# The Conversation Agent Everything your agent says on LinkedIn is written by your AI agent. SDRPilot drives LinkedIn itself, the lead research and the timing, and the agent supplies the words. The **Conversation** tab on an agent's own page edits that agent without leaving SDRPilot. Open it from the [Agents](agents.md) page: click the agent you want in the table, then **Conversation**. There is nothing to pick once you are there. The agent whose page you opened is the agent you are editing, so two agents on the same LinkedIn account can be given different words. There is one record per agent, so an edit made here is live for that agent everywhere it is used. ## Before you start The onboarding wizard builds your first AI agent from your website, so there is normally nothing to set up here. If the editor cannot be loaded, the tab tells you so instead of showing an empty editor. ::: info **Champions Circle members:** your agent lives on your DM Champ account, and an edit made here is the same edit you would make in DM Champ, for every channel that agent answers on. SDRPilot works out which DM Champ agent writes an agent's replies the first time you open it, from whichever agent is set to answer LinkedIn for that account. If you have no DM Champ agent at all, create one in DM Champ first, then come back. ::: ::: ## Business context This is what the agent knows about you and it is the single biggest lever on reply quality. Fill in: - Your **company**: what you do and who you are. - Your **unique selling points** and how you are positioned against the alternatives. - Your **offer**, including what is included and what it costs if you want the agent to say so. - The **persona**: the name, the role and the tone the agent writes in. - **Rules**: what it must always do and what it must never do. - The **goal** of the conversation, for example booking a call or getting a reply to a specific question. There are no message templates and no flow builders anywhere in SDRPilot. The agent writes every message itself from this context plus what it has learned about the lead, which is why a vague business description produces vague messages. ## Knowledge base and FAQs The **FAQs and Knowledge** step is where you add answers the agent should give word for word, link knowledge sources and group them. You can also let it search the web or read specific URLs. **Link sources** opens your whole library: the website pages read during onboarding, grouped by site, and any files already in it. Tick what the agent should use and choose **Save selection**. SDRPilot has no document upload of its own, so to teach the agent what is in a price list, a product guide or a policy, add the facts as FAQs. ::: info **Champions Circle members:** you can upload your own documents in DM Champ: **AI Studio → Knowledge Base → Add source → Document** takes PDF, Word, text, Markdown, CSV and Excel files up to 25 MB each. Then link the file here under **Files**. Google Drive and Google Docs links are not read directly, so download the file as PDF or Word first. ::: ::: FAQs also have their own page in the sidebar, **Knowledge Base**. It is the same set. A question added on either page is what the agent answers from. ## Abilities Under **AI abilities** you decide what the agent can actually do besides talking: look something up, update a contact, create a task, take a booking, or call one of your own tools. You can also connect **MCP servers** here. An MCP server is a bridge to another system, for example your CRM or an internal database, and once connected the agent can use its tools inside a conversation. You choose which of its tools the agent is allowed to use. ## Draft and publish Edits land in a draft band first. The bar at the top shows what is different from the live version, so you can publish when you are happy or discard and start again. Tags are the exception: they save straight away and the tab tells you so. ## Try it out **Try-out** is a sandbox chat with your agent. It uses the same instructions, the same FAQs and the same abilities as the live agent, so it is the honest way to check a change before publishing. Say the things your leads say, including the awkward ones. Above the chat, **Test with a fake lead** runs a made-up person through the whole pipeline. Pick one of the people it writes from your ideal customer profile, or write your own, choose how they turned up (they asked to connect, commented on your post, viewed your profile, or came from a search), and you get the score against this agent's ideal customer profile, the invitation note and the first message it would send. Nobody real is involved and nothing is sent to anyone. You can hand that conversation into the chat below and carry on as the lead. That replaces whatever is already in the sandbox, so it asks first when there is something there. ## Active hours and follow-ups **Active hours** decide when the agent may reply, in your own timezone. **Follow-ups** are the short, AI written nudges it sends when a lead goes quiet. ## When the agent stops and hands over The agent does not try to push through everything. When a conversation needs a person, for example a lead asks something the agent cannot answer or asks for a human, it stops replying and flags the thread. Those threads sit under **Attention** in the [Inbox](../inbox/inbox.md), where you can take over and answer yourself. You can also take over at any time, for any reason, and hand the conversation back when you are done. ## Concluded conversations Archiving a conversation in the Inbox concludes it. Un archiving brings the thread back as a human's conversation, so the agent never quietly starts replying again on a thread you had put away. ::: info **Champions Circle members:** archiving here closes the chat in DM Champ too, and closing the chat in DM Champ archives it here. ::: :::