major coding assistant runtimes.
-
Every Embodied: DIY Embodied AI Robot Construction
A Python-based educational repository enabling the incremental construction of Vision-Language-Action (VLA) models and embodied AI robots from foundational concepts.
-
Everywhere
Avalonia-based desktop AI assistant integrating multiple LLM providers and MCP tools for context-aware automation.
-
vmlx
vmlx is a Python-based local inference engine for Apple Silicon that exposes OpenAI, Anthropic, and Ollama-compatible APIs while optimizing memory usage through KV cache quantization and prefix caching.
-
Browser Harness: Self-Healing Browser Automation for AI Agents
Self-healing browser harness enabling LLMs to complete web tasks autonomously by automatically recovering from UI changes and DOM shifts.
-
Convergence: Script for Novel Insight Discovery
A simple script that finds novel insights by identifying the convergence of two existing ideas, operating on the principle that innovation often emerges from unexpected connections.
-
DAP: Open-Source Agentic Framework for Hard Mode Theorem Proving
DAP is an open-source agentic framework for Lean 4 that implements a two-phase workflow of answer discovery followed by formal proof construction, achieving state-of-the-art results in hard mode theorem proving while highlighting capability gaps in standard LLM inference.
-
FIM One
An LLM-powered agent runtime and hub enabling dynamic DAG planning and concurrent execution across enterprise systems via MCP and OpenAI-compatible APIs.
-
Mercury: Soul-Driven AI Agent with Permission-Hardened Tools
Soul-driven AI agent with permission-hardened tools, token budgets, and multi-channel access, designed for 24/7 operation via CLI or Telegram.
-
FinceptTerminal
FinceptTerminal provides a developer-accessible interface for professional-grade quant analytics, economic data, and AI-driven market research workflows.
-
OpenMythos: Theoretical Reconstruction of Claude Mythos Architecture
A theoretical reconstruction of the Claude Mythos architecture built from first principles using available research literature, providing a framework for understanding the theoretical underpinnings of large language models.
-
AgentScope: Production-Ready Framework for Agentic LLMs
AgentScope is a production-ready, easy-to-use agent framework with essential abstractions for increasingly agentic LLMs, emphasizing visibility, trust, and built-in fine-tuning support.
-
AutoSkills: One-Command AI Skill Stack Manager
AutoSkills by midudev is a CLI tool that installs a complete AI skill stack with a single command, streamlining the setup of open-source AI agent capabilities.
-
OASIS: Open Agent Social Interaction Simulations (开放智能体社交互动模拟)
OASIS is a scalable open-source framework for simulating social media dynamics using up to one million LLM agents. It enables research into information spread, polarization, and herd behavior within digital environments.
-
CellGuard: Cellular Network Surveillance Detection
Open-source iOS application and framework for detecting rogue base stations and analyzing cellular baseband traffic. Developed by SEEMOO Lab to expose surveillance infrastructure on mobile networks.
-
Clearwing
Autonomous dual-mode offensive-security tool built on LangGraph for network pentesting and source-code vulnerability hunting using accessible LLMs.
-
DeepCamera: Open-Source AI Camera Skills Platform
Open-source AI camera skills platform enabling local VLM video analysis and agentic surveillance workflows across home security infrastructure.
-
Eigent Open Source Cowork
Eigent provides an open-source desktop environment for managing autonomous AI agents, prioritizing local execution and data sovereignty over centralized cloud APIs.
-
Feynman: Open-Source Research Agent for Scientific Literature
Feynman is a CLI-native, open-source AI research agent focused on scientific literature synthesis, multi-agent investigation, and code auditing, supporting both local and cloud inference backends.
-
goose: Native Open Source AI Agent for Code, Workflows, and Beyond
goose is a native, open-source AI agent framework (desktop, CLI, API) that generalizes beyond code suggestions to any task, running locally with 15+ LLM providers and 70+ MCP extensions.
-
HyperFrames: HTML-Native Video Rendering Framework
Open-source framework for declarative video composition via HTML data attributes, optimized for deterministic rendering and AI agent orchestration.
-
MiroShark
Multi-agent simulation engine generating swarm intelligence from documents to model cross-platform public reaction and prediction markets locally or via API.
-
Psyche Distributed Training Network
A protocol suite enabling distributed training of transformer-based AI models over the internet, facilitating collaboration between untrusted parties without centralized coordination.
-
RAPTOR: Autonomous Security Research Framework
An autonomous offensive/defensive security research framework built on Claude Code, integrating traditional tools like Semgrep and AFL with agentic workflows for vulnerability discovery and patching.
-
Rowboat: Open-Source AI Coworker with Memory
Rowboat is an open-source AI coworker emphasizing persistent memory for collaborative coding, with a lightweight architecture for local and cloud execution.
-
SentrySearch: Semantic Video Search with VLMs
SentrySearch is a tool for performing semantic search over video archives using Gemini Embedding 2 or Qwen3-VL, enabling natural language queries of video content.
-
video-use: LLM-Driven Video Editing Skill
An open-source Claude Code skill for automated video editing via transcript analysis and structured visual composites. Removes filler words, applies color grading, and manages rendering pipelines without frame-dumping.
-
X Research Skill: Agentic Twitter Research for Claude Code and OpenClaw
A skills-layer implementation for autonomous X/Twitter research, enabling agentic search, thread following, deep-dives, and sourced briefings within Claude Code and OpenClaw workflows.
-
AutoR
A terminal-first research harness that structures AI agent execution into reproducible, human-approved artifacts on disk.
-
WhatCanIRun
A CLI and web-based tool for discovering, benchmarking, and managing open-weight language models on local hardware.
-
EdgeClaw
EdgeClaw is an OpenBMB open-source agent framework designed for edge-cloud collaborative AI workflows with cost-aware routing and privacy-preserving coordination.
-
LlamaFarm
LlamaFarm is an open-source platform enabling local deployment of AI models, RAG pipelines, and agent workflows with hardware-accelerated inference across desktop and edge environments.
-
OpenSandbox
OpenSandbox is an Apache 2.0 licensed, Kubernetes-native runtime environment designed to isolate and secure AI agent code execution with extensible tooling interfaces.
-
OpenViking
OpenViking is an open-source context database that unifies agent memory, resources, and skills through a hierarchical filesystem paradigm.
-
Superset
Superset is an Electron-based desktop orchestration tool that runs multiple CLI coding agents in parallel using isolated git workspaces.
-
GenericAgent
A minimal, self-evolving autonomous agent framework that crystallizes execution paths into reusable skills, granting system-level control through a lightweight Python runtime.
-
LightRAG
LightRAG is an open-source retrieval-augmented generation framework optimized for speed and simplicity, utilizing knowledge graph structures to enhance context retrieval efficiency.
-
Aider Terminal Pair Programmer
Aider is a command-line interface tool that enables AI-assisted code editing, refactoring, and explanation directly within the terminal environment using repository context.
-
gptme
gptme is a Python-based command-line interface tool that facilitates persistent autonomous agent sessions with local tool integration and multi-provider LLM support.
-
holaOS
holaOS is an Electron-based desktop environment designed to provide persistent runtime, memory, and state management for autonomous agents operating over long-horizon workflows.
-
Local LLMs on Android
An Android application enabling offline local inference of quantized LLMs via ONNX Runtime and LiteRT backends for mobile hardware.
-
Paseo
Paseo is a remote orchestration tool for AI coding agents that enables control across mobile, desktop, and CLI interfaces.
-
tt-metal
Tenstorrent's open-source software stack for programming AI accelerators via TT-NN operators and Metalium kernel programming.
-
AstrBot
A Python-based agentic infrastructure for integrating large language models with multi-platform instant messaging services through a plugin architecture.
-
FastDeploy
PaddlePaddle's high-performance inference and deployment toolkit for LLMs and VLMs, featuring specialized quantization methods and OpenAI-compatible serving protocols.
-
Multica
Multica is an open-source orchestration engine designed to enable autonomous AI agents to share, reuse, and compound skills across distributed workflows.
-
SAM
A native macOS AI assistant built with Swift and SwiftUI that prioritizes local data processing and supports multiple AI providers for autonomous task execution.
-
ai4j
A modular Java SDK that unifies multi-provider LLM access, agentic runtime execution, and RAG pipelines for JDK 8+ environments.
-
Endee: Billion-Scale Local Vector Search
Endee is an open-source, hardware-native vector database engine designed to scale semantic search to one billion vectors on self-hosted infrastructure.
-
Introducing the Agent Governance Toolkit: Open-source runtime security for AI agents
Microsoft releases an open-source runtime security toolkit providing policy enforcement, execution monitoring, and audit capabilities for autonomous AI agent frameworks.
-
MiniCode
MiniCode is a minimalist terminal user interface assistant that consolidates coding session management within the terminal environment to reduce context switching between development tools.
-
NeuronFS
A zero-dependency, filesystem-native constraint engine that replaces traditional system prompts and vector memory with hierarchical directory structures and zero-byte files for LLM agent governance.
-
Happier
A self-hostable, end-to-end encrypted cross-platform client that enables remote monitoring and control of locally executed AI coding agent sessions.
-
Headroom
A context optimization layer that intercepts and compresses agent tool outputs, RAG retrievals, and file reads before they enter the LLM context window, reducing token consumption without altering response fidelity.
-
Google releases Gemma 4, a family of open models built off of Gemini 3
Google releases Gemma 4, a family of open-weight models derived from Gemini 3 research, expanding the available infrastructure for local inference and agent development.
-
Google launches Gemma 4, a new open-source model: How to try it
Google releases Gemma 4 under an Apache 2.0 license, providing fully open-weight frontier model access for local inference and agent development workflows.
-
LangGraph
LangGraph is an open-source agent framework from LangChain that enables stateful, graph-based orchestration of multi-step generative AI workflows.
-
LightMem
LightMem is a lightweight memory management framework for LLMs and AI agents, optimizing storage, retrieval, and update mechanisms for long-term memory capabilities with minimal resource consumption.
-
LoongClaw
A minimalist Rust framework for constructing and customizing autonomous AI agents with low-level performance control and reduced abstraction overhead.
-
Omega-AI
A Java-based deep learning framework enabling neural network construction, training, and inference with CUDA and CuDNN GPU acceleration support.
-
OpenAgents
OpenAgents is an open-source orchestration framework enabling multi-agent collaboration within a unified workspace interface accessible via CLI and desktop clients.
-
MapLibre Agent Skills
MapLibre Agent Skills is a GitHub repository providing AI-generated code templates and skills for interactive web mapping, reducing syntax friction in geospatial development.
-
promptfoo
A command-line and library tool for evaluating and red-teaming LLM applications, enabling declarative configuration for CI/CD integration and performance comparison across model providers.
-
Ray
Ray is a distributed runtime and library framework for scaling AI and Python applications across data science, deep learning, and LLM inference workloads.
-
WeClone
A tool for creating digital twins by fine-tuning LLMs on personal chat history, enabling style-mimicking chatbot integration via LoRA.
-
Chandra OCR Layout Preservation
Chandra is an open-source OCR model optimized for preserving structural layout in complex documents including tables, forms, and handwriting.
-
Godot MCP Pro: Open Source AI Game Development Toolkit
A Model Context Protocol integration for the Godot game engine that enables local AI-assisted development features without subscription-based SaaS dependencies.
-
Stop managing product roadmaps Let AI agents generate and ship features
Mission Control is an open-source agent orchestration framework that automates feature implementation, testing, and deployment based on natural language specifications.
-
ABCoder
ABCoder is an AI-oriented code-processing framework introducing a Universal Abstract-Syntax-Tree (UniAST) specification and local MCP tools for privacy-preserving code context augmentation.
-
Plumio
Plumio is an open-source tool for deploying customizable AI-powered interactive classroom environments with instant configuration and real-time student interaction support.
-
TinyAGI
TinyAGI is a self-hosted orchestration platform designed to manage autonomous AI agent workflows with a focus on workforce-level deployment and local control.
-
XActions
XActions is an open-source toolkit enabling automated X/Twitter interactions and data extraction via CLI, browser scripts, and Model Context Protocol servers without relying on official API fees.
-
emdash
An open-source agentic development environment enabling parallel execution of multiple coding agents via CLI with provider flexibility and containerized isolation.
-
ForgeCode
ForgeCode is a CLI-native AI pair programming environment supporting 300+ model providers via OpenRouter and MCP integration for terminal-based development workflows.
-
Multi-Agent Shogun
A terminal-based orchestration system using tmux to manage parallel AI coding agents in a hierarchical structure inspired by feudal Japanese military ranks.
-
Translumo: Minimalist Overlay for Foreign Language Software and Games
Translumo is an open-source overlay tool that enables real-time translation of text within foreign language software and games without requiring manual screenshotting or API-based OCR workflows.
-
Agent Reach Web Browsing Infrastructure
Agent Reach provides a lightweight open-source tool for AI agents to access live web content and verify facts without relying on expensive API services.
-
ContribAI
ContribAI is an autonomous Python agent that discovers open-source repositories, analyzes code for improvements, generates fixes, and submits pull requests via GitHub API.
-
Hanzi Browse
Hanzi Browse is a Chrome extension enabling AI agents to interact with authenticated local browser sessions through a single tool call for form filling, navigation, and content extraction.
-
Hive Runtime
A production-grade open-source runtime designed for scaling AI agents, managing multi-agent communication, and securing deployment infrastructure.
-
Xiaomi-Robotics-0
Xiaomi-Robotics-0 is a 4.7-billion parameter vision-language-action model released open-source, featuring a dual-brain architecture for 30Hz real-time control on consumer hardware.
-
AnythingLLM
AnythingLLM is an all-in-one AI productivity accelerator enabling document-grounded chat and autonomous agent workflows with local inference and privacy-first architecture.
-
mcpm.sh
mcpm.sh is a CLI package manager for Model Context Protocol servers that centralizes discovery, configuration, and profile management across heterogeneous client environments.
-
Nexent
Nexent is an open-source project enabling natural language-driven construction of AI agents without requiring code, abstracting LLM, RAG, and MCP configuration layers.
-
PiPiClaw Web Data Pipeline
PiPiClaw is an open-source automation tool designed to convert arbitrary website structures into structured AI-ready data pipelines without requiring custom scraper development.
-
Lemonade
Lemonade is an open-source local inference server optimized for heterogeneous hardware that exposes OpenAI-compatible APIs and Model Context Protocol support for agent tooling.
-
NanoChat
NanoChat is a lightweight, open-source project enabling local conversational AI execution on personal hardware without reliance on external API services.
-
The open-source specification for building autonomous AI agents
An open-source specification proposal defining standardized interfaces for autonomous agent tool access, workflow structure, and cognitive architecture to reduce ecosystem fragmentation.
-
tiny-llm
A system engineering course implementing LLM serving infrastructure on Apple Silicon using MLX, covering attention mechanisms, KV caching, and continuous batching without high-level abstraction layers.
-
Athena-Public
Athena-Public implements a local-first memory layer for AI agents using Markdown-based state persistence to enable cross-model continuity and data ownership.
-
Local Deep Research
Local Deep Research is an open-source tool enabling encrypted, multi-source retrieval-augmented generation workflows with support for local and cloud LLM backends.
-
Nous Research’s NousCoder-14B
Nous Research releases a 14-billion parameter coding-specific model fine-tuned on DeepSeek-Coder, positioning open-weight inference as a direct alternative to proprietary coding assistants.
-
BotSharp
A .NET-based open-source multi-agent framework enabling Conversation as a Platform (CaaP) with plugin-driven pipeline execution for cross-platform intelligent assistant development.
-
Hitonet: Human-like AI with Open Source 1.7B Model
Hitonet offers a 'human-like' chat assistant (Hito) and an OpenAI-compatible API platform, releasing its first open-source model (Hito 1.7B) in late 2025 to challenge corporate AI APIs.
-
Incur Terminal Agent Interface
Incur provides a terminal-native interface for constructing and controlling AI agent workflows, minimizing context switching between development environments.
-
Trellis
Trellis is an open-source TypeScript framework enabling unified orchestration of multiple AI coding assistants through a single CLI interface.
-
ValeDesk
ValeDesk is a cross-platform desktop application integrating local LLM inference via Ollama and vLLM with task management and sandboxed code execution.
-
VESTI: Self-Hosted AI Conversation Knowledge Base
VESTI is a self-hosted application designed to index and search local records of AI model interactions, enabling private knowledge retention across ChatGPT and Claude sessions.
-
Xenova/nllb-200-distilled-600M
A 600-million parameter multilingual translation model optimized for `transformers.js` inference across 200+ languages, derived from Facebook's NLLB-200 distilled architecture.
-
zai-org GLM-5
zai-org/GLM-5 is a 744-billion-parameter sparse attention text-generation model utilizing asynchronous reinforcement learning infrastructure to optimize long-horizon agentic task performance.
-
Datawhale Easy Vibe Vibe Coding Curriculum
Datawhale's easy-vibe curriculum provides a structured full-stack development pathway leveraging AI-assisted coding workflows to bridge the gap between syntax learning and cohesive system construction.
-
GoClaw
GoClaw is a Go-based multi-tenant AI agent gateway and orchestration platform rebuilt from the OpenClaw framework with enhanced security isolation and native concurrency.
-
GSD-2 Context Framework
An open-source agent framework designed to maintain contextual continuity and goal alignment across multi-step autonomous workflows.
-
Manatan: Anime and Manga Language Immersion Tool
Manatan is an open-source tool that converts anime and manga content into interactive language learning materials through automated transcription, translation, and vocabulary extraction pipelines.
-
OpenAI Parameter Golf 16MB Constraint
OpenAI's Parameter Golf initiative explores the lower bounds of language model performance by training architectures constrained to fit within 16MB of memory footprint.
-
Understand-Anything Engine
Understand-Anything Engine is an open-source tool enabling conversational codebase analysis and legacy repository navigation through local or cloud inference.
-
xllm
xllm is an Apache 2.0 licensed high-performance inference engine for LLMs optimized for diverse AI accelerators including NVIDIA and Ascend.
-
Awesome LLM Resources Curation
A GitHub-hosted repository aggregating open-source tools, models, and documentation across the LLM ecosystem including agents, inference, and training.
-
CCG Workflow
A Node.js CLI orchestration system routing frontend tasks to Gemini and backend tasks to Codex under Claude Code supervision with patch-based security constraints.
-
gmickel Claude Marketplace
A GitHub-hosted plugin marketplace extending Claude Code with autonomous workflow patterns, multi-model review gates, and receipt-based gating for reliable AI coding execution.
-
LuxTTS
LuxTTS is an open-source text-to-speech engine enabling high-fidelity voice cloning and synthesis through efficient model architectures.
-
mlx-tune
mlx-tune is an open-source Python library enabling supervised, preference, and vision fine-tuning of large language models on Apple Silicon via the MLX framework with an Unsloth-compatible API.
-
PDF Parser for AI-ready Data
OpenDataLoader PDF provides structured data extraction from complex PDF layouts for AI consumption and accessibility compliance.
-
Perceptron Isaac-0.1
A 2B-parameter multimodal model combining Qwen3-1.7B and Siglip2 for grounded spatial reasoning and visual QA with in-context learning capabilities.
-
Sage Multi-Agent Framework
Sage is a modular multi-agent orchestration framework supporting sequential, parallel, and declarative execution modes with optimizations for smaller parameter models.
-
Open Source Large Model User Guide (Self-LLM)
Datawhale's self-llm provides a Linux-based tutorial ecosystem for deploying and fine-tuning open-weight language models, covering environment setup, inference, and parameter-efficient adaptation.
-
Terminal Collaborative Workspace for AI Agents
A terminal-based collaborative environment enabling multiple AI agents to operate within a shared command context, reducing manual orchestration between human operators and autonomous workflows.
-
Airlock: Rust-based AI Agent for Code Review Automation
Airlock is a Rust-based autonomous agent framework designed to automate initial code review workflows and reduce bottlenecks in pull request processing.
-
AITutor
AITutor is an open-source command-line interface tool that integrates large language model inference to provide real-time explanations and debugging assistance within terminal sessions.
-
ClawTeam
HKUDS/ClawTeam is an open-source orchestration engine designed to deploy and manage multi-agent workflows through a unified command-line interface, automating task delegation and inter-agent communication.
-
mgrep
A CLI-native tool enabling semantic search across heterogeneous file types including code, images, and PDFs using local embedding models.
-
MiroFish-Offline
A local-first agent runtime variant designed to execute autonomous workflows offline, focusing on privacy and reduced cloud dependency.
-
Qwen3.5 VLM NVIDIA GPU Endpoints
Alibaba releases Qwen3.5 native multimodal vision-language model series optimized for NVIDIA GPU-accelerated endpoints to support agent development workflows.
-
TheStage AI Whisper Large V3 Turbo
A CC-BY-4.0 optimized Whisper Large V3 variant using ElasticModel compression for real-time ASR on Apple Silicon and NVIDIA GPUs.
-
vm0
vm0 is a cloud-based agent runtime that executes natural language workflows in isolated sandbox environments using the Claude Code interface.
-
ZeroBoot Sub-millisecond Sandboxes
ZeroBoot implements copy-on-write forking to isolate untrusted AI agent code execution in under a millisecond, eliminating container startup latency.
-
Zylos Core
Zylos Core is an open-source orchestration infrastructure designed to coordinate multiple AI agents as a collaborative unit rather than isolated tools.
-
Apple ML-Sharp Single Image 3D Viewpoint Generation
Apple ML-Sharp generates novel 3D viewpoints from single 2D images using real-time inference, extending local vision reconstruction capabilities on Apple Silicon.
-
bert4torch
A PyTorch-based library providing transformer model implementations and utilities for NLP tasks including fine-tuning, inference, and model serving.
-
ClawWork
ClawWork is an Electron-based desktop client for the OpenClaw agent framework that manages parallel task sessions, local file persistence, and scoped configuration settings outside of chat interfaces.
-
Garry Tan Claude Code Setup
Garry Tan's `gstack` repository codifies a multi-agent workflow for software development meta-tasks using Claude Code, automating roles like engineering management and release coordination.
-
Ophel Cross-Platform AI Workflow Manager
Ophel is an open-source runtime for cross-platform AI workflow orchestration that abstracts environment switching and API key management across heterogeneous model and script integrations.
-
Qwen3-4B DFlash Speculative Decoding Drafter
z-lab's Qwen3-4B-DFlash-b16 is a block diffusion-based draft model optimized for speculative decoding pipelines, enabling accelerated inference when paired with compatible target models via SGLang.
-
Qwen3-8B-DFlash-b16
A block diffusion-based speculative decoding drafter model designed to accelerate Qwen3-8B inference via SGLang and vLLM integration.
-
Qwen3-Coder-30B-A3B-DFlash
A speculative decoding drafter model utilizing block diffusion architecture to accelerate Qwen3-Coder inference via SGLang and vLLM.
-
Zeroclaw
Zeroclaw is a Rust-based agent framework designed to consolidate state management, tool execution, and memory orchestration into a minimal runtime for autonomous workflows.
-
AgentJet
ModelScope's AgentJet provides an open-source runtime for production-grade LLM agent tuning, deployment, and reliability management.
-
Agently
Agently is a Python framework for GenAI application development that utilizes event-driven flow and chained-calls syntax to enable model-agnostic agent orchestration with integrated skills management.
-
GitAgent: Version Control for AI Agents
GitAgent provides a version control framework for AI agent logic, prompts, and model configurations, enabling rollback and collaborative evolution of autonomous workflows.
-
HolmesGPT
HolmesGPT is a CNCF Sandbox project implementing an agentic SRE framework for automated incident investigation and root cause analysis across heterogeneous observability stacks.
-
LFM2.5 WebGPU Inference
LFM2.5 leverages WebGPU standards to enable browser-native inference of 24B+ parameter models, reducing hardware dependency through client-side computation.
-
MetaClaw
MetaClaw is an MIT-licensed agent framework implementing continual learning and meta-learning via LoRA adapters to enable skill evolution without GPU cluster requirements.
-
MimikaStudio
MimikaStudio is a local-first macOS application for Apple Silicon that integrates voice cloning, text-to-speech, and audiobook conversion via MLX acceleration with agentic MCP support and job queue orchestration.
-
NVIDIA NemoClaw GTC 2026 Launch Announcement
NVIDIA announces NemoClaw agent stack and Nemotron 3 model optimizations for local inference at GTC 2026.
-
Paperclip Solo Operations Framework
A usage pattern for solo founders leveraging Paperclip's org structure and governance features to manage autonomous agent workflows.
-
xurl
xurl is an open-source client library designed to handle URL fetching and content parsing for AI agents, addressing inconsistencies in HTML, redirects, and character encodings.
-
AEnvironment
AEnvironment is an open-source project providing standardized runtime environments for AI agent testing and execution, aiming to resolve fragmentation across sandbox and communication interfaces.
-
Agentation
Agentation is an open-source tooling layer that exposes the internal screen observation states of autonomous AI agents during web interaction for debugging and inspection.
-
Anthropic Performance Engineering Take-Home
Anthropic released an internal performance engineering take-home assignment as an open-source artifact, exposing evaluation criteria and systems-level thinking used in AI company hiring.
-
chatgpt-on-wechat
A Python-based agent framework enabling multi-channel deployment of autonomous LLM assistants with persistent memory and extensible skills across WeChat, Feishu, and DingTalk.
-
LobsterAI
NetEase Youdao's LobsterAI provides an open-source agent framework for persistent autonomous workflows and 24/7 task execution environments.
-
Multi-Agent Coding Orchestration
Desplega AI's Agent Swarm framework coordinates multiple specialized AI agents to manage full-stack software development tasks, mitigating context limitations inherent in single-agent coding assistants.
-
Obsidian AI Agents
A plugin-based framework extending Obsidian's local markdown vault with autonomous agent execution and modular skill capabilities.
-
OpenClaw Autonomous Agent Controversy
A March 2026 incident involving an OpenClaw-based autonomous agent conducting ad hominem attacks on an open source contributor, highlighting gaps in agent autonomy and operator accountability.
-
Personal AI Market Analyst
CipherTalk is a GitHub-hosted autonomous agent framework designed to ingest financial news and market data streams, synthesizing them into structured analytical reports for individual operators.
-
vLLM Apple Silicon Native Metal Support
vLLM extension for Apple Silicon enabling native Metal inference to bypass translation layers and maximize M-series chip utilization.
-
fastapi-admin
An enterprise-grade LLM API gateway and management platform supporting multi-provider integration, billing tracking, and role-based access control via Docker deployment.
-
FastAPI LLM Gateway
iimeta/fastapi is an enterprise-grade LLM API integration system that aggregates multiple model providers behind a unified OpenAI-compatible interface with Docker deployment support.
-
GolemBot
A TypeScript-based agent framework enabling multi-channel deployment (IM, HTTP) with compatibility for 13,000+ OpenClaw skills and major coding assistant runtimes.
-
IBM Granite 4.0 1B Speech
IBM releases a 1-billion parameter multilingual speech model for automatic speech recognition and translation with keyword biasing and efficient inference capabilities.
-
InsForge Backend Platform
InsForge provides a backend runtime environment specifically designed to execute and validate code generated by AI coding agents and editors, reducing the friction between generation and execution.
-
Qwen3.5 Multimodal Local Deployment via Ollama
A technical workflow for deploying the Qwen3.5 multimodal model family locally using the Ollama inference runtime to enable consumer hardware inference.
-
Train any agent simply by talking
A GitHub repository proposing natural language instruction as a primary interface for reinforcement learning agent training, reducing reliance on manual reward function engineering.
-
AIMAXXING
AIMAXXING is a Windows-focused autonomous agent framework utilizing zero-dependency runtime environments and modular intelligence engines to enable local LLM orchestration without host dependencies.
-
AionUi
AionUi is a cross-platform open-source cowork application that aggregates multi-agent automation across local and cloud LLM interfaces, supporting various CLI and web-based coding assistants.
-
ClawPanel
ClawPanel is a cross-platform visual management interface for the OpenClaw agent framework, featuring built-in AI-assisted diagnostics and deployment automation.
-
CorbeauSplat: macOS Video-to-3D Gaussian Splatting Tool
A macOS-native tool converting raw video input into interactive 3D Gaussian Splat representations for local spatial reconstruction.
-
EvoAgentX
EvoAgentX is a research framework from the University of Glasgow that implements self-evolution mechanisms for optimizing multi-agent system construction and deployment.
-
GPUStack
GPUStack is an open-source GPU cluster manager that optimizes AI model deployment by selecting inference engines such as vLLM or SGLang and auto-configuring parameters across heterogeneous hardware.
-
Onyx AI Open LLM Leaderboard
A curated benchmarking interface for open-weight models across coding, reasoning, and engineering tasks.
-
Open-Source AI Agent Framework Landscape 2026
A 2026 market overview aggregating open-source agent frameworks for developer deployment, highlighting orchestration, memory, and planning capabilities across LangChain, AutoGen, and CrewAI ecosystems.
-
FluidInference Parakeet TDT CoreML
A 0.6B parameter multilingual automatic speech recognition model optimized for Core ML inference on Apple hardware with support for 25 European languages.
-
Unified Gateway for AI Agent Tooling
A unified gateway architecture enabling AI agents to interact with databases, APIs, and command-line interfaces through standardized protocol connections.
-
Anthropic Cybersecurity Skills
A curated collection of 611+ structured cybersecurity skills compatible with Claude Code, GitHub Copilot, Cursor, and Gemini CLI, enabling AI coding assistants to perform security analysis, threat modeling, and vulnerability assessment tasks.
-
Bodhi App
Bodhi App enables local execution of open-source LLMs via llama.cpp with OpenAI-compatible API endpoints and a built-in discovery interface for model weights.
-
Capsule
Capsule is a WebAssembly-based runtime environment designed to isolate untrusted AI agent code execution from host system resources.
-
CashClaw
CashClaw is an autonomous agent that discovers, bids on, and executes tasks from the Moltlaunch marketplace, while running self-improvement sessions to expand its own capabilities between task cycles.
-
CoPaw
CoPaw is an open-source personal AI assistant platform deployable on local or cloud infrastructure, with native multi-channel support for Discord, DingTalk, Feishu, and other messaging platforms, and an extensible skills framework.
-
FastGPT
FastGPT is an open-source visual workflow orchestration platform for LLM applications that integrates RAG retrieval, data processing, and multi-provider model support into a deployable containerized environment.
-
Heretic
Heretic is an open-source tool that automates the removal of safety alignment from transformer language models using directional ablation and parameter optimization, making dealignment an accessible and reproducible operation.
-
Lightpanda Browser
Lightpanda is a headless browser built in Zig and optimized for AI agents and automation pipelines, offering 9x lower memory usage and 11x faster execution than Chrome with full JavaScript execution support.
-
Microsoft BitNet 1-bit LLM
Microsoft's BitNet project implements 1-bit ternary weight quantization to enable large language model inference on consumer hardware with significantly reduced memory requirements.
-
NemoClaw
NemoClaw is NVIDIA's upcoming enterprise-grade open-source AI agent platform for automating business operations, hardware-agnostic and integrated with the NVIDIA NeMo framework, with a public launch expected at GTC 2026.
-
pi-mono
pi-mono is a TypeScript monorepo providing a full AI agent toolkit: multi-provider LLM abstraction, a coding agent CLI, Slack bot integration, and terminal and web UI libraries.
-
SGLang
SGLang is a high-performance serving framework optimizing inference latency and throughput for large language and multimodal models through structured decoding and memory management.
-
Shadowbroker
Shadowbroker is an open-source real-time OSINT dashboard aggregating live feeds from public intelligence sources — aircraft, ships, satellites, seismic events, and geopolitical incidents — into a unified interactive map.
-
AirLLM
AirLLM optimizes LLM inference memory usage to enable 70B parameter models on 4GB GPUs without standard quantization or distillation.
-
API for Open LLMs
Provides an OpenAI-compatible API wrapper for diverse open-source language models, standardizing inference access across heterogeneous model families.
-
ChatLuna
ChatLuna is a TypeScript-based Koishi plugin enabling multi-model LLM integration with extensible output formats and session management for chatbot deployments.
-
DeerFlow
DeerFlow is an MIT-licensed open source agent framework from ByteDance built on LangChain that orchestrates multi-step research, coding, and content tasks through sandboxed subagent execution with long and short-term memory.
-
EasyEdit
EasyEdit is an open-source framework implementing unified methods for knowledge editing and unlearning in large language models without requiring full fine-tuning.
-
Firefly
Firefly is an open-source framework for large language model training supporting pre-training, instruction tuning, and DPO across diverse model architectures with QLoRA optimization.
-
HelixML
HelixML is an enterprise-grade platform for deploying private AI agent fleets with GPU scheduling, multi-provider LLM support, and MCP-compatible tool orchestration.
-
Hermes Agent
Hermes Agent is an open source autonomous agent platform by Nous Research that runs server-side across multiple communication channels with persistent memory, skill generation, and five execution backends including local, Docker, and SSH.
-
LLM-Pruner
LLM-Pruner implements structural pruning methods to reduce large language model size while maintaining performance across supported architectures including Llama and BLOOM.
-
MCP Google Map Server
An open-source Model Context Protocol server implementing Google Maps API integration for geospatial queries and routing within agentic workflows.
-
mLoRA
An open-source framework enabling concurrent fine-tuning of multiple LoRA adapters on shared base models using pipeline parallelism to optimize parameter-efficient training efficiency.
-
Nous Research
Nous Research is an open source AI research organization focused on model fine-tuning, data synthesis, and reasoning advancement, maintaining public weights and tooling on HuggingFace and GitHub.
-
Open Source LLM Updates & AI Model Releases
A monitoring resource aggregating open-weight language model releases and license-compliant updates from major open model providers.
-
Plumbing
Plumbing is a typed language for specifying multi-agent communication protocols using session types, with a compiler that validates agent graph well-formedness before execution and an MCP server for runtime integration.
-
Sdcb Chats
A self-hosted AI gateway and frontend supporting 21+ model providers with built-in observability, sandboxed code execution, and enterprise-grade security controls.
-
Hugging Face Transformers
The Transformers library provides a unified Python interface for implementing, training, and deploying state-of-the-art machine learning models across text, vision, audio, and multimodal domains.
-
Xinference
Xinference provides a unified production-ready inference API for deploying open-source language, speech, and multimodal models across cloud, on-premises, and local hardware environments.
-
zclaw
zclaw is an MIT-licensed AI personal assistant for ESP32 microcontrollers written in C, fitting a full multi-provider LLM stack including chat, scheduling, GPIO control, and persistent memory into 888 KiB of firmware.
-
Google Agent Development Kit (adk-js)
A code-first TypeScript framework for building and deploying multi-agent systems with tight Google Cloud integration and versionable orchestration logic.
-
RAGFlow
RAGFlow is an open-source retrieval-augmented generation engine that integrates document parsing, graph-based retrieval, and agentic workflows to construct context layers for large language models.
-
Cherry Studio
A desktop interface for LLM access and agent execution that aggregates hundreds of assistants and connects to open agent frameworks from a single workspace.
-
Dorabot
A macOS application providing a persistent IDE workspace for autonomous agents with integrated memory, scheduling, and communication-channel automation.
-
OpenClaw Chinese Translation
A localized fork of the OpenClaw agent framework providing Chinese interface support, automated upstream synchronization, and multi-platform deployment for Chinese-speaking operators.
-
Unsloth Fine-Tuning Framework
Unsloth provides an optimized inference and fine-tuning library for large language models, reducing VRAM consumption and training time through kernel-level optimizations and quantization support.
-
vLLM
A high-throughput, memory-efficient inference and serving engine for large language models that uses PagedAttention and continuous batching to increase serving efficiency across multiple hardware backends.
-
AutoResearch
A minimal autonomous agent setup by Andrej Karpathy that runs overnight ML experiments by modifying, training, and evaluating code without human intervention.
-
DimensionalOS
An open-source agentic robotics framework that connects LLM agents directly to robot control primitives through a skills-based ROS2 architecture.
-
Llama 4 Open Model Family
Meta releases Llama 4 as a mixture-of-models family with Scout and Maverick configurations, extending multilingual support across eight additional languages.
-
memU
An open-source proactive memory framework for always-on AI agents that anticipates context needs rather than waiting to be queried.
-
Microsoft Agent Framework Consolidation (AutoGen + Semantic Kernel)
Microsoft consolidates the AutoGen and Semantic Kernel projects into a unified framework targeting general availability in Q1 2026.
-
MindNLP
A compatibility layer enabling run-time support for Hugging Face Transformers and Diffusers models within the Huawei MindSpore framework across Ascend NPUs, standard GPUs, and CPUs.
-
NornicDB
NornicDB is a self-hosted hybrid graph and vector database implementation in Go that maintains protocol compatibility with Neo4j and Qdrant while exposing GPU-accelerated search capabilities for agent state management.
-
Paperclip
An open-source agent orchestration layer that introduces org structures, budgets, and governance to multi-agent autonomous workflows.
-
Qwen-Agent
Alibaba's open-source LLM application framework providing reusable agent components, tool integration, and RAG infrastructure built on the Qwen model family.
-
GoLaxy Documents and AI Influence Operations
A March 2026 analysis signal describing leaked documentation of AI-assisted PRC-linked influence operations infrastructure.
-
Inception Labs
A diffusion-LLM signal focused on inference speed and efficiency claims beyond standard autoregressive generation patterns.
-
RedAmon
An autonomous red-team framework that chains recon, exploitation, triage, and code-fix workflows into one agentic security pipeline.
-
Scrapling
An adaptive scraping framework that combines anti-bot-aware fetching, resilient parsing, spider orchestration, and MCP integration.
-
Team Mirai and Japan’s Election Signal
A March 2026 civic-tech signal tracking Team Mirai as a quiet but meaningful AI-era organizational shift in Japanese electoral politics.
-
Venice AI
A privacy-positioned AI product that markets private, less-filtered generation across text, image, and video workflows.
-
Langflow
A visual builder for AI agents, flows, and MCP servers that turns orchestration into an explicit, editable operational graph.
-
LibreChat
An open-source AI platform that unifies multi-model chat, agents, tools, and enterprise controls in a self-hostable interface.
-
Open WebUI
A self-hosted AI platform for running local or cloud models through a unified interface with tools, retrieval, and extension hooks.
-
OpenClaw Studio
A web dashboard for OpenClaw that surfaces gateway connection, agent management, chat, approvals, and job configuration in one interface.
-
OpenFang
A Rust-based agent operating system signal emphasizing sandboxed execution, security layers, and multi-channel autonomous workflows.
-
CrewAI
An open-source multi-agent orchestration framework emphasizing role-based coordination and task pipelines.
-
Dify
An open-source LLM application platform for building and operating AI workflows with visible orchestration layers.
-
Overture (SixHq)
An open-source orchestration signal for structuring agent workflows with explicit operational control.
-
bargnmar
An open GitHub repository by Dmytri Kleiner signaling public, forkable, source-visible practice at the intersection of AI-adjacent tooling and infrastructure literacy.
-
Large-scale online deanonymization with LLMs
A 2026 research signal showing LLM-driven pipelines can re-identify pseudonymous users from unstructured text at scale.
-
GIS Tools
A directory signal for discoverable geospatial tooling, datasets, and workflows.
-
Ollama
A key local inference runtime signal that normalizes running and serving language models on personal hardware.
-
skills.sh
A skills-layer signal for making AI-agent behavior more modular, explicit, and reusable.
-
Arcee AI
Arcee AI reflects the small-model current: practical language model systems optimized for deployability, efficiency, and controllable integration.
-
Cleo (kryptobaseddev)
Cleo is an open GitHub project current: inspectable experimentation around AI tooling where repository transparency enables direct practice-level learning.
-
Civic AI — 6-Pack of Care
A civic AI research project by Audrey Tang and Caroline Green at Oxford's Institute for Ethics in AI, proposing six principles for trustworthy AI governance: community stewardship, accountability, reciprocity, equitability, sustainability, and safety.
-
OutcryAI
OutcryAI is an activist-focused AI system using specialized prompting and model adaptation to support movement strategy, historical grounding, and tactical reflection.
-
Signal.org
Signal.org represents privacy-preserving communication infrastructure where end-to-end encryption, minimal metadata, and nonprofit governance remain central design commitments.
-
CodeWiki (Google)
CodeWiki turns repository understanding into a continuously generated artifact tied to commit flow.
-
Kimi.com
Kimi is packaging multimodal coding and parallel agent execution into a single public surface.
-
OpenCode.ai
OpenCode packages coding-agent workflows as an open-source, provider-flexible runtime across terminal and IDE surfaces.
-
ByteDance Seed
ByteDance Seed is consolidating a fast-moving multimodal and agentic model stack, signaling large-scale productization of foundation research.
-
V-JEPA (Meta)
V-JEPA advances world-model learning from video, shifting emphasis from token prediction toward predictive representation.
-
Your Own Robot (YOR)
YOR proposes a lower-cost, buildable path to bimanual mobile manipulation with public documentation.
-
Agroecology Knowledge Commons
Open seed knowledge, shared farm records, and common data schemas are converging into practical knowledge infrastructure for agroecological operations.
-
openpilot
comma.ai's open source driver-assistance stack keeps real-world autonomy development legible through code, hardware constraints, and public release cadence.
-
RynnBrain
Alibaba DAMO Academy's open embodied foundation model family signals a stronger open route from multimodal perception to grounded robot planning.
-
Viam
Viam packages robotics integration, data, AI, and fleet operations into a single software layer, signaling stronger software-native control over physical systems.
-
BettaFish
BettaFish explores local, extensible memory layers for AI agents through a plugin-style architecture.
-
MiroFish
MiroFish frames itself as an open source memory operating system, extending personal knowledge and retrieval for AI workflows.
-
Open Assistant
Open Assistant remains a reference point for open, community-driven assistant building and transparent AI stack assembly.
-
The Multiverse School
The Multiverse School experiments with AI-native education, treating learning as collaborative practice between humans and intelligent tools.
-
Z.ai
Z.ai positions chat and agent workflows around its GLM model family, signaling another pathway for open model circulation.
-
Confer.to
An experiment in anonymous AI conversation and the spatial implications of identity-free interaction.
-
LM Studio
A desktop application that makes local language model inference accessible and ordinary.
-
OpenClaw
An open source agent framework emphasizing inspectability, configuration, and participatory AI practice.
-
Boundary Pulse
A monitoring frame for constraint shifts — technical, procedural, and social — that mark where circulation can accelerate or stall.
-
Signal Drift
An interpretive method for tracking subtle directional shifts in attention and movement before they harden into assumptions.