--- **date:** 2026-07-01 **status:** Draft **description:** Short presentation of the Workspace menu for configuring LLM providers and documentation agents. **tags:** workspace, agents, llm, mcp, providers, automation, documentation --- ### **Workspace** Menu The Workspace menu lets you create your own documentation agents. These agents can help you write, translate, summarize, improve, or transform your **Markdown** documents. ![Workspace connecting Living Documentation, LLM providers, agents, MCP, and an image provider](/images/DOCUMENTATION/concept-07-workspace-providers-agents.jpg) ### What is the Workspace for? The Workspace connects three elements: - an **LLM provider**: the AI model being used - an **agent**: a reusable business instruction - the **Living Documentation tools**: read a document, update it, generate an image, and so on The idea is simple: instead of rewriting the same prompt every time, you create an agent once, then run it again whenever you need it. --- ### Main nodes In the Workspace, each element is represented by a node. | Node | Role | | ------------------------ | ----------------------------------- | | **Living Documentation** | The center of the Workspace. | | **LLM provider** | The model used by agents. | | **Agent** | A specific documentation task. | | **MCP** | Internal tools available to agents. | An agent is attached to an LLM provider. For example: ```text OpenRouter DeepSeek V4 Pro └── Agent "Improve a Markdown document" ``` --- ### Create an LLM provider An LLM provider contains the information required to call a model: - the endpoint - the API token - the model - the tool usage mode For the token, prefer using an environment variable: ```text env:LLM_API_KEY ``` This avoids writing your API key directly in the configuration. ### Tools or Chat only? A provider can work in two modes. | Mode | Usage | | ------------------- | --------------------------------------------- | | **Allow MCP tools** | The agent can use Living Documentation tools. | | **Chat only** | The agent replies without calling any tool. | If your agent must read or modify a document, it generally needs **Allow MCP tools** mode. If the model refuses tools, use **Chat only** and ask it only for a text response. --- ### Create an agent An agent is a reusable instruction. Example of a simple agent: ```text You are a documentation assistant. Read the document provided by the user. Improve the clarity of the text without changing the meaning. Preserve the Markdown. Update the document when you are done. ``` You can then run this agent by giving it a document id. ### Example uses An agent can be used to: - rephrase a rough document - fix mistakes - translate a document - create a summary - prepare meeting notes - generate an image to illustrate a document - transform text into a table - improve a Markdown document with snippets --- ### Run documents When you run an agent, **Living Documentation** creates a result document. This document contains: - the agent that was used - the provider that was used - the user input - the model response - debug information if enabled These documents are useful for keeping a trace of what the agent did. They are saved in the agent workspace, for example: ```text AI/WORKSPACE/my_agent/ ``` --- ### Image providers Some LLM providers can be configured in **Image generation** mode. They are used to generate images from an agent, for example to illustrate a document. In that case, the provider has a technical id that the agent can use as `imageProviderId`. > Images generated by agents are stored separately from regular attachments. --- ### Simple advice Start with one LLM provider and one agent. When this first agent works well, create more specialized agents: translation, summary, correction, illustration, procedure generation, Markdown improvement.