DeepSeek Harness gained over 60,000 GitHub stars within hours of its August 2026 launch, establishing itself as the first fully modular open-source agent runtime where literally every component is a swappable plugin. Built on the Cordis framework—a programming paradigm for spatiotemporal composability—dsh decomposes the entire agent stack into independently replaceable pieces: model adapters for DeepSeek, Anthropic, OpenAI, AWS Bedrock, Azure, and Google Gemini; tool registries covering bash execution, file system operations, web search, subagent delegation, and todo management; plus session stores, sandboxes, approval policies, orchestration loops, and the user interface itself. Four operating modes serve different workflows: Standard provides the full toolset, Code mode uses model-generated code to compose multi-round tool calls, Minimal strips down to a shell and editor for benchmarking, and Creator mode lets developers inspect the running runtime and test Cordis plugins in memory. The kernel handles plugin mounting, unmounting, and dependency resolution while typed events and services coordinate between components. Profiles and bundles allow the same codebase to produce entirely different products—a terminal coding agent, a browser-based workspace, a headless automation service, or an ACP/JSON-RPC endpoint—by swapping YAML configuration layers. Session history is stored as an append-only event stream for full trajectory replay, and project-level hooks on agent lifecycle events enable fine-grained behavioral customization. MCP client integration connects to external tool servers, while Agent Client Protocol enables programmatic orchestration. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

QM is Y Combinator's internal multiplayer agent infrastructure that shifts AI agents from personal assistants to shared company operating layer. The headless TypeScript core runs on Node.js with Fastify handling HTTP, Slack integration via Bolt, and a web UI built with Vite and Lit. PostgreSQL stores sessions, memory, queue state, and audit logs. Every person and every Slack channel gets an isolated sandbox with its own durable file system, installed tools that persist across runs, private memory, keychain view, permissions, and background crons. The harness-agnostic architecture routes agent tasks through Pi, OpenCode, Codex, or Claude Code without vendor lock-in, with production implementations swapping via a single wiring file. Three org-level security postures gate execution: Strict requires human approval for every tool call, Auto applies automated content screening, and Dangerous removes all pauses. Skills are scope-owned and shareable by grant, with admin-gated promotion to the entire organization and skill packs importable from Git repositories. The web apps feature lets agents spin up custom internal applications published to specific users. The qm CLI bootstraps operator-owned deployment directories with digest-pinned release images, infrastructure rendering, secret management, and live verification checks for Docker, Fly.io, or AWS ECS Fargate targets. On RepoCloud, deploy QM on a dedicated VPS with PostgreSQL persistence, Docker socket access, root SSH access, and complete control over your multiplayer agent infrastructure, all under the MIT license.

With over 14,000 GitHub stars and 41 releases since its May 2026 launch, Prime Agent delivers a fundamentally different approach to AI coding agents by treating the agent's own operating environment as programmable state that improves through use. The Recursive Language Model architecture provides the model exactly one tool — a persistent IPython kernel — where file operations, shell commands, subagent delegation via rlm() function calls, and context management all happen through code rather than rigid tool-calling schemas. Subagents launch as independent sessions with their own model, kernel, and history, communicating results through agent_message.send() without blocking the parent. The Continual Harness stores supplemental prompts, memories, skill descriptions, and reusable subagent specifications as durable state that the /refine command updates through small, evidence-backed edits with full rollback by ID. Daemon-backed sessions keep running when the terminal disconnects, with automatic context compaction summarizing older messages while preserving recent state. The TUI provides an Agent View for monitoring, switching between, and steering multiple live sessions simultaneously. Autonomous mode operates within configurable turn, token, and time budgets with user-defined quality gates. Persistent goals, heartbeats, and scheduled prompts maintain continuity across terminal sessions. Compatible with Anthropic Claude, OpenAI, Google Gemini, local models via Ollama or vLLM, and Prime Inference endpoints. Install via a single curl command on Linux or macOS. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Open source GrokBot, built by the team behind the AG-UI protocol. OpenBot is the open-source enterprise agent platform that gives every AI coworker its own sandboxed computer — a real Chromium browser with its own login sessions, a private filesystem, and only the MCP tools you explicitly grant. The centralized gateway evaluates CEL policy rules against tool name, intent, bot identity, page URL, element attributes, and file paths before any action executes, writing an immutable audit row for every call and outcome. Any agent that speaks AG-UI — LangGraph, Mastra, CrewAI, Pydantic AI, Google ADK, or hand-written endpoints — registers as a Bot and receives its own channel with persistent conversation history. The take-the-wheel system lets humans assume control when an agent encounters login walls or two-factor prompts, recording control transfers as structured audit events. Knowledge documents from Google Drive and OneDrive carry source-based permissions where deny principals always win and ambiguous mappings refuse retrieval entirely. The React and Vite frontend provides live screen viewing of each agent's browser, channel-based chat, admin settings, and component galleries. The Hono API server on port 3001 handles authentication, role-based access, tenant packaging, and credential management backed by PostgreSQL with pgvector. Deploy via Docker Compose with the included supervisor that manages per-bot computer containers. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

OpenRouter's most-used application by token volume — over 17 trillion tokens processed — Hermes Agent is an open-source autonomous agent built by Nous Research that lives on your server and gets more capable every day. Define a goal in natural language and Hermes plans sub-tasks, executes them through tool integrations, observes results, handles errors, and refines until the job is done or it genuinely needs your input. Persistent memory with full-text search and LLM summarization lets it recall context across sessions, and an agent-created skills system self-improves after complex tasks. A messaging gateway connects Telegram, Discord, Slack, WhatsApp, Signal, and 16 more platforms with cross-channel conversation continuity. A built-in cron scheduler runs daily reports, nightly backups, and weekly audits unattended. Subagent spawning parallelizes workstreams, and six terminal backends — local, Docker, SSH, Singularity, Modal, and Daytona — fit any infrastructure. Works with any LLM provider: Nous Portal, OpenRouter for 400+ models from 70+ providers, OpenAI, Anthropic, or your own endpoint. The API key you supply powers all LLM calls; billing goes through your own account. Running on a dedicated VPS with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

With 83,000+ GitHub stars and $18.8M in Series A funding, OpenHands delivers the leading open-source platform for AI coding agents that scored 68.4% on SWE-bench Verified with Claude Opus 4.6, outperforming Devin 2.0's publicly reported 45.8%. The Agent Canvas web UI organizes work into persistent conversations where agents edit files, run shell commands, browse the web, and execute multi-step development tasks inside isolated Docker sandbox containers. The observe-plan-act loop drives agent behavior: the Python controller manages LLM abstraction via LiteLLM routing to 100+ providers including OpenAI, Anthropic, Google, DeepSeek, Qwen, Llama, and local Ollama models. Built-in skills for code review, Docker management, PRD generation, repo-rules enforcement, release notes, and test running attach to conversations automatically via auto-discovery or trigger-based activation. The Automations system schedules recurring agent tasks with configurable templates for CI workflows, dependency updates, and documentation generation. MCP server integration enables agents to access external tools and data sources. The REST API powers an OpenAI-compatible endpoint for connecting agents to chat UIs, IDEs, and voice platforms. GitHub, GitLab, Slack, and Jira integrations enable pull request reviews, issue resolution, and team notifications. The SDK provides Python and REST APIs for embedding agents in custom tools with local or cloud execution, custom agent behaviors, and Kubernetes deployment. On RepoCloud, deploy OpenHands on a dedicated VPS with Docker socket access, persistent project storage, root SSH access, and complete control over your AI development infrastructure, all under the MIT license.

Reaching 45,000 GitHub stars within seven months of launch, Multica is the fastest-growing open-source platform for managing AI coding agents as first-class teammates — assign an issue to Claude Code, Codex, Cursor, Copilot, Kimi, or any of 21 supported agent CLIs and it picks up the work, comments progress in real time via WebSocket, raises blockers, and hands the result back for human review before anything merges. The Go backend (Chi router, sqlc-generated type-safe queries, gorilla/websocket) connects to PostgreSQL 17 with pgvector for semantic search across workspace history, while the Next.js 16 App Router frontend delivers workspace dashboards showing per-agent token spend, execution time, daily cost charts, and runtime status across unlimited connected machines. Agent Skills provide reusable methods, reference material, and supporting files that compound across runs — a persistent knowledge layer that makes each subsequent task faster and more accurate. Squads let a leader agent select the right specialist for subtasks, creating multi-agent workflows without manual orchestration. Review gates ensure no AI-generated code ships to main without explicit human approval. Self-host via Docker Compose or Kubernetes with full Git integration across GitHub, GitLab, Gitea, and Forgejo including self-hosted instances. The CLI and REST API make every surface scriptable, and Autopilot automations trigger agent runs from events. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed with additional conditions.

Perplexity's answer-engine experience, self-hostable and open-source: Morphic searches the web and writes cited answers. Instead of returning a list of links, it searches the web, reads the sources, and generates a complete answer with inline numbered citations. The generative UI streams rich components, source cards with thumbnails, image grids, syntax-highlighted code, and LaTeX math, rather than plain markdown. Quick mode answers fast; Adaptive mode runs deeper multi-step research. Search backends are pluggable: the Docker Compose bundle ships with a private SearXNG instance so no search API key is required, and Tavily, Brave, and Exa are supported alternatives. LLM providers include OpenAI, Anthropic, Google, Ollama, and any OpenAI-compatible endpoint, with per-mode model mapping - fast, cheap models for quick searches, stronger models for adaptive research, tuning the cost-quality trade-off per query type. An inspector panel exposes tool execution during multi-step research, and AI-suggested follow-up questions keep an investigation moving. Chat history persists in PostgreSQL, results are shareable by URL, file uploads feed context into queries, and optional Supabase authentication adds multi-user or guest access. Because the default search path is your private SearXNG instance, research topics never hit a commercial search API - and with local Ollama models the marginal cost of a query approaches zero. Built with Next.js, TypeScript, and the Vercel AI SDK under Apache 2.0.

Designed as a unified personal AI workstation, QwenPaw operates as an autonomous digital copilot that coordinates scheduled automations, interactive coding sessions, and multi-channel team communication across everyday messaging apps. Users can dispatch long-running research tasks, process office documents including PDF, Excel, and Word files, and execute browser-based data collection without manual intervention. The built-in web console and terminal interface provide full visibility into agent reasoning, allowing operators to inspect intermediate thinking steps, review proposed file edits, and approve sensitive tool calls. Through direct connectors for Discord, Telegram, DingTalk, Lark, and WeChat, teams can trigger specialized skills or query shared workspaces directly from their existing group channels. A self-evolving personal memory system continuously indexes chat interactions and local resources into editable Markdown files, ensuring knowledge persists across restarts and task delegations. Built-in guardrails including a sandboxed execution runtime, tool access policies, and automated skill scanners protect underlying host files from unauthorized modifications during autonomous scripting runs. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.

Instead of a separate chatbot for each channel, Synaplan runs one self-hosted AI assistant for your team, website visitors, WhatsApp line, and inbox, and shows what every reply cost. In the web app, connect OpenAI, Anthropic, Gemini, Groq, Mistral, xAI, Hugging Face, or local Ollama. Each kind of work gets its own model, from chat and image to video and audio, and every answer carries a Free, Low, Mid, or High cost badge. A planner splits a big request into a task graph and sends each step to the model that fits. On your website, one script tag adds a chat widget, and a human operator can take over any conversation live. On WhatsApp Business and email, the assistant replies on the channel where the question arrived. Behind every channel sits one knowledge base: upload PDFs, Word and Excel files, or images with OCR, and semantic search feeds them into answers. Saved tasks run on demand or on a schedule, approvals pause any action that writes, and watched pages mail you a diff when a URL changes. Connections reach Microsoft 365, Dropbox, Nextcloud, ownCloud, WebDAV, CalDAV, Jira, and Confluence. An MCP server and an Anthropic-compatible endpoint let Claude, Cursor, and Claude Code use your instance. Users and groups share folders and assistants, and admins see usage and health. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

With over 24,000 GitHub stars and origins in the MemGPT research paper on virtual context management, Letta has evolved into the leading open-source platform for building AI agents that maintain persistent memory, identity, and continuity across sessions rather than operating as stateless prompt-response loops. The core architecture uses memory blocks — structured, labeled text chunks that reside permanently in the agent's context window — allowing agents to programmatically rewrite their own memory, learn new skills, and improve through a sleeptime dreaming process that runs reflection and memory organization during idle periods. The self-hosted App Server deploys via Docker and exposes a WebSocket API on port 4500, letting the TypeScript Agent SDK connect from any application using local, remote, or cloud backends. Agents support git-versioned memory through MemFS where every memory change is tracked and auditable, multi-agent communication via subagents, scheduled tasks, and integration with messaging platforms including Slack, Discord, Telegram, WhatsApp, and Signal. The platform is fully model-agnostic, routing to OpenAI, Anthropic, xAI, or self-hosted open-weight models through Ollama depending on cost, performance, and data residency requirements. The Agent File format serializes complete agent state — memory, skills, prompts, and conversation history — into portable snapshots. Desktop applications for macOS, Windows, and Linux provide native interfaces alongside the terminal CLI and web chat at chat.letta.com. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Pi Web is a browser interface for the Pi coding agent ecosystem, providing a visual workspace that reads the same local configuration and session files as the CLI. The session workspace groups conversations by project with running state indicators, context usage percentages, cost tracking, and compaction details, while two branching modes let users create independent session files from earlier messages or fork branches within existing sessions to explore alternative coding directions. Real-time streaming via Server-Sent Events delivers agent responses with structured Markdown rendering, thinking steps, tool call visualization, and image drag-and-drop input. The project file explorer browses working directories with syntax-highlighted source preview, Git diff inspection, and rendering for Markdown, images, audio, PDFs, and DOCX files with automatic refresh. Git worktree support switches checkouts from the sidebar while keeping sessions from the same repository grouped together. The Models panel manages provider authentication via OAuth and API keys, model selection, model smoke tests, and models.json configuration shared bidirectionally with the CLI agent. The Skills panel lists, searches, installs, and toggles agent skills without terminal access. The interface ships with English and Simplified Chinese translations, light and dark themes, a chat minimap, keyboard shortcuts, and completion sounds. Basic Auth protects remote access when binding to non-loopback addresses. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

With 15,500 GitHub stars and over 1,100 forks, Plandex delivers a terminal-based AI coding agent purpose-built for the complex, multi-file tasks that overwhelm single-file AI assistants. The Go-powered server maintains a cumulative diff review sandbox that quarantines all AI-generated changes from your project files until you explicitly approve them — enabling 20-file refactors where you cherry-pick good changes and reject bad ones without touching git. A 2M token effective context window loads only what each step requires, while tree-sitter project maps index repositories exceeding 20M tokens across 30+ programming languages, providing structural awareness of class hierarchies, function signatures, and import graphs without burning tokens on full file content. The configurable model pack system assigns different models to different roles — Claude for planning, GPT for coding, Gemini for summarization — supporting Anthropic, OpenAI, Google, OpenRouter, Azure OpenAI, AWS Bedrock, DeepSeek, Perplexity, and Ollama for local models. Full auto mode handles end-to-end autonomous workflows including high-level planning, context loading, implementation, terminal command execution, and automated debugging of both terminal and browser applications. The interactive REPL provides fuzzy auto-complete, version-controlled sandbox branching, rewind to any previous point, and Git integration for commit message generation. The Plandex Server exposes 60+ REST API endpoints for programmatic orchestration across organizations, projects, plans, and branches. Deploy via Docker Compose for self-hosted operation with your own API keys. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Your team's group chat could have a colleague who has read every email, ticket, and document, remembers all of it, and answers in seconds; MindRoom puts AI agents like that into your chat rooms as full participants. Each agent is a real user on the open Matrix protocol, so you talk to it in MindRoom Chat on web or mobile, in any other Matrix client, or through bridges in Slack, Telegram, WhatsApp, and Discord. Give every agent its own model: a local model for private rooms, a frontier model for hard coding problems, and switch per room or thread whenever the work changes. Agents connect to more than 100 tools, including Gmail, calendars, GitHub, Google Drive, Home Assistant, and any MCP server, and they search your own documents and knowledge bases for answers with sources. Memory persists across conversations and platforms, so an agent remembers people, preferences, and project context months later and can turn repeated work into reusable skills. Before an agent sends an email or runs a command, an approval card shows exactly what will happen; you tap once to allow it, and sandboxed workers keep executed code away from your credentials. Schedule morning briefings, silent inbox checks, and recurring digests in plain language, or wake an agent from any other system with a single HTTP request. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Roughly $1M of development work, open-sourced: Maybe Finance began as a $249/year commercial personal finance product before the company released it all. It aggregates bank accounts, credit cards, loans, investments, crypto, and real estate into a single net worth dashboard with historical trend charts - replacing the spreadsheet that usually glues a whole portfolio together. Transactions are categorized and tagged with rules, with merchant tracking and search across imported or synced activity; budgets track spending by category against plan; and the investment view follows holdings, cost basis, and returns across brokerage accounts. Multi-currency support converts accounts held in different currencies into a single reporting currency, bank synchronization works through Plaid where supported, and manual CSV import covers any institution. An optional AI assistant answers questions grounded in your own financial data. Because the app was built as a paid product with professional design before being open-sourced, its interface quality exceeds most community finance tools - and self-hosting means your balances and transactions are not monetized by a free app or gated behind an annual subscription. The stack is Ruby on Rails with Hotwire on PostgreSQL, licensed AGPL-3.0 and deployed via Docker. The original repository is archived; development continues in the community fork Sure, compatible with the same self-hosted setup.

JobNavigator runs your whole job hunt as an automated pipeline, scraping career pages and job boards, scoring every posting against your résumés, and tracking each application to its outcome. Discovery fans out across company career pages through twelve ATS handlers, aggregators like LinkedIn, Indeed, and ZipRecruiter, and a Chrome extension that saves any posting you stumble on. Each new job is scored against your base résumés, lightly for a quick number or in full, with keyword coverage and requirement mapping that show exactly where you match and where you do not. Strong matches get a tailored résumé from eight PDF templates and a cover letter written from your persona, ready for your review before anything is sent. The extension then auto-fills application forms and drafts answers to free-text questions from your stored Q&A bank. Every application moves through stages and interviews in one tracker, while Gmail polling spots replies and updates the status for you, and Telegram delivers new-job alerts and a daily digest. A household can share one install with up to five accounts, each keeping its own jobs, résumés, and PIN, while scraping runs once for everyone. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

After the bills are covered, how much is actually safe to spend? monetr organizes an entire budgeting application around that one question. Inspired by the defunct Simple bank, it budgets paycheck by paycheck rather than month by month. Funding schedules encode when you get paid - including multiple schedules per account and a weekend-exclusion option for deposits that land early - and every expense or savings goal is tied to one. monetr then calculates how much of each recurring obligation (rent, car payment, subscriptions, on any repeat interval) to set aside from each paycheck, so a large bill never has to come out of a single check. What remains after allocations is surfaced as Free-To-Use, and a forecasting timeline projects contributions, due dates, and leftover funds forward so low-balance periods are visible before they happen. Transactions arrive either as OFX uploads from your bank or automatically through a Plaid connection using your own API credentials, keeping balances and transactions synced without manual entry. The app is mobile-friendly and installable as a PWA on desktop or phone. Self-hosting via Docker Compose is completely free, with all source code public and your financial data on your own hardware.

Perplexity's search experience without Perplexity: Vane deploys Perplexica, an open-source AI answer engine built as the self-hosted alternative. Instead of returning a page of links, it reads your question, searches the live web through the SearxNG metasearch engine, and composes a direct answer with cited sources. Retrieval quality comes from embeddings and similarity search: fetched pages are re-ranked against the query so the model answers from the most relevant passages rather than whatever ranked first. Two query modes cover different needs - Normal mode runs a straightforward web search, while Copilot mode generates multiple reformulated queries and actively pulls content from top matches for harder questions. Focus modes specialize retrieval for academic papers, YouTube, Reddit discussions, Wolfram Alpha calculations, or the general web. The answering model is your choice: OpenAI-compatible APIs or fully local LLMs such as Llama 3 and Mixtral through Ollama, which keeps queries entirely on your infrastructure. Because SearxNG pulls live results, answers reflect current information, and no search history is tracked.