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.

Prompt, run, edit, and deploy full-stack Node.js applications from a browser tab: Bolt.diy is the official open-source version of Bolt.new's AI coding agent. Its foundation is StackBlitz's WebContainer technology - a sandboxed in-browser Node.js environment where the AI controls the whole stack: filesystem, npm, dev servers, terminal, and browser console. That means the agent does not just generate code; it installs dependencies, runs Vite or Next.js, reads errors, and fixes them. The defining difference from Bolt.new is model choice per prompt: 19+ providers including OpenAI, Anthropic, Gemini, DeepSeek, Groq, Mistral, Amazon Bedrock, and local models via Ollama or LMStudio, extensible through the Vercel AI SDK. Development ergonomics include live preview, a diff view of AI changes, codebase search, file locking to prevent generation conflicts, 15+ starter templates (React, Vue, Next.js, Astro, Svelte, Expo), and MCP support for external tools. Projects integrate with Git and Supabase, and deploy in one click to Vercel, Netlify, or GitHub Pages.

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.

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.

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.

OpenCode Manager is a mobile-first command center for AI coding agents. Install the PWA on your phone or tablet and you get real-time streaming chat, multi-repository Git operations, and scheduled automation right in your pocket. Git integration handles SSH-authenticated repo cloning, worktree management, unified diffs, and branch operations across all your projects in one dashboard. Chat with coding agents through Server-Sent Events streaming that supports slash commands, @-mentions for files, Plan and Build modes, and Mermaid diagram rendering for architecture discussions. Schedule reusable prompts to run against any repository on intervals or cron expressions, with each run tracking history and linking to sessions so you can pick up exactly where automation left off. MCP server configuration adds local and remote HTTP servers with OAuth support, plugging into the broader Model Context Protocol ecosystem. A dedicated assistant workspace provides an isolated AI environment with auto-provisioned skills for managing schedules, notifications, and settings. Multiple AI providers are supported including Anthropic, GitHub Copilot, and OpenAI-compatible services, each configurable with custom system prompts and granular tool permissions. Push notifications alert you to session events, agent questions, errors, and task completions across all managed repositories.

Open-sourced by the Databricks AI team under Apache 2.0 and reaching over 8,500 GitHub stars within two months of launch, Omnigent introduces the meta-harness concept: a common orchestration layer that sits above existing AI coding agents and makes them interoperable parts of a governed, collaborative system. The platform wraps Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and any custom agent defined in a simple YAML configuration file into sandboxed sessions with a uniform API, then exposes each session through the terminal, a web UI, a native desktop application, mobile interfaces, and a REST API. Built-in multi-agent workflows include Polly, a coding orchestrator that delegates tasks to parallel sub-agents in separate git worktrees and routes each diff to a reviewer from a different vendor, and Deep Research, which plans sub-queries, searches the live web through MCP servers, reads full pages, and cross-checks claims across independent sources. Contextual security policies go beyond static allow/deny rules by maintaining per-session state to enforce spend caps, model routing, approval gates for destructive actions, PII blocking, and repository-scoped write restrictions across server-wide, per-agent, and per-session levels. The OS sandbox restricts filesystem and network access while intercepting egress requests to inject credentials only on approved calls. Cloud sandbox providers including Modal, Daytona, E2B, CoreWeave, Kubernetes, and Databricks launch disposable execution environments per session. 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 6,500 GitHub stars, HolaOS bills itself as an "open agent computer" that reimagines the traditional operating system as a shared workspace where humans and AI agents collaborate across files, browsers, and 100+ integrated tools simultaneously. Unlike chat-only interfaces, HolaOS places live application UIs—Notion-style editors, browsers, custom workspace apps—side by side with the agent conversation, so operators always see what agents are doing and can intervene at any moment. The persistent memory system stores workspace knowledge locally as Markdown files and embedded vectors via SQLite vec, enabling RAG-powered recall that survives session boundaries without the typical context window bloat. Safe Session Compaction reserves roughly 70% of the model context window for fresh reasoning while folding older history into structured checkpoints that retain goals, constraints, progress, and decisions. Agents connect to Linear, GitHub, Slack, Jira, HubSpot, Gmail, and dozens more through one-click OAuth, automatically fetching relevant signals and converting scattered app data into working memory. BYOK support for Claude, GPT, and Gemini models lets operators use their own API keys at zero markup, while built-in Kimi K3 and GLM-5.2 models provide ready-to-use alternatives. Skills package reusable workflows that any agent can invoke on demand, and scheduled triggers enable autonomous digests, monitors, and reports. The runtime supports independent server deployment alongside the desktop client. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Modified Apache 2.0 licensed.

Kandev provides a command center for orchestrating AI coding agents across parallel workstreams. The Go backend paired with a Next.js frontend delivers kanban boards with drag-and-drop columns, pipeline workflow definitions with per-step agent handoffs, and an IDE-like review workspace combining file editor, file tree, terminal, browser preview, and unified git diffs. Multi-provider support connects Claude Code, GitHub Copilot, Codex, Qoder, Grok, and custom agents through configurable profiles with per-agent prompts, runtimes, and review gates. Tasks execute in isolated git worktrees with multi-repository support, letting agents work on separate branches simultaneously while changes surface in a consolidated review interface. Native integrations with GitHub, GitLab, Jira, Linear, Sentry, and Slack pull external issues into the kanban and link tasks to pull requests. Kandev exposes streamable HTTP and SSE MCP endpoints, enabling external clients — Cursor, Claude Desktop, Augment — to create tasks and read workspace context programmatically. Workflow definitions export as portable YAML for sharing across installations. Agentic workflows chain multi-step pipelines mixing different models per step — Opus for architecture, Sonnet for implementation, with human review gates between stages. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 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.

With 2,400+ GitHub stars, HolyClaude replaces a two-hour manual setup process with a single docker compose up command that launches a fully configured AI coding workstation. The container ships with Claude Code pre-installed alongside seven additional AI CLIs — Gemini CLI, OpenAI Codex, Cursor, TaskMaster AI, Junie, OpenCode with OpenRouter multi-provider support, and Pi Coding Agent — all accessible through the CloudCLI browser-based web interface on port 3001. A headless Chromium browser running on Xvfb display :99 comes pre-configured with Playwright 1.61 for automated screenshots, testing, and web scraping without additional setup. The development toolchain includes Node.js 26, Python 3, TypeScript, git, GitHub CLI, database clients for PostgreSQL, SQLite, and Redis, plus deployment CLIs for Vercel, Wrangler, Netlify, and Azure. Process management uses s6-overlay for automatic restarts and graceful shutdown, while bind mounts to ./data/claude and ./workspace ensure credentials, configuration, and project files persist across container rebuilds. The full image supports AMD64 and ARM64 architectures, running on Linux, macOS with Docker Desktop, Windows with WSL2, and Synology or QNAP NAS hardware. A slim tag at roughly 2GB omits the browser stack for environments that do not need it, while the full image at 4GB includes everything. Users authenticate with their existing Anthropic account directly through Claude Code with no credential relay or proxy involved. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.