# What TulipFarm is > TulipFarm is a control panel for business operations carried out by AI agents. Learn how chat, definitions, data, and approvals fit together. Source: https://tulipfarm.ai/docs/concepts/what-is-tulipfarm/ TulipFarm is a control panel for business operations carried out by AI agents. People ask for an outcome in ordinary language. The product can define the data they need, configure a specialist agent, and turn repeatable work into an automation. People then use chat and generated screens to work with that system. A business user can build and operate their system without writing code or editing configuration files. Those files exist underneath, and agents maintain them through product tools. ## Who it is for Business users who build and run business operations through chat without writing code. The audience is not restricted to an industry, company size, or specific job title. ## The work it is meant to handle Examples include customer and sales tracking, support tickets, employee directories and leave, expenses and invoices, and engineering operations. These are examples of systems people can build from shared capabilities, not a list of separate prebuilt applications. A request such as "track our customers and follow up when we have not heard from them" brings several parts of TulipFarm together: 1. Define customer and follow-up tables with fields and relationships. 2. Store actual customers and follow-ups as records. 3. Give an agent instructions, relevant skills, and limited access. 4. Connect the external account needed for the work. 5. Use knowledge such as contact policy or approved messaging to guide the agent. 6. Run the work on request or through a scheduled or event-triggered workflow. 7. Ask for human approval where required and record what actually happened. This is an illustrative product journey. ## Chat builds and operates the business system Chat is where people see progress, answer questions, approve actions, inspect results, and stop work. A browser integration is not the job itself. Work is persisted as runs and events so it can wait and resume independently of the UI. ## Definitions and business data have different lives TulipFarm stores authored definitions and their history: what tables, agents, skills, and automations exist. The runtime loads those definitions and does the work. Records, messages, execution history, knowledge, and uploaded files have their own storage. A customer definition describes valid customer data. A customer record holds one customer's actual information. Changing an agent's instructions changes a definition; sending an email is an action with a separate outcome. Saving a definition is not enough to call it active. See [Saved, activated, and running](https://tulipfarm.ai/docs/concepts/saved-activated-running/index.md). ## Reusable agents, skills, and workflows An **agent** is a named worker with instructions and permissions. A **skill** is a procedure an agent can use. A **workflow** describes repeatable work and how it starts. For example, a support agent can use a ticket-triage skill during a daily triage workflow. The skill does not grant access to tickets. The agent and the initiating person or service must already have the required authority. Automations can combine deterministic computation, tool calls, AI judgment, and human decisions. See [Agents, skills, and workflows](https://tulipfarm.ai/docs/concepts/agents-skills-and-workflows/index.md). ## Knowledge and memory support the work Knowledge holds reusable business information and source-backed answers. Uploaded files can be read in chat without becoming searchable knowledge. Personal memory is different. It is a member-authored document scoped to one workspace membership. It belongs only in that member's private chats and private tasks, not in shared project context. Projects hold human-maintained notes, instructions, and files shared by selected participants. Project context is not automatic memory. See [Files and Knowledge](https://tulipfarm.ai/docs/concepts/files-and-knowledge/index.md). ## Autonomy has boundaries TulipFarm's product model gives agents explicit tools and limited authority. Permission checks, guardrails, approvals, and execution records control what they can do. An instruction in a prompt is not a substitute for a permission check. Personal integrations, private chats, and raw credentials stay with the person who owns them. Sharing an answer requires attention to the audience as well as the person who asked for it. See [Approvals](https://tulipfarm.ai/docs/concepts/approvals/index.md). People describe work, agents operate on business data and connected services, reusable definitions make that work repeatable, and people can inspect and control the result.