Agent harnesses, ranked by category
Ranked from the best-of-Agent-Harnesses list (167 harnesses, rescored weekly). Stars and licenses are facts; fit-by-category is our rubric, not a verified benchmark. Pick the harness and the model as a pair. We price the model side.
Last refreshed 2026-09-26 10:42 UTC · 87 of 167 harnesses researched in depth
Personal agents (13)
| Harness | What it is | Stars | Tier | Autonomy | Recovery | License |
|---|---|---|---|---|---|---|
| OpenClaw(opens in a new tab) | Self-hosted, always-on personal agent (formerly Clawdbot/Moltbot): a gateway + event-loop runtime that treats messages, heartbeats, crons, a | 390,149 | complex | headless | resumable | open |
| Hermes(opens in a new tab) | Nous Research's self-improving agent: a learning loop turns experience into reusable skills, builds a persistent user model across sessions, | 247,414 | slightly complex | headless | resumable | open |
| AnythingLLM(opens in a new tab) | Self-hosted "AI second brain" **harness**: chat with your documents, run built-in agent skills (web search, code execution, browsing), and m | 66,253 | complex | headless | resumable | open |
| nanobot(opens in a new tab) | Ultra-lightweight, self-hosted personal agent framework: the **harness** is a Python daemon wiring tools, memory, and MCP into chat/webhook | 48,409 | mostly simple | n/a | n/a | unknown |
| CowAgent(opens in a new tab) | Self-hosted **harness** (formerly chatgpt-on-wechat) that plans tasks, runs tools/skills, and self-evolves via memory; multi-model, multi-ch | 47,053 | slightly complex | n/a | n/a | unknown |
| Khoj(opens in a new tab) | Self-hostable "AI second brain": answers over your docs and the web, custom agents, scheduled automations, and multi-client reach (web, Obsi | 37,429 | complex | headless | resumable | open |
| Eliza(opens in a new tab) | Open "agentic operating system" (elizaOS): persistent multi-agent runtime with character files, a plugin ecosystem, and social/platform inte | 19,395 | complex | headless | resumable | open |
| Agent Zero(opens in a new tab) | Organic, prompt-defined personal agent framework: hierarchical sub-agents, persistent memory, browser and code tools, and self-modifying beh | 19,216 | slightly complex | bounded | resumable | unknown |
| OpenHarness (HKUDS)(opens in a new tab) | Open agent harness with a built-in personal agent ("Ohmo") that runs across Feishu, Slack, Telegram, and Discord; core tool-use, skills, mem | 15,810 | complex | bounded | resumable | open |
| QM(opens in a new tab) | Y Combinator's multiplayer agent **harness** for work, open-sourced from months of internal use: every person and room gets scoped memory, f | 15,181 | complex | headless | durable | open |
| OpenJarvis(opens in a new tab) | Stanford Hazy Research's local-first personal AI **harness**: on-device model inference (Ollama built in), agent execution, memory, and lear | 10,019 | slightly complex | headless | none | open |
| AIlice(opens in a new tab) | Fully autonomous general-purpose agent; one binary, Docker-ready, for when you want "set goal and walk away" without a framework. | 1,417 | slightly complex | bounded | none | open |
| Talon(opens in a new tab) | Multi-platform personal agent living in Telegram, Discord, Teams, and the terminal. The **harness** is a pluggable-backend loop (Claude, Kil | 83 | slightly complex | headless | resumable | open |
Enterprise harnesses (18)
| Harness | What it is | Stars | Tier | Autonomy | Recovery | Sandboxing | License |
|---|---|---|---|---|---|---|---|
| Codex(opens in a new tab) | OpenAI's terminal coding agent. The **harness** is the sandboxed tool-call loop with multi-provider support; the CLI is the shell. Reference | 125,493 | slightly complex | bounded | resumable | strong | open |
| Open Interpreter(opens in a new tab) | Lightweight terminal coding agent oriented to open models (DeepSeek, Kimi, Qwen). The **harness** is a code-execution loop — the model write | 68,390 | mostly simple | bounded | resumable | strong | open |
| autogen(opens in a new tab) | Conversable agents and group chats; code execution and human-in-the-loop; Microsoft origin, AG2 ecosystem. ⚠️ In maintenance mode since late | 61,079 | complex | bounded | resumable | strong | open |
| langgraph(opens in a new tab) | State-machine graphs over LLM steps; checkpointing, human-in-the-loop, and durable execution so workflows survive restarts. | 42,015 | slightly complex | headless | durable | basic | open |
| Khoj(opens in a new tab) | Self-hostable "AI second brain": answers over your docs and the web, custom agents, scheduled automations, and multi-client reach (web, Obsi | 37,429 | complex | headless | resumable | strong | open |
| deepagents(opens in a new tab) | LangChain's Python+TypeScript agent harness on top of LangGraph: planning tool, virtual filesystem, shell sandbox, sub-agent spawning—the "C | 29,596 | slightly complex | bounded | durable | unknown | open |
| openai-agents-python(opens in a new tab) | Handoffs, guardrails, and multi-LLM routing; minimal surface so you own the loop. | 29,582 | mostly simple | bounded | resumable | strong | open |
| letta(opens in a new tab) | Python agent runtime with tool use and control flow; lean API; stateful agents with long-horizon memory. | 24,809 | mostly simple | headless | durable | basic | open |
| Google ADK(opens in a new tab) | Google's official Agent Development Kit: code-first Python toolkit for building, evaluating, and deploying agents. Optimized for Gemini but | 21,581 | complex | headless | resumable | strong | open |
| SWE-agent(opens in a new tab) | LM-driven harness built for SWE-bench: edit state, command execution, and issue-focused loop—the reference agent stack next to the benchmark | 20,370 | slightly complex | headless | resumable | strong | open |
| pydantic-ai(opens in a new tab) | Type-safe Python agents with Pydantic I/O; multi-provider, MCP, Logfire observability, and human-in-the-loop. | 20,072 | slightly complex | bounded | durable | unknown | open |
| OpenHarness (HKUDS)(opens in a new tab) | Open agent harness with a built-in personal agent ("Ohmo") that runs across Feishu, Slack, Telegram, and Discord; core tool-use, skills, mem | 15,810 | complex | bounded | resumable | strong | open |
| QM(opens in a new tab) | Y Combinator's multiplayer agent **harness** for work, open-sourced from months of internal use: every person and room gets scoped memory, f | 15,181 | complex | headless | durable | unknown | open |
| Cloudflare Agents(opens in a new tab) | Persistent, stateful agents on Durable Objects: state, websockets, scheduling, and AI chat baked in. The serverless answer to "where does th | 5,609 | slightly complex | headless | durable | unknown | open |
| AgentStack(opens in a new tab) | Scaffolds full agent projects; plugs in CrewAI, LangGraph, OpenAI Swarm, LlamaStack and wires AgentOps observability from day one. | 2,191 | slightly complex | n/a | n/a | strong | open |
| langgraph-bigtool(opens in a new tab) | Build LangGraph agents with large tool sets; retrieval and on-demand tool loading so agents scale beyond context without stuffing every sche | 558 | slightly complex | bounded | durable | none | open |
| Proliferate(opens in a new tab) | Open-source AI IDE for Claude Code, Codex, OpenCode, and more. The **harness** contribution is the workspace/session orchestration layer: ru | 505 | complex | bounded | resumable | strong | open |
| AgentBox(opens in a new tab) | Runs multiple coding agents in parallel, each in its own sandboxed VM, locally or in the cloud, from one command. The **harness** contributi | 467 | slightly complex | n/a | n/a | strong | open |
Small business harnesses (12)
| Harness | What it is | Stars | Tier | Autonomy | Recovery | Build vs buy | License |
|---|---|---|---|---|---|---|---|
| Open Interpreter(opens in a new tab) | Lightweight terminal coding agent oriented to open models (DeepSeek, Kimi, Qwen). The **harness** is a code-execution loop — the model write | 68,390 | mostly simple | bounded | resumable | blueprint | open |
| LiteLLM(opens in a new tab) | One interface to 100+ LLMs; routing, caching, budgets. Not an agent framework—the pipe every agent framework uses. | 59,233 | mostly simple | n/a | retry | unknown | open |
| nanobot(opens in a new tab) | Ultra-lightweight, self-hosted personal agent framework: the **harness** is a Python daemon wiring tools, memory, and MCP into chat/webhook | 48,409 | mostly simple | n/a | n/a | unknown | unknown |
| openai-agents-python(opens in a new tab) | Handoffs, guardrails, and multi-LLM routing; minimal surface so you own the loop. | 29,582 | mostly simple | bounded | resumable | build | open |
| smolagents(opens in a new tab) | Code-as-action agents: model outputs Python executed in sandbox (E2B, Modal, etc.); ~1k LOC core. | 29,413 | mostly simple | bounded | none | unknown | open |
| letta(opens in a new tab) | Python agent runtime with tool use and control flow; lean API; stateful agents with long-horizon memory. | 24,809 | mostly simple | headless | durable | blueprint | open |
| botpress(opens in a new tab) | Visual bot builder and runtime; multi-channel, open-source alternative to commercial bot platforms. | 14,919 | complex | headless | resumable | managed | open |
| PraisonAI(opens in a new tab) | Autonomous multi-agent teams with a single entry point; emphasis on minimal config. | 9,080 | mostly simple | bounded | none | blueprint | open |
| strands-agents(opens in a new tab) | Model-driven Python SDK; decorators for tools, native MCP, multi-agent; "minimal code" without sacrificing provider choice. | 7,377 | mostly simple | bounded | resumable | unknown | open |
| youtu-agent(opens in a new tab) | Tencent Cloud's agent framework: a minimal tool-calling **harness** designed to perform well with open-source models, positioned as a lighte | 4,614 | mostly simple | bounded | retry | blueprint | unknown |
| AgentSilex(opens in a new tab) | ~300 lines of readable agent code on top of LiteLLM; the "I want to see the whole loop" option for learning or minimal production. | 456 | super simple | bounded | none | build | open |
| SuperAgentX(opens in a new tab) | Lightweight multi-agent orchestrator with an AGI-angle; minimal surface, docs-first, for teams that want orchestration without the kitchen s | 203 | mostly simple | bounded | none | blueprint | open |
How the derived tables are filtered
Enterprise harnesses
Include if the license signal is open-source AND at least one of: strong sandboxing (container/VM/OS-level isolation) or durable recovery (execution state survives restarts).
Small business harnesses
Include if the harness is a runnable runtime (not a skill pack, curated list, benchmark suite, or observability tool) AND either managed (buy, not build) or simple to adopt (installable without a platform team), AND at least 100 GitHub stars.
Personal agents
Direct category: always-on, self-hosted agents run as a daemon and talked to from chat apps. No filter.
A research-harness table is prepared but withheld: the source list is still thin. It releases when the category grows.
Full rubric wording: Methodology, Harness Tables.
Data: best-of-Agent-Harnesses (https://ryanalberts.github.io/best-of-Agent-Harnesses/), CC-BY-SA-4.0 · CC-BY-SA-4.0. Fit-by-category is InferenceIndexer's rubric applied to that data; stars and licenses are the source's facts.
Continue: For agents · Model rankings · API docs