Managing servers

Agents

Creating and configuring OpenClaw agents — identity, model, skills, sandbox, API keys, Telegram channels.

What an agent is

In OpenClaw, an agent is a configured LLM persona with its own identity, model, skills, memory, and (optionally) Telegram channel bindings. EXMER exposes the full agent lifecycle as Mini App sections.

Creating an agent

On the Server page, tap Agents → + Add agent. Minimum info: an agent ID (slug like main or support). Optionally: workspace path, default model.

Identity

The first section sets the agent's human-facing identity — display name and emoji. These show up in Telegram channel messages and in the EXMER UI.

IDENTITY.md, SOUL.md, USER.md, HEARTBEAT.md, TOOLS.md

OpenClaw agents have several markdown files that shape their behavior. EXMER edits them inline:

IDENTITY.md — who the agent is, their character, style, backstory.
SOUL.md — core rules, values, what they must never do.
USER.md — who the agent is talking to (the operator's info).
HEARTBEAT.md — periodic tasks the agent should check on a schedule.
TOOLS.md — notes about the agent's environment that aren't worth a full skill.

Model + thinking

Select the underlying LLM (Claude Sonnet, Opus, or any OpenClaw-compatible model) and the default thinking level (off / low / medium / high). These are per-agent, not per-server.

Skills

OpenClaw skills are composable capability modules. EXMER's Skills section has three tabs:

Bundled — skills that ship with OpenClaw.
ClawHub — public skill registry.
Custom — skills you've uploaded yourself into workspace/skills/<name>/.

Sandbox

Configure the agent's filesystem sandbox: isolation mode, workspace access level (none / read only / read-write), and scope (session / persistent).

Tools profile

Presets that map to specific tool permission sets. Choose based on what the agent should be allowed to do.

Telegram bindings

Bind the agent to a Telegram channel so messages in that channel get routed to this agent. EXMER automates the grammy setup. Unbind at any time.

API keys

Per-agent Anthropic / OpenAI / other provider keys. Keys are stored in OpenClaw's auth profile files and never exposed back to the API response (masked in the UI).

RAG / Memory

Tap the 🧠 RAG / Memory button to open the RAG manager for this agent. See the RAG memory page for the full flow.