How Zimmer Works
Workspaces and agents
Everything in Zimmer lives inside a workspace. A workspace belongs to a company or brand and contains all your agents, knowledge bases, and data.
An agent is a single AI-powered experience — one chatbot, one use case, one deployment. You can have multiple agents in a workspace. For example: a product advisor on your homepage and a support agent on your help page.
Zimmer has two agent types. Knowledge agents are conversational — they chat with users in real time using your knowledge bases and journeys. Action agents are autonomous — they run workflows in the background on triggers like schedules, webhooks, or data changes. Both types share the same AI models, knowledge layer, and flow editor.
Templates and the visual layer
Each agent has a template — the visual appearance users see. It includes the hero section (banner image or carousel, headline, subtitle), the chat widget styling, quick reply buttons, and product cards.
You customise templates in the Template Studio. Changes are drafts until you publish.
Journeys and the conversation layer
A journey defines how a conversation flows. It's a visual node graph — you connect nodes like greeting → question → AI response → product card. The journey controls when the AI speaks freely, when it collects data, and when it shows a product.
Each agent has one active journey at a time.
Knowledge and data
Agents pull from two types of knowledge:
Knowledge Bases hold unstructured content — website pages, PDFs, FAQs, policies. The AI uses semantic search (RAG) to find relevant answers.
Product Catalogs hold structured data — products, plans, rooms, menu items. The AI uses this to recommend specific items and render product cards.
Both live at the workspace level and can be shared across multiple agents.
Projects and access control
Workspaces can have multiple projects (brands) and multiple team members. Access is role-based:
- Owner and Admin — see and manage everything across all projects
- Editor — can build and update agents, journeys, and knowledge, but only in projects they're assigned to
- Viewer — read-only access to assigned projects
Project-level membership is managed in Settings → Members. When you invite someone as an Editor or Viewer, you choose which projects they can access.
Publishing
Agents exist in Draft or Published state. Drafts never affect the live experience. When you publish, the current configuration goes live and your agent gets a landing page URL and widget embed code.
Agent Groups and multi-agent coordination
Agents can work together through Agent Groups. When agents share a group, they automatically share context — a Knowledge agent can see recent run results from Action agents in its group, giving it awareness of what's happening across your automation.
You can also use the Trigger Agent node in the journey builder to have one agent invoke another directly. This works in two modes:
- Sync — wait for the triggered agent to finish and use its result
- Async — fire-and-forget, continue the flow immediately
Manage groups from the Groups section in any agent's Settings tab.