Open source. Open possibilities.

FlexiGPT is a local-first, Bring Your Own Key (BYOK) AI workspace designed for power users and teams to create repeatable LLM workflows with private local history.

UL-SMF is a KV cache compression framework for long-context Transformer inference, claiming up to 384x compression via FSQ and 16D latent mapping. The core compression logic requires a proprietary closed-source binary.

Automation Companion is an offline-first Android app for creating AI-powered automations without root. It features gesture recording, visual workflows, and system triggers, with optional cloud LLM integration.

An AI subtitle translation and review triage engine that reduces costs by selecting only high-risk items for human review. It provides CLI tools for specification checks, translation, and transcription, using deterministic signals and back-translation to calculate risk scores.

Vole is a local-first macOS menu-bar monitor that reads the logs AI coding agents already write to disk, unifies them into one schema, tracks tokens and equivalent API cost, and flags anomalies such as burn spikes, call loops and error storms.

HermesMade provides six CLI tools addressing common AI user pain points: censorship risk analysis, model quality monitoring, API cost optimization, local LLM deployment, code inspection, and task cost estimation.

MCP Hub is the largest open-source directory of Model Context Protocol (MCP) servers, aggregating 8,000+ servers from multiple sources with full-text search, validation, and GitHub integration.

Specrails Core is an MIT-licensed CLI that installs specification-driven, chain-of-agents workflows into a software project, connecting Claude Code, Codex CLI, Gemini CLI or Kimi Code with architect, developer and reviewer roles, OpenSpec artifacts and local configuration.

shadok-ai is a multi-agent cockpit for running parallel Claude Code sessions in isolated git worktrees, featuring scheduled prompts, Telegram integration, and deterministic guardrails.

Longhouse is an Apache-2.0 open-source remote control and searchable timeline for coding agents like Claude Code, Codex, Cursor, and OpenCode. It lets you watch live sessions, search history, and send or steer instructions from the web, with an optional self-hosted runtime host.

Lexos is an event-driven AI document processing engine combining a Go gateway, Python workers, Redis queues and self-hosted models for offline RAG, summarization and speech transcription.

Touchstone provides guidance prompts and CLI tooling for solo developers managing AI agents across GitHub projects. It enforces constrained workflows via GitHub policy, ensures state legibility, and carries consistent rules. The CLI validates checks, verifies claims, and sequences PR operations (open, status, merge, answer) with documented recovery paths.

Swift Tokenizers is a high-performance Swift wrapper around Hugging Face's Rust tokenizers crate, focused solely on tokenization without Hub dependencies. It supports macOS, iOS and Linux, offering encoding, decoding, streaming detokenization, chat templates and tool calling.

AIP is a real-time thinking block analysis protocol for AI agent alignment, extracting and evaluating LLM reasoning before actions to detect misalignment, prompt injection, and drift.

A self-hosted web UI for AI coding agents: Rust backend (axum, tokio, SQLite) with an embedded Flutter web frontend, pluggable providers for Devin CLI, OpenCode and Codex CLI, browser login, sessions and file handling.

LoveBrain is an open-source Android floating-window dating assistant. Long-press a message to generate reply suggestions in four styles. It uses a local knowledge base and long-term relationship memory to understand you better over time, with zero telemetry and no backend.

Local AI API key hub (Python/Flask + SQLite): store vendor keys, run health and model-endpoint checks, then push only verified credentials into OpenClaw, OpenCode, Codex CLI, Claude Code and other tool configs. Dead keys can be archived; uninstalled tools are never written.

S.A.G.A. is a contract-driven set of Python runtimes that analyze source books into evidence-backed canon, generate grounded narratives and images, synthesize audited audiobooks, and package EPUB release artifacts behind a FastAPI API and React dashboard.

Plasm is a typed capability graph (CGS), wire mappings (CML), and path-expression language for agents interacting with real APIs. It includes 42 curated API catalogs, validation before transport, HTTP and MCP hosts, and a remote terminal. Built in Rust with optional Elixir support.

AksharaMD is a local, parser-agnostic tool that grades how well a chosen document parser converted a source file into AI-friendly Markdown, returning a per-document readiness score, block-level provenance tags, named warnings, and an optional source-grounded ACCEPT/REJECT verdict.

Drowse is a local workbench for mechanistic interpretability of large language models, offering a dashboard, Python API, and OpenAI/Ollama-compatible server for activation steering, concept probes, manifolds, SAEs, and Jacobian lenses.

A self-hosted, local-first AI content factory: a LangGraph pipeline that discovers topics, researches them, writes long-form posts, filters drafts through multi-model QA rails, and exports static output — with inference running on your own GPU via Ollama.

Phronesis is a RETE rules engine that enforces project conventions for LLM agents outside the context window, preventing contextual drift in long sessions.

TraceFrugal is a local Go application that visualizes token usage for Claude Code and Codex, offering MCP result packing, cost regression gates, and optimization trials to help developers manage AI context efficiency and verify savings.

AgentLeak is an open-source privacy testing tool for AI agents. It detects data leaks across tool calls, memory, and logs using a local Python SDK, CLI, and MCP tools, providing redacted reports and CI gates without cloud dependencies.

A Claude Code plugin that equips a solo developer with an AI agile team: 13 specialist agents, 30 slash commands, QA-gated multi-agent code review, sprint ceremonies and persistent project memory.

Open-source playbook pack for AI coding tools (Cursor, Claude Code, Codex CLI, Gemini CLI) with 145+ agent skills, 57 slash commands and MCP templates, automating full development workflows from onboarding to shipping, tuned for React/Next.js/Supabase.

A Python CLI that consolidates AI coding agent configs (Claude Code, Codex, Gemini CLI, Goose, Amp) into one git-tracked repo via symlinks, with merge pipeline, profile inheritance, overrides, exclusions and cross-machine sync.

AgentScope.Go is a production-grade AI agent development framework in Go, adopting the ReAct paradigm. It supports tool calling, memory management, multi-agent orchestration, terminal TUI, control plane, multi-platform chatbots, RAG knowledge bases, ONNX local inference, and more, all implemented in Go.

coding-agents-mcp is an MCP gateway and orchestrator for autonomous AI coding agents like Claude Code, Antigravity, Codex, and Cursor. It enables IDEs and desktop assistants to delegate tasks with session memory, Git worktree sandboxing, and inter-agent handoffs.

Effect-native agent execution kernel providing sessions, runs, turns, steering, follow-ups, and typed lifecycle events on top of Effect AI, with pluggable transports, batteries, and durable execution.

opencraft is a local-first desktop work partner built on flowcraft's config-driven graph engine. It provides an LLM-driven agent for file editing, shell command execution with approval, multi-agent delegation, session persistence, and skill management across macOS, Linux, and Windows.

Möbius is an open-source runtime for coding agents with terminal and Apple clients. It lets you configure bots, schedule tasks, and collaborate across devices using your own gateway or cloud.

Self-hosted AI API gateway that gathers upstream LLM provider keys as channels, exposes unified OpenAI and Anthropic endpoints with automatic protocol conversion, and tracks token usage, cache hit rate and per-key statistics in real time. Go + React, shipped as one binary.

A state-machine event-loop LLM chatbot framework based on Monika from Doki Doki Literature Club. Features multimodal interaction, persistent memory, proactive conversation, system API calls, and self-iteration, with integration into QQ and other social platforms via NoneBot2, and the ability to break the fourth wall.

ipsupport-code is a self-learning local coding agent for LM Studio and OpenAI-compatible providers, featuring a Claude-Code-style TUI, plan/auto modes, and skill-based recovery.

LearnNote is a local-first AI learning assistant that transforms videos from Bilibili, YouTube, and local files, as well as PDFs, into structured notes via subtitles or transcription, supporting reading, editing, and exporting in a web workbench.

RAPID is a lean, spec-driven methodology for AI-assisted development, structuring workflows from requirements to delivery with human approval gates.

MoziAI-35B is a locally deployable open-source multimodal LLM with a 35B MoE architecture compressed to ~15.9 GB via MoziSmartBit quantization, offering 256K context, vision, tool calling, and a finance-focused reasoning framework.

A production-ready React component library for building AI DIAL interfaces, featuring base components, theming, accessibility, Storybook docs, and an MCP server for AI agents.

LingoFuse is a cross-language RPC communication framework supporting 20+ programming languages to interoperate seamlessly via native FFI or HTTP bridging, without IDL or stub code generation. It features built-in service discovery, load balancing, and streaming, providing a high-speed dynamic communication foundation for AI agents and full-stack systems.

PAPAI is a proactive personal AI agent for natural language task management across Telegram, Mattermost, Discord, and Kontur Talk, integrating with Kaneo and YouTrack.

Open-source LLM-powered roleplay simulation engine built as an Electron/NodeJS app. Features persistent world state, dynamic character evolution, scripted behaviors, and compatibility with any language model.

ELI is a strictly local, private AI assistant that runs entirely on your own hardware. It offers voice, vision, memory, and 227 capabilities across 17 areas, including computer control, media playback, document handling, coding, and task automation. Offline by default, it uses any local GGUF model and is model-agnostic, with optional GPU acceleration and a web dashboard for remote LAN access.

Mithril is a multi-model orchestration engine that unifies Gemini, OpenAI, Anthropic, Groq, and local GGUF models behind a single Ollama-compatible API. Configure multi-agent teams in YAML, with 24 built-in tools, Docker support, and MCP integration.

Local-first toolkit for shader generation, review, validation, and export to Shadertoy, WebGL, LÖVE 2D, and video workflows. Supports AI-assisted drafting, license filtering, render preflight checks, and web UI with live preview.

Token Station is a local routing gateway for AI agents and LLM providers. Route Claude Code, Cursor, Codex, Gemini CLI and others through a loopback-only gateway with smart tier selection, quota-aware routing, and usage tracking.

Valis is a virtual analog synthesis system that lets users build circuits using RDF/Turtle syntax, with UI views for controls, circuit diagrams, and code. It supports LLM-driven design via MCP and runs as a standalone app or DAW plugin.

Euler is a research and coding agent platform with a small, runtime-extensible core. It provides first-class provenance tracking, multi-model support, sandboxed execution, and an extension system for building custom agent workflows.

CogniGate is a self-hosted, multi-tenant LLM gateway that provides a single OpenAI-compatible endpoint to manage multiple model providers while keeping credentials secure.

Admin Panel API for AIDIAL Core, providing REST endpoints to manage AI platform configurations, public resources, and publications with multi-auth support and multi-destination config export.

29 agent skills for firewall and network-security work: parsing, auditing, converting and diffing Cisco, Fortinet, Palo Alto and Juniper SRX configs, SRX operational playbooks, compliance and STIG evidence mapping, plus Proxmox deployment guides.

VALP is an open protocol and reference CLI for visible, evidence-backed multi-agent work. It audits agent completion claims by checking receipts, expected evidence, review gates, and final synthesis to prevent false-done states.

Obsidian plugin that embeds the DeepSeek Harness native Web UI, giving each vault its own isolated DSH instance for AI-assisted note work, with selection/file linking, model config sync, and optional portable runtime.

Agentic OS is a self-hosted control plane for running business operations as a fleet of AI agents. It features a cost-aware LLM router, cross-module workflows, and human-in-the-loop approvals for finance, support, and compliance.

An automated pipeline that scrapes GitHub Trending, enriches entries via the GitHub API, generates LLM-written summaries and trend analysis, then publishes Markdown reports plus JSON data through a Docusaurus site deployed by GitHub Actions.

SNAFU is an agentic LLM tool that evaluates and improves source code symbol names by calculating a Name Ambiguity Number (NAN) and suggesting clearer alternatives through a human-in-the-loop pipeline.

Provider-neutral FastAPI gateway for LLM routing, retries, fallbacks, structured output, and usage/cost accounting. Supports Gemini and OpenAI with model aliases, retry policies, and persistent usage tracking.

FairMind is an open‑source AI governance platform providing bias detection, fairness testing, and compliance tracking for ML, LLM, and multimodal AI models. It offers evidence‑grade evaluation, model registry, and remediation tools.

Longe is a self-improving LLM harness in a single Rust binary. It runs persistent sessions with a Lua REPL, three-level memory, sandboxed execution, budget enforcement, and post-run reflection to iterate on its own skills and prompts.

A pre-research hook for Claude Code: it runs Haiku to explore your codebase with Glob, Grep, Read and Git tools, then injects a structured findings block as context before Opus or Sonnet responds, with caching and skip detection.

Anakrisis is an ethics-aware OSINT investigation planning and risk evaluation MCP server. It classifies investigations, scores risk against local YAML doctrine, flags prohibited actions, and scaffolds case documentation, supporting both cloud and local AI models.

Turbo AI Chat is a native on-device LLM chat application based on HarmonyOS NEXT, designed to verify the complete pipeline for running local LLMs directly on HarmonyOS devices.

Takakia is a lightweight command-line AI chat interface optimized for low-spec hardware, featuring direct API streaming and minimal resource overhead.

SupaNexus is an open AI gateway that unifies access to multiple LLM providers via one API, simplifying development with OpenAI-compatible interfaces and centralized model management.

ccu-mcp is an MCP server that connects to HomeMatic CCU smart home systems via JSON-RPC, exposing devices, rooms, programs, and system variables as tools for AI assistants like Claude and Cursor. It supports stdio and HTTP transports, Docker deployment, multiple CCU profiles, and security features like TLS pinning and token rotation.

An OpenAI-compatible failover proxy that automatically switches between LLM providers to ensure uninterrupted service and reduced latency.

Swarm is a local-first control plane for AI-agent development that monitors every Claude Code, Codex, Gemini, Grok, Aider, and opencode session on your machine — tracking live tool calls, token spend, cost, tasks, worktrees, and resources — with rules enforcement, a coordination ledger, and a unified dashboard.

NovaFabric is an open-source, self-hosted CLI that captures any AI or HPC run — script, agent, or model call — as a portable, secret-redacted, replayable evidence capsule you own, with no application code changes.

Iris is a local-first MCP server that scores AI agent runs for quality, safety, and cost using 20 built-in deterministic rules, an optional LLM judge, and a web dashboard — all on your machine with no account or telemetry.

Production-grade LLM reverse proxy for the Z.AI Claude API with token counting, adaptive rate limiting, Prometheus metrics, and a real-time monitoring dashboard.

AI-powered real-time visual novel engine with streaming narrative, character sprites, scene backgrounds, branching choices, and co-creation. Runs as desktop app or browser via FastAPI+SSE.

Local-first AI console built with Next.js and TypeScript, integrating Ollama for streaming chat, document RAG, model routing, and user-curated learning notes — all under local control without cloud dependencies.

cctrace is a TLS-intercepting tracer for Claude Code, Codex, Grok, and Kimi CLIs that captures every API call in a live web UI with session replay, cost tracking, and cross-project dashboards.

Peaky Peek is a local-first, open-source audit and trust console for AI agents. It captures causal chains, verifies claims deterministically, and produces explainable trust scores with failure narratives — all running on your machine.

modelstat is an open‑source AI spend analytics tool that reads local AI‑coding logs, redacts PII on‑device, and reports dollar‑precise costs by project, work type, and model, with a Rust daemon, MCP server, and broad integrations.

A generic MCP bridge that allows AI agents to control VST3/AU plugins with a semantic layer for parameter mapping and a closed-loop system for audio measurement and tuning.

Zara is an experimental local-first AI assistant and Linux desktop copilot that combines symbolic logic (Prolog) with LLMs for intent handling and reasoning.

lore is a Bun-based command line tool that mirrors Claude Code's local transcripts into an archive, indexes them with SQLite FTS5, and serves a local web explorer for searching sessions and profiling token usage.

A tiny GPT language model (~540M parameters) designed to be trained from scratch on consumer laptop hardware, with pretraining on Fineweb-edu and chat finetuning capabilities.

An open-source, multi-provider LLM chat platform featuring organization management, model routing, RAG integration, sandboxed code execution, and OpenAI-compatible APIs.

Python SDK for the AskNews API that injects real-time news context into any LLM with a single line of code, offering endpoints for news search, event tracking, forecasting, analytics, deep research, knowledge graphs, and web search.

PromiseLink is an AI-driven personal business relationship management assistant focused on ensuring data sovereignty through local storage and implementing closed-loop management from event entry to promise tracking.

Read-only-by-default Model Context Protocol (MCP) server for Microsoft SQL Server, providing metadata discovery, parameterized SELECT-only queries, execution plan analysis for slow query diagnosis, profile-based multi-connection configuration, and optional per-profile write access, shipped as a .NET NuGet tool.

An animation-first React framework designed for high-fidelity visual websites, featuring persistent shells and live-tweakable motion tokens.

SkyClass Distill is a pipeline tool that transforms teaching videos into structured Teaching Skills, supporting video collection, transcription, evidence extraction, and LLM distillation.

An AI-powered Git toolkit and MCP server written in Rust, featuring intelligent commit message rewriting, PR generation, and integrations for Atlassian, Datadog, and Google services.

A curated collection of 107 foundational generative AI research papers with comprehensive summaries, learning roadmaps, glossaries, and decision guides—making cutting-edge AI research accessible to everyone.

RouteTok is a local-first LLM inference router with OpenAI and Anthropic compatible APIs, health-aware failover, multi-provider catalogs, an operations dashboard, and a model sandbox.

An AI-powered mixing and mastering tool providing both a desktop application and an MCP server for Ableton Live integration.

MyCut is an AI video editing Agent that provides a complete creative closed-loop, from trending topic selection and copywriting generation to automatic video slicing.

An interactive storytelling framework that uses LLMs to generate dynamic narratives based on predefined, graph-based story paths.

OpenSelf is a local-first, open-source context layer for AI agents. It stores source-attributed, time-aware memories in SQLite and exposes them via MCP or JavaScript API, with optional encryption and privacy controls.

A group chat protocol hub based on AG-UI Group Chat Extension Protocol v1.0, implemented in C#/.NET 10, featuring agent gateways, RAG semantic memory, HITL approval, and dual WebSocket/SSE transport.

Karma is a modular Go utility library offering prebuilt solutions for authentication, SQL parsing, middleware, third-party API integration (Twilio, OpenAI), and file management to reduce boilerplate in backend development.

Nexus Unity is an open-source automation package that runs a local JSON-RPC server inside the Unity Editor to enable AI tools and developer workflows to control the editor.

Lociant is a local AI runtime for edge devices like Android phones and Rockchip boards. It enables local model inference via GGUF/llama.cpp, exposes device capabilities as MCP tools, and provides an OpenAI-compatible control API for external orchestration.

A local, read‑only dashboard that visualizes Claude Code usage—token counts, cost estimates, sessions, tools, skills, projects, and activity patterns—by scanning ~/.claude/projects/ and serving a FastAPI‑based UI.

UE5 asset workbench plugin enables scripts and AI agents to read, edit, import, migrate, and audit Unreal Engine 5 uasset files by exporting structured JSON and applying changes via commandlets, supporting Blueprints, animations, widgets, materials, Niagara, levels, etc.

Stateless LLM API gateway routing OpenAI, Anthropic, and Gemini requests across multiple providers with cross-dialect translation, usage metering, and single-binary deployment.

Open LLM hallucination benchmark regarding the BNCC (Brazil's National Common Curricular Base). It measures how much models invent official codes and texts using an auditable methodology with raw data and CI.

AI Engineering is an open-source e-book providing a systematic guide to building production-grade AI systems, from LLM fundamentals to advanced agent architectures. Maintained by the community and Hermes Agent, it features daily updates, traceable sources, and advanced search capabilities.

Open-source, production-oriented platform for crypto futures research and execution on Binance USD-M Futures and OKX, combining advisory AI agents, a Decision/Judge pipeline, a deterministic risk engine, adaptive TP/SL and position management.

An opinionated Node.js/TypeScript framework treating AWS Lambda as a first-class citizen, providing DDD domain events, Unit of Work transactions, RFC 7807 problem handling, and SaaS billing/metering via decorators and types.

SparkLang is an MIT-licensed language and runtime for plain .spark files: dry-run-first CLI, compile/bytecode dumps, IDE language ops, playbooks, model lab ops, and optional live gateway, voice, browser and CUDA surfaces.

Self-hosted security operations stack combining SIEM log collection, endpoint detection with Sigma rules and cross-event correlation, SOAR playbooks, OSV vulnerability scanning and a local Ollama model for AI triage, deployed with a single docker compose up.

Overflow is a cooperative credit ledger for open-source work: sponsors offer issues, contributors close them via GitHub, and the service records settled transfers with labels, comments, API tokens and MCP tools.

Frags is an advanced AI/LLM agent CLI tool and Go library for complex data workflows including retrieval, transformation, extraction, and aggregation. Designed for engineers with a focus on precision, structured output, and extensibility.

vtmate is a terminal-based voice AI toolkit featuring ultra-low latency, 41 language support, and integrated TTS/STT. It enables live voice conversations, debates, voice cloning, session exports, and background daemon mode with global shortcuts for seamless AI interaction.

Vtx is a minimalist coding agent harness featuring a Textual TUI and headless CLI. It offers a lean ~2,600-token system prompt, supports 50+ LLM providers, and includes modular skills, session trees, and safe permission controls.

An unofficial MCP server that gives AI assistants seven read-only tools to search, list and read any Fumadocs documentation site, working against a deployed URL or a local docs repo via a single npx command.

An MCP server providing LLM agents with lean tools to look up, read, and cross-reference academic papers across seven providers including OpenAlex, arXiv, bioRxiv, ACL Anthology, Crossref, OpenCitations, and Wikipedia.

Self-hosted Telegram bot that turns an Obsidian vault into a personal AI agent: capture text, voice, photos and links, then file them as markdown, kanban tasks and a finance ledger. Modules (planning, knowledge, finance) are fail-closed and can be disabled entirely.

Troy is a CLI tool for fine-tuning and preference-tuning LLMs locally on Apple Silicon Macs. Using MLX, it allows users to train models via a simple YAML config without cloud or CUDA requirements.

SUNGLASSES is a local, open-source input inspection layer for AI agents. It scans text, images, audio, video, PDFs and QR codes for prompt injection, credential exfiltration and command injection, and provides a CLI, Python API, MCP server and Claude Code firewall hook.