The Two Assistants Everyone Is Self-Hosting in 2026
If you want a personal AI assistant that lives on your hardware, answers in your chat apps, and remembers everything without sending your life to a SaaS company, two open-source projects dominate the conversation: OpenClaw ("the AI that actually does things") and Hermes Agent by Nous Research ("the agent that grows with you").
Both are free to download. Both are bring-your-own-API-key. And that means the real comparison isn't the price tag — it's features, architecture, and what your token bill looks like once the agent is doing real work.
Feature Comparison at a Glance
| OpenClaw | Hermes Agent | |
|---|---|---|
| License | Open source, free | Open source, MIT |
| Runs on | Your Mac/Windows/Linux machine | Your own server (macOS/Linux/Windows) |
| Chat channels | WhatsApp, Telegram, Discord, Slack, Signal, iMessage | Telegram, Discord, Slack, WhatsApp, Signal, Email, CLI |
| Integrations | 50+ (Gmail, GitHub, Spotify, Obsidian, Twitter…) | Web search, browser automation, vision, image generation |
| Skills | Self-hackable — writes its own skills; NVIDIA-backed Skill Cards for security | Self-improving — auto-generates new capabilities |
| Parallelism | Background tasks, cron jobs | Isolated parallel subagents |
| Sandboxing | Local-first, opt-in auto-execution policies | Local, Docker, SSH, Singularity, Modal backends |
| Models | Claude, GPT, local models | Bring your own API key |
The Cost Is the Same Question for Both
Neither project charges a cent. Your entire running cost is the LLM API behind it — so the monthly bill depends on usage volume and model choice, not on which assistant you picked. At June 2026 prices:
| Usage Pattern | Tokens/Month (in / out) | DeepSeek V4 Flash | Claude Opus 4.8 |
|---|---|---|---|
| Light — a few questions a day | 1.8M / 0.24M | ~$0.32 | ~$15 |
| Daily driver — ~50 exchanges/day | 9M / 1.2M | ~$1.60 | ~$75 |
| Heavy — subagents, browsing, automations | 30M / 4M | ~$5.30 | ~$250 |
The spread is the story: the same assistant doing the same work costs anywhere from pocket change to a car payment depending on the model behind it. Both projects let you mix models, so the smart configuration is identical for both: a capable flagship for the main loop, a budget model for background tasks and subagents. We ran the full numbers in How Much Does Hermes Agent Cost to Run?
Which One Should You Pick?
Pick OpenClaw if you live in iMessage (it's the only one of the two that supports it), you want the larger plug-and-play integration ecosystem, or you want your assistant on the machine you already use daily. Its Skill Cards security model is also the more mature answer to the "my agent installed a malicious skill" problem. Start with our OpenClaw Setup 101 guide.
Pick Hermes Agent if you want a server-side agent with serious workload features — parallel subagents, multiple sandboxing backends (Docker, SSH, Modal), email and CLI access, and scheduled automations as a first-class concept. It's the more infrastructure-minded of the two.
Either way, the model decision matters more than the assistant decision. Browse the live pricing table, use the Model Picker to match a model to your workload, and estimate your loop costs with the Agent Cost Calculator.