Self-hosted AI agents move fast in 2026. Projects pivot governance, ship breaking changes, get acquired or fade. Migration is now part of the ops budget.
Migrating from one self-hosted AI agent to another is mostly about three things in 2026: porting tool definitions, re-encrypting credentials in the new vault, and re-registering MCP servers under the destination project's spec.
The OpenClaw to Hermes Agent path is the most documented: Hermes ships a command for it. hermes claw migrate shows a preview, writes a backup of ~/.hermes, then imports the OpenClaw persona, memory files, skills, model and provider settings and approval mode, plus the API keys if you add --migrate-secrets. An OpenClaw Docker sandbox maps to Hermes Agent's Docker terminal backend; without one, Hermes runs commands on the host, as OpenClaw did. Smoke test, then decommission the old install. Until October 2026 this paragraph described a manual port of tool YAML files with new sandbox declarations; Hermes documents no such files.
OpenClaw to Nanobot is a different kind of move: you trade a polished plugin marketplace for a lighter personal agent with a browser UI, and you switch on its shell sandbox yourself. Custom plugins port with effort. Built-in features like browser automation don't.
OpenClaw to ZeroClaw is the move to a single Rust binary with approvals and an OS sandbox on by default. Pair it with Ollama if you also want the model on your machine, and accept the capability hit on multi-step planning that comes with a local model. Until October 2026 this paragraph called it the privacy migration that drops cloud models: ZeroClaw works with cloud providers as readily as with Ollama.
Hermes to IronClaw is the move to tools sandboxed from the first run: WebAssembly tools that only reach what they declare, secrets injected outside them, a log of every tool execution. Check which of your tools exist as IronClaw extensions before you move. Until October 2026 this paragraph promised gVisor and RBAC, which IronClaw does not document.
Each of these has its own playbook. Start with the migration that fits your departure point, follow the explicit steps, don't skip credential rotation.
What this page says about each agent's defaults was read in its own documentation on 2026-10-18: Hermes Agent docs: migrate from OpenClaw, Nanobot README (v0.3.5), IronClaw README (main, b0b999d), ZeroClaw README (v0.8.6).
Guides
- How to migrate from OpenClaw to Hermes Agent: step-by-step: Migrate from OpenClaw to Hermes Agent with hermes claw migrate: preview, import, what gets archived, where commands run, re-pairing, retiring OpenClaw.
- The OpenClaw timeline: from Clawdbot to a foundation (late 2025 to July 2026): OpenClaw's history from sources: Clawdbot in late 2025, two renames in January 2026, CVE-2026-25253, 135,000 instances exposed, its creator's move to OpenAI and the OpenClaw Foundation of July 2026.
- 5 best OpenClaw alternatives in 2026, compared: Hermes Agent, Nanobot, NanoClaw, IronClaw and ZeroClaw compared on setup, security model, model support and cost: which OpenClaw alternative fits your threat model.
Agents on this topic
- OpenClaw: Self-hosted AI agent at the centre of the January 2026 security crisis. Its tool sandbox is off by default: the main session's tools run on the host until you configure one.
- Hermes Agent: Agent from Nous Research (MIT, Python). Runs shell commands on the host by default, with an approval step for dangerous ones; Docker and cloud sandboxes are options. Works with many LLM providers.
- Nanobot: Personal AI agent in Python (MIT) with a browser UI and chat apps. Native Anthropic, OpenAI and OpenRouter providers plus local servers; its shell sandbox is off until you switch it on.
- ZeroClaw: Single Rust binary from ZeroClaw Labs, MIT or Apache-2.0. Works with Anthropic, OpenAI, Ollama and about twenty other providers, and turns on an OS sandbox it detects.
- IronClaw: Rust agent from NEAR AI, MIT or Apache-2.0. Untrusted tools run in a WebAssembly sandbox with declared capabilities, and secrets are injected outside them.
Terms
MCP (Model Context Protocol) · MCP server · Tool call · AI agent