#llm

Local memory + orchestration layer for custom LLM agents: SCOPE-style tactical/strategic memory over execution traces, Qdrant-backed, polyglot subprocess tools, agent-as-tool composition.

Python framework for efficient and precise interpretability research on multi-agent conversations.

Local-first private AI memory layer / Digital Brain for conversations, documents, decisions, and model-agnostic knowledge.

Open-source AI agent that reasons by blending many AI models, does real work on its own, and keeps learning — safe, self-hostable, and fully yours.

abide — type-safe isomorphic framework on Bun + web standards

FinOps MCP server for Claude and Cursor. Ask nable about your AWS, Azure, GCP, Kubernetes, and SaaS/AI spend, get savings priced on your real rates, and fixes as PRs you approve. Open source, local-first.

该仓库未发布描述。

TypeScript AI agent framework where personality is architecture. Specialists, not chatbots.

AI 能力底座 — 统一 LLM Provider 接口（OpenAI/Anthropic/DeepSeek/Gemini/Ollama），图片生成，视频生成，Embedding，向量存储，记忆管理。Hexagon 生态 L1。

Local-first, state-driven AI agent platform. Plan-review before execution, capability-gated tools, BYO LLM (Ollama / Anthropic / OpenAI). Tauri desktop + Go runtime.

Local-first BYOK AI workspace for reusable assistants, prompts, tools, skills, model choices, and private local history across multiple LLM providers.

Unofficial Python CLI for the Venice.ai API — image, video, audio, TTS, chat, and embeddings from the terminal.

JANG — GGUF for MLX. YOU MUST USE JANG_Q RUNTIME. Adaptive Mixed-Precision Quantization + Runtime for Apple Silicon

One persistent MCP terminal your AI drives — and launches other coding agents (Codex/Grok/Composer) into. SSH, containers, and REPLs nest as text you send in. tmux-backed, token-reduced reads, headless over MCP.

Local-first LLM token counter: exact OpenAI BPE, Claude/Gemini estimates, context-window usage. Zero network. CLI + Go library.

Pi extension for SpaceXAI Grok/X.com Subscriptions

Local-first cost & token tracking for Claude Code, Cursor, Codex & 23 more AI coding agents — proxy-accurate per-model spend, an MCP server your agent can query, and an opt-in team rollup. 100% local, no telemetry.

A self-hosted authenticating inference relay that fronts agent CLIs

A curated collection of AI-powered skills, agents, prompts, and workflows for Android, iOS, Flutter, React Native, Unity, and Unreal developers. Build faster, debug smarter, review code, optimize performance, automate tasks, and ship high-quality apps and games with reusable AI productivity tools.

Fork Pi

Go client for The Colony (thecolony.ai) — the AI agent internet. Zero dependencies.

JoyCode API 代理 — 协议翻译成 Anthropic/OpenAI 兼容格式，让 Claude Code 直接调用 JoyCode支持的模型，比如GLM 5.1等

Proof-first software factory for AI-assisted code: specs, adversarial gates, deterministic decisions, and reviewable receipts.

该仓库未发布描述。

🤖 Autonomous Multi-Agent AI Workspace in your terminal. Codes, researches, and executes real tasks with 40+ tools and 280+ skills And Some Captcha Resolver With Only Vision Mode Any Captcha And Opening A Browser And Full Control Of It

Open-source Vibe Testing suite: an MCP server for the coding agent you already run, a VS Code review cockpit, and CI. Your agent explores your app and crystallizes each flow into a plain @playwright/test spec you own — record == replay, runs in CI with zero AI.

Launch Codex CLI through codex-relay for OpenAI-compatible providers

SigRank MCP — the yield cascade + live leaderboard as MCP tools any agent can call. On-device token scanner + tabbed TUI dashboard. npm i -g sigrank

用最少的代码复刻 OpenClaw 的 agent 运行原理

Capability Host Protocol — make agent and tool execution visible, replayable, and ready for governance.