Work with the agent

The agent is one chat that follows you through Staffer. Open it from any page and it knows where you are: on a job, it works on that job. Anywhere else, it can look across your whole workspace, but only at the jobs you can see. See Who can see a job.

​

Open the agent

• Click Agent in the bar at the bottom right of any page. It always starts a new chat, unless the one that’s open is still empty. To go back to an earlier chat, click its tab in the bar.
• Click Agent in the sidebar to start a new chat full screen. The workspace home page is the same full-screen chat.

​

Three ways to see it

The chat opens in a panel in the corner, over the page you’re on. You can also dock it beside the page, or open it full screen. Docked or floating is remembered in your browser, so it stays the way you left it after a reload. It doesn’t follow you to another computer. In full screen, click Exit full screen at the top right to go back to the page you came from, with the chat where it was. The button only shows when you came from another page.

​

What it can see from here

A Context line sits above the message box and says what the agent can see. On a job it shows the job’s name and the tab you’re on, for example “Context: Senior Backend Engineer · Brief”. On a segment it shows the segment’s name, and on a project it shows Project, so the agent knows which one you mean. On other pages it shows the page, such as Talent pool or Segments. In full screen it says Whole workspace. On a job, the agent drafts the brief, searches, shortlists and changes the process for that job. Open a different job and ask again in the same chat, and it works on that one instead. With no job open, it looks things up across your workspace. Ask which jobs need attention this week or for a summary of your pipeline across open jobs. It can also open a page for you: the chat follows you there, in the corner. See What the agent can do for what it’s allowed to change.

​

Start a job from any chat

With no job open, describe the job you’re hiring for, or paste a link to a job post you already have. The agent asks a few questions first, then drafts the brief, creates the job and opens it. See Create a job.

​

Conversation starters

A new chat in the corner panel suggests a few things to ask. Click one to send it.

• On a job: “Summarise where this role stands”, “Find more candidates for this role”, “Who in the pipeline needs a follow-up?” and “Draft an outreach message for this role”.
• Everywhere else: “Help me write a brief for a role I’m hiring for”, “Which of my roles need attention this week?” and “Summarise my pipeline across open roles”.

​

Your chats

Your chats are private. Nobody else in the workspace can open them, admins included. Every chat is saved as you go, and the agent gives it a name once it’s clear what it’s about. Click the name at the top of the chat to rename it. Once you’ve renamed a chat, the agent stops renaming it.

• Recent chats. Your latest chats show as tabs in the bottom bar, just left of Agent, up to four. Click a tab to reopen that chat. Hover over a tab and click the × to take it off the bar. That doesn’t delete it: it stays in your history and comes back to the bar when it gets a new message.
• Chat history. Click the clock icon next to Agent to search your chats by name, start a New chat or reopen an older one. Chats are grouped into Today, This week and Older, and the list holds your 50 most recent chats.

Because chats are private, your team doesn’t see what you asked the agent or what it told you. Put anything the team needs in the job’s notes.

​

Attach files

Click + (Add to message), then Attach a document or image. You can attach up to 5 files to a message, up to 3 MB each. The agent reads PDF, Word, Excel, PowerPoint, OpenDocument, RTF, CSV, text, Markdown, JSON and HTML files, and PNG, JPEG, WebP and GIF images. Large images are scaled down for you. A scanned PDF with no text in it is sent as an image, so the agent can still look at it. A document’s text counts toward the message’s 8,000-character limit, so the agent only reads the start of a long document. Older .doc, .xls and .ppt files can’t be read: save them as .docx, .xlsx or .pptx first.

​

Voice input

Click the microphone (Record voice message) to dictate instead of typing. It works in every chat in Staffer and in every browser, including Arc, Brave and Dia. Your words appear in the message box as you speak, and Norwegian is recognized reliably. Holding ⌘D (Ctrl+D on Windows) in any chat composer works as push-to-talk: release the keys and the transcript stays in the message box, ready to send. If dictation can’t carry on, for example because the microphone is blocked, you’re out of credits or the connection drops, it tells you why and keeps what it already transcribed.

​

Ask permission or Auto

The switch next to + shows your mode: Ask permission or Auto. In Ask permission, the agent shows an approval card before actions that change things. In Auto, it goes ahead. See What the agent can do.

​

Message length and stopping a reply

A message can be up to 8,000 characters. A counter appears once you’re close to the limit, and going over it disables Send rather than cutting your text short. While the agent is replying, the send button becomes Stop. Stopping ends the reply in your chat. A search that’s already running keeps going in the background, and its results still land on the Scouting tab when it finishes. To stop the search itself, use Stop search on the Scouting tab. See Search for candidates.

​

What it uses to answer

Every reply draws on your company knowledge and your own Communication tone and Notes for the agent (set in Settings → Account → Preferences). On a job, it also reads the job’s brief and notes and, if the job belongs to a team, that team’s knowledge.

​

What it can’t see

The agent can’t see your email, Slack or calendar. It only knows what’s in Staffer, so a decision made in a call or a DM is invisible to it until someone writes it down, for example in the job’s notes.

​

”Tell me more” on a candidate

Click Tell me more on a candidate’s profile panel to ask the agent about that candidate. It sends a message naming them to your current chat and opens the panel, which is what lets the agent read their full record and answer questions about them specifically.

​

Credits

Every message you send the agent uses credits, in the corner panel and full screen alike.

​

Use Staffer from your own assistant

You can also connect Claude, ChatGPT, Cursor and other assistants to a workspace, in beta. That’s a separate assistant working in Staffer for you, not this chat. See Connect Claude, ChatGPT or Cursor.

​

Common questions