Agents
Five agent templates ship in the Gentity catalog. Each one is its own Docker image, its own first-run experience, and its own set of model options. The pages below cover what each agent expects on first boot and which models actually work well with it in practice.
Catalog
OpenClawLocal-first AI assistant gateway. OS-agnostic. The lobster way. 🦞
2vCPU · 4096MB · 10GB · 385 models
HermesSelf-improving agent by Nous Research — persistent memory, skills, OpenRouter-native. 🪶
2vCPU · 4096MB · 10GB · 6 models
AiderPair-programming agent for the terminal. Edits files, commits to git. 🤖
1vCPU · 2048MB · 5GB · 7 models
Claude CodeAnthropic's agentic coding CLI. Reads codebase, edits files, runs commands. 🪐
1vCPU · 2048MB · 5GB · 3 models
browser-useBrowser-controlling agent. Fills forms, navigates pages, scrapes data autonomously. 🌐
2vCPU · 4096MB · 10GB · 6 models
Picking an agent
- Writing code, terminal-first: Claude Code or Aider. Aider is more focused (git-diff workflow, edit-in-place); Claude Code covers more — multi-file edits, agentic loops, MCP tools.
- General LLM gateway / harness orchestration: OpenClaw. Comes with codex, claude-code, and pi harnesses bundled — useful when you want one container that can switch between agent runtimes.
- Self-improving, memory + skills: Hermes. Persistent memory across sessions, a skills library that grows, OpenRouter-native so you can swap models without re-keying.
- Browser automation: browser-use. Headless Chromium inside the container; the agent fills forms, clicks links, scrapes structured data.