Kortix is the open source OpenClaw alternative to pick when the agent work belongs to a company rather than a laptop. In Kortix, every agent, skill, memory file, connector and trigger lives in one git repo you own, each session runs on its own isolated cloud computer, and nothing reaches the main branch until a person merges the change request. OpenClaw is the self-hosted personal assistant these alternatives are measured against, and the projects below cover the other reasons teams look past it: a self-improving agent, container isolation, a tiny Rust runtime and a coding-agent control center.
How we rank these alternatives
Three questions separate these projects for real work. Self-hosting is one of them, and it is not the most important.
Where the configuration lives decides what you actually own. When agents, skills, memory and connectors are files in a git repo, you can grep the whole company, diff any change and roll one part of it back. When they sit in an application database or a single config file, the machine or the vendor holds the source of truth.
Who reviews the change decides how much you can trust the agent. A capable agent can edit production. The difference is whether its output lands directly or waits behind an approval gate a person reads as a diff before it merges.
Where it runs decides the operational and compliance burden. Self-hosting on a laptop, in your VPC or on-prem changes who patches the host, who holds the keys and what an auditor can see.
The shortlist, ranked
1. Kortix: the open source alternative for a whole company
Best for: a company that needs to own every agent, skill, memory file and connector, and approve every change before it lands.
Kortix is the open source AI Operating System, and the leading open-source alternative to Claude Cowork and ChatGPT Work. It is built on the idea that a company should own all of it: every agent, all of their data, every skill, every connector, the memory and the whole configuration, on your own infrastructure if you want it there. Kortix is open source (Elastic License 2.0) — self-host, read and modify the code.
Six things set Kortix apart from every project on this list:
- The company is one git repo. Agents and skills are markdown, memory and connector configuration are files, and kortix.yaml declares the machine image, the connectors and the triggers.
- Every tool the company runs on. 3,000+ apps, plus any MCP, OpenAPI, Postman, GraphQL or raw HTTP API, with connector credentials brokered server-side and never entering the machine.
- Allow, ask or block on each tool call, down to the arguments it was given.
- Any model with your keys. Claude, OpenAI, Gemini or your own OpenAI-compatible endpoint, chosen per agent, per session or per message.
- One isolated Linux computer per session, each on its own branch, thousands running in parallel.
- One gate to land work. Sessions start from web, Slack, Teams, email, mobile, CLI or API, or from cron and signed webhooks, and the result lands as a change request a human merges.
You can run Kortix on a laptop, a VPS, your VPC or on-prem, or use the managed cloud. The platform lives in the Kortix on GitHub repository, the documentation covers install and self-hosting, and the head-to-head is in the full Kortix vs OpenClaw write-up.
2. OpenClaw: the self-hosted baseline
Best for: a solo user or small team that wants a private, always-on assistant on their own hardware, across the chat apps they already use.
OpenClaw is an open source AI assistant that runs on your own computer and meets you where you talk: Discord, iMessage, Slack, Microsoft Teams, Telegram, WhatsApp and more than 20 other channels, with native apps for macOS, iOS, Android, Windows and Linux. It is MIT licensed and stewarded by the OpenClaw Foundation, an independent 501(c)(3), with no paid tier, no hosted service and no token. A single Gateway runs it as a personal assistant on a laptop or as a shared team deployment, and the only difference is configuration.
OpenClaw's models and agent harnesses are plugins you swap without changing anything else, and it works with hosted and local providers. Its security is application-level, built on allowlists, pairing codes and DM pairing; tools run on the host for the main session unless you configure sandboxing. There is no review gate between the agent and the result: it acts, and you get the output. The project's own site and repository state all of this plainly, including the advice to read the security and sandboxing guides before exposing the Gateway to other users. If you are still deciding whether a personal assistant is what you need, what OpenClaw is explains the model in full.
3. Hermes Agent: the self-improving agent you host yourself
Best for: developers who want an agent that learns their workflows and runs on their own infrastructure with any model.
Hermes Agent is an MIT-licensed AI agent from Nous Research that creates skills from experience and improves them during use, instead of starting from scratch each session. It connects Telegram, Discord, Slack, WhatsApp, Signal, email and a terminal interface from one gateway process, keeps memory in a searchable store with cross-session recall, and runs scheduled automations written in natural language.
Hermes Agent is model-agnostic: Nous Portal, OpenRouter, OpenAI or your own endpoint, switched with a command and no code changes. It deploys on a $5 VPS, a GPU cluster or serverless infrastructure, with seven terminal backends including local, Docker, SSH, Modal and Vercel Sandbox, and it ships a migration command for people arriving from OpenClaw. Configuration is a config file on your machine or server. Hermes Agent supports command approval, DM pairing and container isolation, and it does not provide the shared company repo or the change-request gate that Kortix does. Its repository documents the learning loop and the backends.
4. NanoClaw: the container-isolated alternative
Best for: security-first users who want every agent sealed in its own container on a codebase small enough to audit.
NanoClaw is an MIT-licensed OpenClaw alternative that runs each agent in its own Linux container with filesystem isolation, built directly on Anthropic's Claude Agent SDK. Its stated reason for existing is trust: a codebase small enough to understand, one process and a handful of files, with customization by forking instead of configuration sprawl.
NanoClaw connects WhatsApp, Telegram, Discord, Slack, Microsoft Teams, iMessage, Matrix and more, gives each agent its own workspace and memory, and routes outbound requests through a credential gateway so agents never hold raw API keys. Scheduled tasks and agent templates ship in the box. The trade-off mirrors the strength: configuration and behavior are code you maintain per fork, there is no central repository of company configuration, and there is no review gate before an agent acts. The repository documents the isolation model.
5. ZeroClaw: the lightweight Rust runtime
Best for: edge, low-power and resource-constrained deployments where a Node.js runtime does not fit.
ZeroClaw is an agent runtime written in Rust and shipped as a single binary, dual-licensed under MIT or Apache 2.0. It talks to Anthropic, OpenAI, Ollama and around 20 other providers, reaches the world through 30+ channels including Discord, Telegram, Matrix, email and webhooks, and is configured in one TOML file at ~/.zeroclaw/config.toml.
ZeroClaw's security default is supervised autonomy: medium-risk operations require approval and high-risk calls are blocked, with workspace boundaries, OS-level sandboxes such as Landlock, Bubblewrap and Docker, and signed tool receipts on every action. An SOP engine runs event-triggered procedures with approval gates, and hardware support reaches Raspberry Pi and microcontrollers. All of that runs on your own machine with your keys, but as a single-user runtime it stops short of a shared company repo or a merge gate. The repository is the only official source.
6. OpenHands: the coding-agent control center
Best for: engineering teams that want a self-hosted control center for coding agents and code review.
OpenHands is an MIT-licensed developer control center for coding agents. It runs the OpenHands agent, Claude Code, Codex, Gemini or any ACP-compatible agent across local, Docker, VM and cloud backends, and lets teams add automations that trigger from Slack, GitHub, Linear or a schedule.
OpenHands defaults to running the agent server on your machine, where the agent has full filesystem access, and offers one Docker container per conversation for isolation. Its documented automated pull-request review workflow posts inline comments on your pull requests, so code review runs inside the platform while the merge decision stays with your team. What it does not cover is the rest of the company. It is a control center for code and developer workflows, not for agents, skills, memory and connectors across every team. The repository documents the backends and the self-hosting security posture.
OpenClaw alternatives compared
| Project | Open source | Where configuration lives | Review gate before changes land |
|---|---|---|---|
| Kortix | Yes: Elastic License 2.0 | One git repo you own: agents, skills, memory, connectors | Yes: work lands as a change request a person merges |
| OpenClaw | Yes: MIT | Gateway config on your own machine or server | No: agents act directly, with allowlists and pairing |
| Hermes Agent | Yes: MIT | Config file on your machine or server | Approval prompts per command, no change-request review |
| NanoClaw | Yes: MIT | A small code fork, with per-agent workspace and container | Container isolation, no change-request review |
| ZeroClaw | Yes: MIT or Apache 2.0 | One TOML file at ~/.zeroclaw/config.toml | Risk-based approval, high-risk calls blocked |
| OpenHands | Yes: MIT | Agent Canvas config and a workspace per conversation | Automated PR review with inline comments; merge stays with you |
How to choose
Pick Kortix when the agents are company infrastructure. You need the configuration in git, a human gate on every change, per-tool allow, ask or block rules, connectors across the tools the business already runs on, any model with your keys, and a choice of self-hosting in your VPC or on-prem. This is the case the other projects on the list are not built for.
Pick OpenClaw when you are one person who wants an always-on assistant on your own hardware and you are comfortable with application-level security. Pick Hermes Agent when you want the agent to build up its own skills and you will run it yourself. Pick NanoClaw when container isolation on a small, auditable codebase is the priority. Pick ZeroClaw when the hardware is small or the runtime has to be a single binary. Pick OpenHands when the work is code and pull-request review is the workflow.
If two projects still look close, ask who owns the configuration and who reads the diff. That pair of questions points to Kortix for company work, and to the others for personal or single-domain jobs. If you have already narrowed it to the two open source leaders, the Kortix vs OpenClaw comparison settles that choice, and the self-hosting guide covers running Kortix on your own infrastructure.
FAQ
Is OpenClaw open source?
Yes. OpenClaw is MIT licensed and stewarded by the OpenClaw Foundation, an independent 501(c)(3), with no paid tier or hosted service. Its licence is not the question for a company. The question is governance: OpenClaw's security is application-level and the agent acts without a review gate, so nothing stands between the agent and the result the way a change request does in Kortix.
Which OpenClaw alternative gives a person final approval before changes land?
Kortix does, for every job. An agent works on an isolated cloud computer, commits to its own branch, and opens a change request; the change reaches the main branch only when a person merges it. OpenHands automates pull-request review for code and posts inline comments, which is code-review feedback rather than a company-wide change-request gate. OpenClaw, Hermes Agent, NanoClaw and ZeroClaw act directly, with approval prompts or isolation but no merge review.
Can I run an OpenClaw alternative in my own VPC or on-prem?
Kortix runs self-hosted on a laptop, a VPS, your VPC or on-prem, or in managed cloud, and the closed platforms it competes with cannot be self-hosted. OpenClaw, Hermes Agent, NanoClaw and ZeroClaw are all self-hosted by design. Choose by whether you also need a shared company repo and a review gate, which is the case Kortix is built for.
Kortix is the open source alternative for teams that want the whole system rather than a single assistant: one repo you own, any model with your keys, 3,000+ connectors and a human gate on every change. Try Kortix to start, or go deeper with the Kortix vs OpenClaw comparison and the self-hosting guide. If you are new to OpenClaw itself, begin at the openclaw-alternative.com home page.