Trending open-source AI

DSH Go is an unofficial Android client for DeepSeek Harness. It provides a chat interface and can run a DSH engine on the phone through a built-in Linux environment, without root or Termux, or connect to DSH on a computer or server over SSH or DSHarness Remote. Users can switch models, send images, use voice input, handle tool approvals, and share or replay conversations through encrypted links. It also offers optional phone-control features that read the screen and tap or type when permissions are enabled.

• Runs a DSH engine on-device in a built-in Linux environment without root or Termux
• Connects to remote DSH hosts via SSH or DSHarness Remote
• Chat supports model switching, images, voice input, and tool approvals
• End-to-end encrypted conversation share links and browser replay

★ 12kDartAGPL-3.0since Oct 25, 2026

Mika is an open-source Android app that acts as an AI assistant for your phone. You describe a task in plain language and it opens apps, taps, types, scrolls and browses the web on a hidden background screen, so you can keep using the phone normally. It works with any app or site without per-app integration, and can run scheduled or conditional tasks. It needs your own model API key or a provider sign-in, and it keeps data on the device.

• Operates any Android app by reading the screen and tapping, typing or scrolling
• Built-in desktop-class browser with web tasks and uploads/downloads
• Schedules and conditions such as recurring reminders
• Linux sandbox per conversation for Python, Node and shell on your files

★ 7.3kKotlinApache-2.0since Oct 29, 2026

magpie is a local gateway and menu bar app that lets coding agents such as Claude Code, Codex and Gemini CLI run on any model you have access to. It rewrites each agent's own config so you can pick a model per agent, use one API key or subscription everywhere, and move to another account or model when a quota runs out. It also reports token usage and cost, and can mask secrets in prompts before requests leave your machine.

• One local gateway at 127.0.0.1:3425 that translates between OpenAI, Anthropic and Gemini protocols
• Per-agent model picker that edits each agent's config file in place, with profiles for saved setups
• Routing groups with smart, order, rotate, usage and pace failover across keys and accounts

★ 31kGoMITsince Sep 23, 2026Website

Agent Autoloop is a lightweight local automation hub for command-line AI agents such as Claude Code and Codex. It routes scheduled tasks or external events to an agent based on rules, lets the agent exit after each run, and sends an approval card through chat or desktop notifications only when a decision is needed. After you answer, it resumes the original session and continues. It runs on macOS or Linux, uses plugins for event sources, agents, channels, and delivery, and has a zero-dependency kernel.

• Rule-based dispatch of events to command-line agents
• Approval cards through Feishu, DingTalk, Slack, Discord, or desktop notifications
• Plugins for event sources, agents, notification channels, and CI/CD delivery
• Resumes the original agent conversation after a human answer

★ 3.5kTypeScriptMITsince Nov 3, 2026

AnyModelBot is an open-source, local-first AI agent app for Windows, macOS, and Linux. It runs persistent bots, each with its own persona, default model, workspace, and memory. Users chat in the app or through messaging channels, and bots can read and write files, run commands, and control the desktop, asking for approval before going outside their workspace. Models come through Modelflare or the user's own provider API keys, with usage billed per token.

• Multiple bots with editable personas, default models, workspaces, and memory
• File read/write and sandboxed command execution with approval cards
• Connections to WeChat, Feishu, Telegram, Slack, Discord, and Notion
• Background tasks, scheduled rules, and Modelflare or official API keys

★ 2.7kPythonMITsince Oct 27, 2026Website

A set of skills, subagents, and a Python CLI that helps AI coding agents identify a PC game's engine, choose a modding route, inspect code, build a small mod, generate art and sound, test it in the running game, and package it. It includes field notes where agents record working versions, routes, verification steps, and fixes so later agents can reuse them. It is for developers and modders who want an agent-assisted workflow for single-player game mods.

• Agent skills and subagents for a staged modding workflow
• um scan fingerprints game engines, loaders, anti-cheat, and saves
• Asset generation commands for fal, ComfyUI, Sprixen, and Voidless
• Knowledge base of field notes with verification and gotchas

★ 9.5kPythonMITsince Sep 30, 2026

REA is an MCP server that connects AI coding agents to reverse-engineering tools. An agent can inspect native binaries, JavaScript and Electron apps, .NET assemblies, Android packages, firmware and websites, then report findings together with the evidence and limitations behind each conclusion. Analysis runs locally, and the same tools are usable from a terminal. It suits developers who want to understand how an existing app implements a feature and build their own version, and teams turning recorded app traffic into reinforcement-learning environments.

• Native binary pseudocode, assembly, strings and symbols through Hopper, Ghidra, IDA or Kuna
• Static JavaScript and Electron inspection of modules, imports, IPC and source maps
• .NET assembly metadata and CIL instruction analysis without running the code
• Exports recorded API requests and responses as tools for agent RL environments

★ 436kTypeScriptMITsince Apr 13, 2026Website

ECC is an installable plugin and toolbox that gives AI coding agents a structured engineering workflow. It adds agents, skills, commands, hooks, rules, and memory so an agent can plan before building, verify changes with tests, review its work, and reuse successful patterns. It supports Claude Code, Codex, Qoder, and adapters for other coding harnesses. AgentShield scans prompts, hooks, MCP configuration, permissions, secrets, and agent files. It is aimed at developers who use coding agents and want a consistent process instead of rebuilding one in every prompt.

• Guided setup for Claude Code, Codex, and Kimi Code
• Agents and skills for planning, review, security, architecture, and domain work
• Hooks, memory, rules, and continuous learning
• AgentShield scans prompts, hooks, MCP config, permissions, secrets, and agent files

★ 298kJavaScriptMITsince Jan 17, 2026Website

DeepSeek Harness (dsh) is an open-source agent harness from DeepSeek AI, built on an everything-is-a-plugin architecture and powered by Cordis. It runs locally: install Node.js, launch from npm, and the command starts a Web UI and opens it in your browser, with an option to skip opening the browser or use an SSH launch. Developers can also build and run it from a repository checkout. The project is a developer preview, so breaking changes are expected, and it is MIT licensed.

• Everything-is-a-plugin architecture
• Runs from npm with a local Web UI
• --no-open flag to skip opening the browser
• Build and run from a repository checkout

★ 274kTypeScriptMITsince Aug 13, 2026Website

Hermes Agent is an open-source AI agent from Nous Research that runs in a terminal or through messaging apps like Telegram, Discord, Slack, and WhatsApp. It keeps memory across sessions, creates and refines skills from completed tasks, searches past conversations, and can run scheduled jobs unattended. It works with many model providers and terminal backends such as local, Docker, SSH, and serverless sandboxes. Developers who want a persistent, self-hosted assistant on a server rather than tied to one laptop are the intended users.

• Built-in learning loop that creates and improves skills from experience
• Persistent memory and full-text search over past sessions
• Single gateway connecting Telegram, Discord, Slack, WhatsApp, Signal, and CLI
• Cron scheduler for unattended tasks with delivery to any platform

★ 261kPythonMITsince Jul 22, 2025Website

Ponytail is a skill and prompt set you install into an AI coding agent such as Claude Code, Codex or Cursor. It pushes the agent to write only the code the task needs, reusing what already exists, while keeping validation, error handling, security and accessibility. It also adds commands that review a diff, audit a whole repository, collect deliberately deferred shortcuts into a ledger, and show a measured scoreboard. Developers using AI coding assistants who want smaller, better-checked changes would use it.

• Intensity levels: lite, full, ultra, off
• Review command covering bugs, security, load and missing tests on a diff
• Repo-wide audit that ranks findings by importance
• Claude Code status band showing task cost, tokens and agent time

★ 184kJavaScriptMITsince Jun 11, 2026Website

CC Switch is a manager for AI coding and agent tools such as Claude Code, Codex, Gemini CLI, OpenCode, and others. It lets users switch API providers in one click and manage MCP servers, Skills, and Prompts in a single place, so they do not have to hand-edit JSON, TOML, or YAML configuration files. It is aimed at developers who use multiple AI command-line or desktop tools and want centralized configuration and provider switching.

• One-click API provider switching
• Manage MCP, Skills, and Prompts in one place
• Avoids hand-editing JSON, TOML, or YAML config files
• Supports Claude Code, Claude Desktop, Codex, Gemini CLI, OpenCode, and more

★ 152kRustMITsince Aug 4, 2025Website

Graphify is a command-line tool and assistant skill that builds a knowledge graph from a software project so you can query it instead of searching files. It parses code locally with tree-sitter and can include docs, PDFs, images, and media through a semantic pass. Developers using AI coding assistants can ask plain-language questions, trace connections between concepts, and see whether each link was extracted or inferred. It supports many assistants and platforms.

• Local tree-sitter code parsing with no LLM
• Query, path, and explain commands over the graph
• EXTRACTED or INFERRED confidence tags on edges
• Supports docs, PDFs, images, and video/audio inputs

★ 143kPythonApache-2.0since Apr 3, 2026Website

Caveman is a tool you install on your machine that changes how coding agents such as Claude Code, Codex, and Gemini CLI use tokens. It makes agent replies short while leaving code, commands, and error messages untouched, shrinks logs, JSON, CSV, and test output before the model reads them, and picks the model and effort level for each request. It also keeps scripts the agent writes so they can be reused. Developers use it to cut agent cost and time, and it adds commands for commits, reviews, and session token stats.

• Short agent replies with code and commands kept exact
• Compresses logs, JSON, CSV, and test output before the model reads them
• Auto mode picks the model and effort for each request
• Blocks keep agent-written scripts in the repo for reuse

★ 118kGoApache-2.0since Apr 4, 2026Website

Orca is a desktop application for running multiple command-line AI coding agents side by side. Each agent works in its own isolated git worktree, so you can send one prompt to several agents, compare their output, and merge the best result. It bundles split terminals, a built-in editor, GitHub and Linear views, diff annotation, and a CLI for scripting workflows. A mobile companion app lets you monitor agents remotely. It is aimed at developers who already use terminal-based coding agents and want one place to manage them.

• Parallel git worktrees for running several agents at once
• Mobile companion app for monitoring and steering agents remotely
• Design Mode captures HTML, CSS and a cropped screenshot into a prompt
• Orca CLI commands for scripting worktrees, snapshots, clicks and fills

★ 117kTypeScriptMITsince Mar 16, 2026Website

Claude-Mem is a memory system for Claude Code and other coding agents. It uses lifecycle hooks to capture tool usage observations during a session, generates semantic summaries, and stores them in a local SQLite database. A worker service provides search endpoints and a web viewer, while MCP tools let the agent search an index, view a timeline, and fetch full observations. Developers use it to preserve project context across sessions and retrieve past work with token-efficient, layered lookups.

• Persistent memory across sessions
• Lifecycle hooks capture tool usage observations
• MCP search, timeline, and get_observations tools
• SQLite storage with Chroma hybrid search

★ 114kTypeScriptApache-2.0since Aug 31, 2025Website

Paperclip is an open-source Node.js server and React UI for coordinating multiple AI agents as a team. Users bring their own agents or runtimes, assign goals and tasks, and track work and spending from a dashboard. It provides an org chart, roles and permissions, approval gates, budgets, scheduled routines, shared skills, and connections to services. It is aimed at people running several coding or business agents and wanting one place to monitor activity, review outputs, and control costs.

• Bring-your-own-agent orchestration
• Org chart with roles, permissions, and reporting lines
• Budget tracking with threshold alerts and pauses
• Scheduled routines with task owners and run history

★ 112kTypeScriptMITsince Mar 2, 2026Website

Taste Skill is an Agent Skill for coding assistants such as Claude Code and Codex that guides them through building or redesigning company websites. It classifies the client's business, deals a palette, font pairing and hero layout, and starts from a starter HTML page with a shared CSS system.

• Deals palettes, font pairings and hero layouts by page kind
• Starts from a starter HTML page that enforces layout rules
• Resolves real brand icons and vetted photographs into local files
• Browser checker renders pages at several widths and reports pass, warn or fail

★ 111kJavaScriptMITsince Feb 19, 2026Website

Agent Reach is a command-line tool that installs and manages read-only connectors so AI agents can search and read sites they normally cannot reach. It supports default channels such as Bilibili, V2EX, GitHub, YouTube, and web search, with optional channels like Xiaohongshu, Xueqiu, LinkedIn, Reddit, and Boss. It also handles captions, speech-to-text, structured data, and login reuse through a dedicated background browser. Developers using CLI agents can add, remove, check, upgrade, and uninstall channels by asking the agent in plain language.

• Default connectors for Bilibili, V2EX, GitHub, YouTube, and web
• Optional channels including Xiaohongshu, Xueqiu, Boss, Xiaoyuzhou, LinkedIn, and Reddit
• Video captions, speech-to-text, and podcast transcription
• Read-only searching and reading with a dedicated background browser and per-site login reuse

★ 109kPythonMITsince Feb 23, 2026

Archify is an open-source tool that works with AI coding agents to turn a description or repository into an interactive HTML diagram or explainer page. It supports architecture, workflow, sequence, data flow, and other diagram types, and lets users refine results through chat. Developers and teams can use it to understand systems, document processes, and share visual explanations as self-contained files with themes and export options.

• Creates interactive HTML diagrams and explainer pages from descriptions or repositories
• Supports architecture, workflow, sequence, data flow, ERD, class, tree, timeline, and waterfall diagrams
• Refines diagrams through chat requests such as adding services or highlighting paths
• Exports PNG, WebP, JPG, PDF, and share cards, with light and dark themes

★ 94kJavaScriptApache-2.0since Apr 15, 2026Website

World Monitor is a real-time intelligence dashboard that combines curated news feeds, geopolitical monitoring, market and infrastructure tracking, and cross-stream signals in one interface. It offers web, desktop, terminal, CLI, SDK, and MCP access, with a local AI option via Ollama. It is aimed at developers, analysts, and agents who want situational awareness and programmatic access to live global data.

• Curated news feeds synthesized into briefs
• 3D globe and WebGL flat map engines
• Country Instability Index scores and movement
• Local AI via Ollama with no API keys

★ 90kTypeScriptAGPL-3.0since Jan 8, 2026Website

RTK is an agent-agnostic tool that reduces the token volume of command output before it reaches an AI coding agent. It ships as a Rust binary and uses hooks or agent plugins to rewrite shell calls, then filters, groups, truncates, and deduplicates output across many developer commands. Developers using coding assistants can install it to cut context size without changing their usual shell workflows. An optional separate proxy daemon also compresses provider API traffic.

• Compresses and filters command output before it reaches the LLM context
• Supports 100+ commands across Git, cloud, language, and container tools
• Optional RTK Proxy daemon compresses provider-API traffic

★ 86kRustApache-2.0since Jan 22, 2026Website

HyperFrames is an open-source framework for turning HTML, CSS, media, and seekable animations into deterministic MP4 videos. Developers can author compositions as plain HTML with timing attributes, preview them in a browser, and render locally, in Docker, or on AWS Lambda. It includes a CLI, reusable catalog blocks, and skills that help AI coding agents plan, write, lint, preview, and render video projects. It is aimed at developers and teams building repeatable motion graphics and automated video pipelines.

• HTML-native compositions with data attributes for timing and tracks
• Deterministic frame capture in headless Chrome with FFmpeg encoding
• CLI for scaffold, preview, lint, inspect, and render
• Agent skills for Claude Code, Codex, Cursor, and other coding agents

★ 80kTypeScriptApache-2.0since Mar 9, 2026Website

Headroom is a software toolkit that compresses the text an AI agent or application sends to a large language model, including tool output, logs, retrieved chunks, files, and chat history, before the request reaches the model. It works as a Python or TypeScript library, a local proxy, or a wrapper for coding agents, and it can cache originals for retrieval. Developers building LLM apps or using agent tools can use it to reduce input tokens, with optional output-token shaping and shared memory across agents.

• Compresses tool outputs, logs, RAG chunks, files, and conversation history
• Python and TypeScript libraries plus a local proxy
• Reversible compression with local cache and retrieval
• MCP server and cross-agent shared memory

★ 79kPythonApache-2.0since Jan 7, 2026Website

An AI-powered job application framework that runs on Claude Code. After you fork it and fill in your profile, Claude can search job portals, evaluate postings against your background, draft a tailored CV and cover letter, review and revise them, compile the PDFs, and check ATS readability. It also ranks scraped jobs, prepares interview packs and mock interviews, records outcomes, and can sync status from Gmail or publish a read-only Notion view. Aimed at job seekers willing to set up the Claude Code toolchain.

• Searches job portals and ranks postings by fit
• Drafts, reviews, and revises tailored CVs and cover letters
• Compiles and visually inspects LaTeX PDFs, then checks ATS text extraction
• Builds interview prep packs and runs mock interviews

★ 70kPythonMITsince Mar 18, 2026

MemPalace is a local-first memory system for AI coding assistants and other MCP clients. It stores conversation history as verbatim text and retrieves it with semantic search, organizing content into scoped wings, rooms, and drawers. It ships a CLI, Docker image, and MCP server, supports pluggable vector backends, and keeps data on your machine unless you opt in. Developers using Claude Code, Codex, Cursor, or local models would use it to retain and recall long-running project context.

• Verbatim conversation storage with semantic search
• Structured palace index with wings, rooms, and drawers
• Pluggable vector backends including ChromaDB and pgvector
• MCP server with tools for recall, diaries, and knowledge-graph operations

★ 60kPythonMITsince Apr 4, 2026Website

Hypit is an open-source skill and system that lets coding agents such as Claude Code and Codex create video workflows. You can give an agent a reference video to clone, start from templates, or describe a video; the agent builds an editable, re-runnable composition with footage, captions, B-roll, effects, and code-rendered visuals. It supports variants, localization, and pluggable components, and generation models are optional. Developers and creators use it to produce social ads, UGC, podcast clips, and similar videos.

• Clone a reference video into an editable workflow
• Generate a video from a text description
• Word-anchored captions, B-roll, and effects
• Reusable variants and localized versions

★ 28kTypeScriptsince Jul 29, 2026Website