> ## Documentation Index > Fetch the complete documentation index at: https://octop.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Agents: Your Personal AI Specialists in Octop > Each Octop agent has its own workspace, LLM provider, skills, channels, and conversation history. Learn how to create and manage agents. Agents are the core unit in Octop. Each user can create and run multiple agents simultaneously, and every agent operates in complete isolation — its own workspace directory, its own LLM provider settings, its own enabled skills, and its own conversation history. This lets you maintain, for example, a focused coding assistant alongside a general-purpose research assistant, without either one bleeding context into the other. ## Agent attributes When you create an agent, you configure the following attributes: | Attribute | Description | | - | - | | **Name** | A short identifier used in the UI, CLI, and channel routing | | **Persona (MBTI)** | An optional personality profile that shapes the agent's communication style | | **Default model** | The LLM provider and model the agent uses (e.g. `openai:gpt-4o`) | | **System prompt** | Free-form instructions appended on top of the persona backbone in `SOUL.md` | | **Workspace backend** | Where the agent's files are stored — local disk, Docker sandbox, OpenSandbox, COS, S3, or another named store | | **Shared** | When enabled, other users on this instance can open the agent (writes still require the owner) | > [!NOTE] > > Each agent maintains its own conversation history, channel connections, and cron jobs. Switching between agents in the web UI or CLI does not mix their histories or settings. ## Create an agent Open the web UI at `http://127.0.0.1:8088` and go to **Agents**. The page has three tabs — **My agents**, **Built-in experts**, and **Expert market** — and a **Create from built-in expert** button that seeds a new agent from a template. ![Agents](https://mintcdn.com/octop/P_aBdthWeK4YndCE/images/ui-agents.png?fit=max\&auto=format\&n=P_aBdthWeK4YndCE\&q=85\&s=a8e3d7027bcf003a0e5f47f015298feb) `octop init` creates a default `main` agent so you can start chatting immediately. That default agent uses home-directory storage, the same as agents you create in the web UI. > [!TIP] > > For CLI create, list, start, stop, and reload commands, see [Agent commands](https://octop.mintlify.site/cli/agent-commands.md). ## Agent lifecycle Each agent card shows its runtime state and carries a toggle to start and stop it, plus a reload control for picking up configuration changes. | State | Meaning | | - | - | | **Running** | The agent is loaded and handling requests | | **Stopped** | The agent is disabled and will not answer messages | | **Error** | The agent hit a startup or runtime failure; check logs for details | > [!WARNING] > > Stopping an agent terminates its active sessions on all connected channels. Any in-progress message will not receive a reply. Restart the agent to resume. > [!WARNING] > > Deleting an agent permanently removes its workspace under `~/.octop/agents//`. The web UI and CLI both warn before confirming. This cannot be undone. ## Per-agent configuration Skills, sub-agents, personas, memory, and channels are all configured per agent under **Personalization**. Pick the agent at the top of the page, then switch between the tabs: | Tab | What it configures | | - | - | | **Skills** | Installed, built-in, and marketplace skills, plus skill packages. You can copy skills between a package and this agent's workspace. | | **Sub-agents** | Helper agents this agent can delegate to. In chat, `@` mention a running sub-agent, or use `ask_agent` to open a separate thread. | | **Plugins** | Per-agent plugin toggles (moved here from the Tools panel) | | **MBTI** | The persona that shapes tone and reasoning style | | **Memory** | What the agent retains across conversations. While Octop is compacting memory, chat shows a progress banner and pauses send. | | **Channels** | Messaging platforms routed to this agent | ## Expert library Octop ships a bundled **expert library** — pre-configured agent templates for coding, research, writing, and more. Experts are templates, not running agents. Instantiating an expert creates a new, independent agent seeded with that template's configuration. Browse them under **Agents → Built-in experts** and **Agents → Expert market**, or see [Experts & Personas](https://octop.mintlify.site/concepts/experts-personas.md). ## Share an agent When you create an agent you can attach default knowledge bases and connector instances. New chats for that agent start with those selected. You can still change the selection per turn. The chat sidebar and `@` picker list **running** experts only. Clicking a skill in chat inserts `/slug` into the composer. When you create or edit an agent, turn on **Shared** so other people on this instance can open it. Only the owner can change its settings or files. Shared-expert skills are available in chat. ## Run an agent in Docker Administrators add a **Docker Sandbox** or **OpenSandbox** under **Management → Storage**. When you create an agent, pick that storage backend so the agent runs in an isolated sandbox instead of on the host disk. OpenSandbox creates a remote sandbox when the expert starts and destroys it on stop; enabling the backend installs the official SDK. * **Sandbox scope** chooses how containers are shared: one per expert, one per user, or a fixed sandbox. * Deleting the agent does **not** delete the container. * You can browse container files in the workspace viewer when the backend uses a **Fixed** scope. In the workspace viewer you can open Word (`.docx`) files as Markdown and save them back. Headings, bold, italic, lists, and tables are kept; complex formatting is simplified. ## Skills Open **Personalization → Skills** to enable skills for the current agent. Use the **Skill market** tab to install from SkillHub. Reusable sets for the whole instance live under **Skill packages** in the sidebar: import a ZIP, then attach the package to an agent, or copy selected skills either way (`copy` into the workspace, or push a workspace skill into a package). > [!TIP] > > Experts are a great starting point when you want a task-specialist agent. Pick the closest template, instantiate it, and then fine-tune the system prompt and skills rather than building from scratch. ## Next steps > [card] [Experts & Personas](https://octop.mintlify.site/concepts/experts-personas.md) > > Learn how MBTI personas shape agent behaviour and how to use the full expert template library. > [card] [Channels](https://octop.mintlify.site/concepts/channels.md) > > Connect your agent to Feishu, WeCom, DingTalk, Telegram, and other messaging platforms. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.