Agents

The roster: every agent, online or offline, with its emoji, model, and status. From here you create agents (from a ready-made archetype or from scratch), start, stop and restart them, and open one for editing.

Stopping is not deleting. A stopped agent keeps its config, workspace, memory and history, and stays offline across runtime restarts until you start it again. Prefer stopping over deleting when you just want an agent out of the way.

Opening an agent gives you a detail page with these tabs:

Tab	What you set there
Overview	Name, emoji, colour, avatar, description, model, system prompt, and the core behaviour switches.
Subagents	Specialists this agent can delegate to (see Agent-to-agent and the org chart).
Tools	Which tools the agent may use, by category, with a description of each.
Connectors	Messaging channels — Telegram, Slack, WhatsApp, email and the rest (see Channels and connectors).
MCP	External MCP servers this agent connects to, with live connection health (see MCP servers).
Webhooks	Inbound and outbound webhooks (see Webhooks).
Commands	Which slash commands are exposed, plus your own custom ones. What you enable here is what the messengers' slash menus offer; changes apply without a restart.
Monetization	Paid subscriptions for this agent's end users (see Subscriptions and monetization).
Multi-user	Per-end-user isolation for a shared agent (see Multi-user: two different features), plus Message policy — message length caps, silence rules and a reply-length ceiling, which apply to any agent (see agent.json5 reference).
Skills	Reusable procedures the agent can follow. Every skill opens: see Reading a skill.
Talk to Agent	Who may message this agent and who it may message, plus its place in the org chart (see Agent-to-agent and the org chart).
Pages	Published pages this agent serves.
Variables	Per-agent environment values.
Advanced	Backend/sandbox, ACP, budgets, prompt mode, iteration limits.
Config	The raw agent.json5, when the raw editor is enabled.

Most changes need a Restart of that agent to take effect. The UI tells you when one is pending.

Reading a skill

The Skills tab lists every skill this agent can actually reach, and each row carries two labels that tell you what you are looking at before you open it:

• Where it lives. Common skill — the shared set every agent on the box loads. Agent skill — this agent's own, in its workspace. Learned skill — one the agent wrote for itself. Configured skill — from an extra directory named in the agent's skills setting.
• Whether you can change it here. Editable or Read-only. Common skills are read-only from an agent's page on purpose: they are shared, so editing one from a single agent would quietly change what every other agent loads. Change those in the shared skills folder on the box, or ask AgentFather (olano_install_skill for every agent, olano_create_skill for a new one).

Switching a skill on or off

Every row also carries an on/off switch. It decides whether this agent loads that skill — nothing more:

• Off means the skill is not described in the agent's prompt, cannot be delegated to, and is ignored by the agent's self-learning. The agent behaves as though it does not have it.
• The skill is not deleted. It stays on disk and stays in this list, marked Off, so you can switch it back on whenever you want.
• Turning off a common skill affects only this agent. Every other agent on the box keeps it. That is the whole reason the switch exists: common skills live in one shared folder, so before this the only way to take one away from a single agent was to delete it for everybody.
• All skills are on by default.

Changes save immediately and apply the next time the agent restarts — the page reminds you when a restart is pending. Under the hood this writes the agent's disabled_skills list (see agent.json5 reference), so you can also set it by hand in the raw config.

A row marked Needs admin tools has no switch. That skill declares the tools it is written for (the AgentFather playbooks olano-admin and olano-agent-builder need the olano_* tools) and this agent does not have them, so it is never described in the agent's prompt whatever the switch would say. Granting the tools — the agent's Tools tab, or tools in its config — is what makes it appear; Turn all on leaves such rows alone.

An agent can work its own switches from a conversation too ("stop using the spreadsheet skill"), if Manage its own skills is on in its self-configuration settings — see Ask the agent: self-configuration, magic links and admin tools. It can list its skills and switch them on and off for itself; it can never install, edit or delete one that way.

Click any row to open the skill. The viewer takes most of the screen, because a SKILL.md is a document you are meant to read, not a summary to glance at. You get it rendered — the actual instructions the agent follows — with the YAML header folded away above it, and a list of everything bundled with the skill beside it: scripts, forms, reference documents, example images. Click a file to preview it; images render inline, and anything the browser cannot show offers a link to open it raw. On a phone the file list becomes a short index above the text.

Under the labels, every skill explains in a line where it lives and what that means for changing it. Editable ones also get an Open in Files button that jumps straight to the folder in the Files tab, which is where you edit it.