The AI-agent ecosystem is moving ridiculously fast. New tools appear every week, established projects change direction, and yesterday’s exciting experiment can become tomorrow’s abandoned repository.
This comparison covers all products discussed, including open-source, source-available, and proprietary tools. Their licensing status is stated explicitly rather than hiding materially different licenses under the increasingly elastic label “open.” Products are ordered by GitHub stars.
Data checked on July 22, 2026. GitHub stars measure repository visibility and community interest—not software quality, user count, security, maturity, or suitability for production. For proprietary products, stars belong to the vendor’s public repository and do not mean that the complete product is open source.
| Product | Developer | Category | Interfaces and Integrations | Model Support | License / Source Status | GitHub Stars | Target Audience |
|---|---|---|---|---|---|---|---|
| OpenClaw | OpenClaw community | Personal AI assistant | CLI, gateway, WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Teams, Matrix, IRC, Google Chat | Multiple providers | MIT — open source | 383,752 | Users wanting a self-hosted personal assistant available through messaging platforms |
| Hermes Agent | Nous Research | Self-improving personal agent | CLI/TUI, WebUI, API, Telegram, Discord, Slack, WhatsApp, Signal, email, Home Assistant, MCP | Provider-independent; API and local models | MIT — open source | 218,499 | Technical users wanting a persistent, self-hosted agent with memory, skills, automation, and remote access |
| n8n | n8n GmbH | Workflow and AI automation | Visual workflow editor, Web UI, API, CLI, webhooks, custom nodes, 400+ integrations | Provider-independent through integrations | Sustainable Use License — source-available/fair-code, not OSI open source | 197,396 | Automation builders and teams connecting applications, data, APIs, and AI agents |
| OpenCode | Anomaly | Coding agent | CLI/TUI, desktop application, IDE integration, client/server mode | Provider-independent; API and local models | MIT — open source | 188,387 | Developers seeking an open-source, terminal-first alternative to proprietary coding agents |
| Claude Code | Anthropic | Coding agent | CLI/TUI, VS Code, JetBrains, desktop and web workflows, GitHub Actions, SDK | Anthropic Claude models | Proprietary — public repository is not the full open-source product | 138,615 | Developers prioritizing strong agentic coding and deep repository reasoning within the Claude ecosystem |
| Codex CLI | OpenAI | Coding agent | CLI/TUI, IDE integration, GitHub, CI workflows, non-interactive execution | Primarily OpenAI models | Apache-2.0 — open-source client | 100,466 | Developers and engineering teams already using OpenAI models and services |
| OpenHands | OpenHands | Agent development platform | Web UI, CLI, REST API, GitHub, GitLab, Slack, Linear, webhooks, CI, ACP-compatible agents | Provider-independent | MIT core; separately licensed enterprise components | 81,615 | Teams operating autonomous coding agents and self-hosted engineering automation |
| Pi | Mario Zechner and community | Agent toolkit and coding agent | CLI/TUI, TypeScript libraries, agent loop, SDK, extension system | Multiple providers | MIT — open source | 74,881 | Developers preferring a minimal, programmable agent foundation over a large platform |
| Paperclip | Paperclip AI | Multi-agent control plane | Web UI, API, CLI adapters, shell commands, HTTP/webhooks, Claude Code, Codex, OpenClaw, custom agents | Runtime-agnostic; orchestrates external agents | MIT — open source | 74,407 | Individuals and teams coordinating multiple agents with tasks, roles, budgets, and governance |
| Cline | Cline | Coding agent | VS Code, JetBrains, CLI, SDK, browser automation, MCP, terminal execution | Provider-independent; API and local models | Apache-2.0 — open source | 64,913 | Developers wanting an autonomous agent directly inside their IDE |
| Goose | Block / Agentic AI Foundation | Extensible general-purpose agent | CLI, desktop application, MCP, developer tools, extension system | Provider-independent; API and local models | Apache-2.0 — open source | 51,429 | Developers and power users building extensible local agent workflows |
| Aider | Aider community | AI pair programmer | CLI, Git integration, browser-assisted workflow, scripting | Provider-independent; API and local models | Apache-2.0 — open source | 47,599 | Developers wanting a mature, focused, Git-native terminal coding assistant |
| Cursor (public repository) | Anysphere | AI-native code editor | Desktop IDE, VS Code ecosystem, terminal, Git, remote development, background agents | Vendor-managed model selection and supported providers | Proprietary | 33,054 | Developers wanting a polished, integrated AI-first editor experience |
| Kilo Code | Kilo | Agentic engineering platform | VS Code, JetBrains, CLI, cloud agents, terminal and browser tools | Provider-independent; API and local models | MIT — open source | 26,438 | Developers wanting one agent platform across IDE, terminal, and cloud environments |
Focus: cross-platform personal assistance through messaging applications.
OpenClaw is a self-hosted personal AI assistant built around messaging-platform integration. It can operate through WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and numerous other communication channels. Its primary focus is making an autonomous assistant continuously accessible across devices instead of limiting it to an IDE or terminal session.
Its enormous community footprint is impressive, but GitHub popularity should not be confused with engineering maturity. Deployment quality, extension quality, security, and operational reliability still need to be evaluated independently.
Focus: persistent personal assistance, learning, automation, and tool use.
Hermes Agent is the self-improving AI agent built by Nous Research. Its distinguishing feature is a built-in learning loop: it creates reusable skills from experience, improves them during use, persists knowledge, searches previous conversations, and gradually develops a richer model of the user.
It can run on a small VPS, workstation, GPU cluster, Docker environment, or serverless infrastructure and remain accessible through its terminal interface, WebUI, API, and messaging platforms. Unlike agents focused exclusively on coding, Hermes is designed as a long-lived personal and technical assistant.
Focus: deterministic workflow automation and application integration with AI capabilities.
n8n is a visual workflow-automation platform combining more than 400 integrations, webhooks, APIs, custom code, and native AI capabilities. It is especially useful for deterministic business processes in which agents are only one component of a larger workflow involving databases, SaaS applications, infrastructure, approvals, and human review.
Unlike the autonomous agents in this list, n8n is primarily an orchestration engine: the workflow graph remains explicit and reproducible. Its code is publicly available under the Sustainable Use License, commonly described as fair-code or source-available, but it is not OSI-approved open source.
Focus: provider-independent, terminal-first software engineering.
OpenCode is a model-independent open-source coding agent designed around the terminal. It combines an interactive TUI with desktop and IDE interfaces while allowing users to select commercial APIs or local models.
Its main strength is delivering a polished coding-agent experience without locking the workflow to one model vendor. It is particularly attractive to developers who like the Claude Code style of working but want control over models, deployment, and source code.
Focus: high-capability agentic coding within Anthropic’s Claude ecosystem.
Claude Code is Anthropic’s proprietary coding agent for understanding repositories, planning changes, editing files, executing commands, running tests, and handling Git workflows. It is available through terminal, IDE, desktop, web, SDK, and automation-oriented interfaces.
Its main appeal is the tight integration between Anthropic’s models and its coding harness. The public GitHub repository is useful for distribution, issues, plugins, and related resources, but the complete product is not released under an open-source license.
Focus: agentic software development using OpenAI models.
Codex CLI is OpenAI’s open-source terminal coding agent. It can inspect repositories, modify files, execute commands and tests, and operate interactively or in automated workflows.
Its strongest fit is in environments already standardized on OpenAI models, GitHub, and CI automation. The agent implementation is Apache-2.0 licensed, although practical use generally depends on OpenAI’s commercial model infrastructure.
Focus: autonomous software-development agents and engineering automation.
OpenHands is a platform for autonomous software-development agents and engineering automation. It provides self-hosted agent execution, a developer control center, APIs, and integrations with systems such as GitHub, GitLab, Slack, and Linear. It can coordinate OpenHands agents as well as external ACP-compatible agents.
Its scope is broader than pair programming: OpenHands targets delegated engineering tasks, scheduled automations, and persistent teams of coding agents. The core is MIT-licensed, while some enterprise components use separate licensing terms.
Focus: minimal and programmable building blocks for coding agents.
Pi is a deliberately minimal toolkit for building and running coding agents. It provides a unified LLM API, an agent loop, a terminal interface, a coding-agent CLI, and reusable TypeScript packages.
Instead of imposing a large orchestration platform, Pi gives developers compact building blocks that can be modified or embedded into custom systems. Its strengths are programmability, simplicity, and experimentation with agent architecture.
Focus: organizational control and governance for teams of agents.
Paperclip is a control plane for teams of AI agents. If an individual agent is an employee, Paperclip aims to be the company around it. It provides tasks, organizational structures, roles, delegation, budgets, permissions, governance, audit trails, and scheduled heartbeats.
Paperclip does not replace coding or personal agents—it coordinates them. Claude Code, Codex, OpenClaw, shell commands, HTTP services, and custom agents can participate through adapters or webhooks. This makes Paperclip especially interesting for users who already operate several specialized agents and need structure above them.
Focus: transparent, autonomous software engineering inside the IDE.
Cline is an autonomous coding agent built primarily for IDE-based work. It can inspect a project, edit files, execute terminal commands, interact with browsers, and extend its capabilities through MCP servers.
Unlike terminal-first tools, Cline keeps the agent’s plans, changes, and approval flow close to the developer’s editor. Its focus is interactive software engineering with broad model-provider support and visible human control over potentially destructive actions.
Focus: extensible local automation extending beyond code generation.
Goose is an open-source, extensible AI agent originally created by Block and now developed under the Agentic AI Foundation. It is available through CLI and desktop interfaces and uses extensions and MCP integrations to interact with development tools and external systems.
Goose is not restricted to suggesting code. It can install software, execute commands, edit files, run tests, and automate general technical workflows using commercial or local models.
Focus: controlled, Git-native pair programming in the terminal.
Aider is one of the most established open-source AI pair-programming tools for the terminal. It works directly inside Git repositories, creates coherent multi-file edits, and integrates generated changes with normal Git workflows.
Aider supports numerous commercial and local models while remaining smaller and more focused than full autonomous-agent platforms. Its strength is practical and controlled code editing rather than elaborate multi-agent orchestration.
Focus: an integrated, polished AI-first code-editor experience.
Cursor is a proprietary AI-native editor based on the VS Code ecosystem. It combines code completion, repository-aware chat, multi-file editing, terminal access, Git workflows, remote development, and background agents in one desktop application.
Its strength is product integration and user experience rather than openness or provider independence. Cursor maintains a public GitHub repository, but that repository is not the complete source code of the commercial editor; its star count should therefore be interpreted as community interest around the product, not as adoption of an open-source codebase.
Focus: agentic development across IDE, terminal, and cloud environments.
Kilo Code is an open-source agentic engineering platform spanning IDE, CLI, and cloud workflows. It targets developers who want autonomous code generation, terminal execution, and project assistance without committing to a single model provider or interface.
Compared with narrowly focused editor extensions, Kilo aims to provide a broader environment covering local development and delegated cloud execution.
These tools do not all compete directly. They occupy different layers of the emerging agent stack:
| Layer | Representative Products | Purpose |
|---|---|---|
| Personal agent | OpenClaw, Hermes Agent, Goose | Persistent assistance, communication, tools, and automation |
| Coding agent | OpenCode, Claude Code, Codex CLI, Cline, Aider, Cursor, Kilo Code | Repository analysis, code editing, testing, and software delivery |
| Agent framework | Pi | Building custom agents and agent runtimes |
| Engineering platform | OpenHands | Operating autonomous development agents and automations |
| Multi-agent control plane | Paperclip | Coordinating agents through tasks, roles, budgets, and governance |
| Workflow automation | n8n | Connecting applications, APIs, events, data, and agent workflows |
For a technically oriented, self-hosted setup, a particularly interesting combination is:
That architecture avoids expecting one agent to do everything—a design pattern that remains popular mainly because disappointment has not yet been added as an observability metric.
The comparison intentionally distinguishes three materially different categories:
Commercial APIs or hosted services used by an open-source client do not automatically make the client proprietary. Conversely, visible source code does not automatically make a product open source.