Agent Serve · Chat
Chat works with your tables, files, knowledge bases and connected tools. Answers come with their sources, and charts with the data behind them.
Agent Serve Chat is the conversational surface of Agent Serve, the AI workspace where teams build, deploy, and manage AI agents. Teams ask questions in natural language, see the sources behind answers, generate charts and documents from workspace data, and build, run and deploy workflows and tables by describing them.
How it works
Agent Serve Chat keeps the work close to the conversation: the context it used, the result it made, and the next step.
Type or paste the question the way you would ask a colleague, or start from a Playbook.
Type @ to add a table, a file, a folder or a knowledge base, or attach files to the message.
Chat queries the table, searches the documents and calls connected tools to build the answer.
Save the answer as a chart, a table or a document in the Library, and share it by link.
In depth
Agent Serve Chat reads and changes the same workflows, tables, files and knowledge bases your team works in.
Describe the process and Agent Serve Chat proposes the blocks in the workflow builder. You choose Accept or Reject for each change.
When Agent Serve Chat answers from a document or knowledge base, source chips under the reply link back to the document it used.
Agent Serve Chat creates tables, imports data, adds rows and columns, and turns a question about the data into a chart.
Agent Serve Chat saves Word, PDF, PowerPoint, Markdown, HTML and CSV files to the Library, where the team can open and share them.
Capabilities
Ask Agent Serve Chat for a breakdown and get a chart built from the table or source you named.
“Show renewal exposure by region.”
Agent Serve Chat shows the documents it used as source chips, so anyone can check the answer.
Answer → Renewal policy
Agent Serve Chat calls the integrations your workspace has connected, and can connect new apps and MCP servers, a standard way to plug tools into AI.
Salesforce · Slack · Google Drive
Agent Serve Chat searches and reads the web, and runs code when a question needs a calculation.
“Compare these two rate sheets.”
Agent Serve Playbooks start common jobs the same way each time, such as connecting an integration or deploying a workflow.
Research · Create a table · Deploy a workflow
Agent Serve Chat generates images, audio and video, and reads attached recordings as a transcript and sampled frames.
Call recording.mp4 → transcript
Examples
Three requests analysts at banks, insurers and payment companies bring to Agent Serve Chat, each ending in something they keep.
An analyst mentions the disputes table in Agent Serve Chat, asks for the top merchants by chargeback rate, and shares the chart.
An underwriter attaches the submission folder in Agent Serve Chat and asks for a brief checked against the underwriting guidelines knowledge base.
An operations lead describes the KYC refresh process, and Agent Serve Chat proposes the workflow, which the lead accepts and deploys.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
Under the hood
| Models | Brain Auto picks for you, or choose Brain Fuji (most capable), Brain Hikari (deepest reasoning), Brain Sora (balanced) or Brain Hayate (fastest). |
|---|---|
| Effort | Quick, Balanced, Deep, Deeper and Max. |
| @mentions | Workflows, blocks, tables and table selections, files and folders, knowledge bases, integrations, MCP servers, past chats, run history, Playbooks and docs. |
| Attachments | Documents, code, images, zip archives, audio and video. Audio and video are read as a transcript and sampled frames. |
| Files it creates | .docx, .pdf, .pptx, .md, .html, .csv and chart files, saved to the Library. |
| Built-in Playbooks | Connect an integration, research, create a table and deploy a workflow. |
| Message actions | Copy message, Good response, Bad response and Fork in new chat. |
| History | Grouped by Pinned, Today, Yesterday, This week and Earlier. |
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| A question about the data | BeforeA ticket to the data team and a wait for a spreadsheet. | With Agent ServeMention the table in Chat and get the figure and the chart. |
| Trusting an AI answer | BeforeA confident paragraph with nothing to check it against. | With Agent ServeSource chips that link to the documents behind the answer. |
| Turning an answer into a deliverable | BeforeCopy the reply into a document and format it by hand. | With Agent ServeChat saves the .docx, .pdf or .pptx to the Library. |
| Automating the next request | BeforeExplain the process again to whoever builds the automation. | With Agent ServeDescribe it once and accept the workflow Chat proposes. |
Governance
Agent Serve Chat works under the same roles, access policies and data limits as the rest of the workspace.
Security and governanceAgent Serve Chat uses the resources and tools available to the person asking. People with a limited data view are kept to tables and dashboards that apply their limit.
Agent Serve access policies let admins hide Chat for a group, along with the models, blocks and integrations that group can use.
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Sign in through your identity provider with SAML 2.0 or OIDC, and provision people with SCIM.
Run Agent Serve in your own cloud or data centre, next to the systems it reads.
Works with
Build, edit, run and deploy workflows from a conversation, and accept each change in the builder.
Create and query tables, add rows and columns, and chart the result.
Search knowledge bases, cite the source, and add new documents to them.
Attach files and folders as context, and keep the documents Chat writes in the Library.
Call the tools your workspace has connected, from Salesforce to Slack.
Further reading
Most complaint breaches are visible weeks before they happen, in a queue nobody has time to read. Here is a setup that computes every deadline, flags risk early and puts the at-risk list in front of the right person.
Retail banking service · 12 min readThe morning spreadsheet is out of date before the meeting starts, and two people bring two RTO numbers. This playbook moves daily decisions onto dashboards that refresh every 15 minutes or query live, with one definition per number.
Ecommerce leadership · 5 min readPerformance teams spend their mornings pulling the same numbers out of Ads Manager. These field notes cover connecting Meta ads to Agent Serve, asking questions in Chat, and letting a Relay send a daily review that a person acts on.
Performance marketing · 6 min readQuestions
Agent Serve Chat is the conversational surface of Agent Serve. Teams ask questions in plain language and build, run and deploy workflows, tables and documents by describing them.
Agent Serve Chat works with connected tools and workspace resources such as files, tables, workflows and knowledge bases. The resources and permissions available to the person asking decide what it can use.
Yes. When Agent Serve Chat uses a document or knowledge base, the answer shows source chips that link to the document it used.
In the workflow builder, Agent Serve Chat proposes changes and shows Accept and Reject, so a person decides what goes into the workflow.
Agent Serve Chat offers Brain Auto, which picks a model for each request, or a fixed choice of Brain Fuji, Brain Hikari, Brain Sora or Brain Hayate, with an effort level from Quick to Max.
Agent Serve Chat saves Word, PDF, PowerPoint, Markdown, HTML and CSV files and charts to the Library. Spreadsheets in .xlsx format are not generated today.
Last updated
Enterprise access
Agent Serve is available to teams by invitation. Tell us what you want to build and we will set up your workspace with you.