Unique AI Conduct

Overview

Unique AI Conduct is an autonomous AI agent that can tackle complex, long-running tasks on your behalf. Unlike Unique AI Chat — which provides quick, conversational answers — Conduct gives the agent a full coding environment, access to your organization's tools, and the ability to plan and execute multi-step workflows independently.

Think of Chat as asking a colleague a question. Conduct is more like handing a colleague a project brief and letting them work through it — writing code, processing files, searching your knowledge base, and delivering finished results.

The key difference from Chat: Conduct doesn't just tell you how to do something — it does it. The agent has access to a full coding environment and can run long-running tasks that would be impractical in a basic chat interaction.

Who is it for?

Data analysts, developers, compliance officers, operations teams, and any professionals whose work involves data processing, file manipulation, code execution, or complex multi-step tasks. Conduct is for teams that need the AI to produce deliverables — not just answer questions.

If you're unable to access certain features or sections of this article, it's possible that your firm doesn't have access or hasn't upgraded to the latest version. Please reach out to your internal support team for further assistance.

When to Use Conduct

Use Conduct when your task requires the agent to do work, not just respond:

• Data analysis and transformation — "Take this CSV, clean the data, calculate monthly growth rates, generate charts, and produce an Excel report" — Conduct writes and runs the Python to do it all.

• Document processing at scale — Extract tables from PDFs, convert file formats, merge datasets, or generate formatted reports from raw data.

• Multi-step projects — Tasks that require planning, iterating, and building on results across multiple steps — like building a financial model, running statistical forecasts, or prototyping an algorithm.

• File generation — Create deliverables: Excel workbooks, charts, cleaned datasets, HTML previews, or processed documents ready for downstream use.

• Research and synthesis — Search your organization's knowledge base, pull information from multiple sources, and synthesize findings into a structured output.

Use Chat instead when you just need a quick answer, a summary, or a simple conversation.

How Conduct Works

When you give Conduct a task, the agent:

1. Plans the approach — Breaks your request into steps, deciding which tools and code to use

2. Executes autonomously — Writes and runs code in a sandboxed environment (Python and Bash), searches your knowledge base, and processes files — all without needing step-by-step guidance from you

3. Handles complexity — If a step fails, the agent recovers and tries alternative approaches. It can iterate on results, refine its output, and chain multiple operations together

4. Delivers results — Returns finished files, charts, analysis, and structured output directly in the conversation

What Conduct Can Do

For a detailed explanation of how referencing and the hallucination check work, see How Referencing & Hallucination Checks Work.

Step-by-Step Guide

Getting Started

1. Open the Unique AI platform and navigate to a Unique AI Conduct space.

2. Describe your task. Be specific about what you want the agent to accomplish — not just what question to answer. The more context you provide, the better the agent can plan its approach.

   1. Good prompt: "Take the attached Q4 sales report (Excel), calculate year-over-year growth by region, identify the top 3 underperforming regions, and generate a summary chart plus a cleaned dataset."

      Less effective: "Analyze this file."

   2. Uploading Files: Attach files by clicking the attachment icon or dragging files into the chat. The agent can read your files, process them with code, and generate new files based on them.

3. Let the agent work. Conduct will show its progress through execution steps as it plans, writes code, and processes data. You can monitor the steps as they execute.

4. Review and iterate. Once the agent delivers results, you can ask for changes like we see below: "Clean up the graph, there are some overlapping labels." The agent maintains context and can build on its previous work.

When the agent creates files (charts, reports, processed data), they appear in the conversation:

• Images, Documents, and data files appear as in-line cards that can be downloaded.

Tips & Tricks

• Be task-oriented, not question-oriented. Instead of "What does this data show?", try "Analyze this dataset, find outliers, and create a summary report."

• Provide context upfront. Tell the agent what the data is, what format you want results in, and what matters most.

• Iterate naturally. The agent keeps your files and context within a session, so you can refine results across multiple messages.

• Check the execution steps. You can see what code the agent wrote and ran — useful for verifying the approach or learning from it.

• Use Skills when available. If your admin has configured skills for common workflows (report generation, data cleaning), the agent will use them automatically for higher-quality results.

Limitations

• Session timeout: Your sandbox shuts down after 10 minutes of inactivity. Downloading important files before stepping away is still the safest option. For every timeout you might encounter, see Conduct Timeouts and Limits.

• Storage is per chat: Your workspace is checkpointed automatically, so files the agent created earlier are restored when you continue the same chat — including after an idle shutdown. A new chat always starts from an empty workspace.

• Egress restrictions: The agent cannot browse the open web or make arbitrary external API calls. Web search and knowledge base access are available within your organization's policies.

• File type limits: Uploads limited to supported file types, max 10 files at once, up to 512 MB per file.

• Harness-dependent models: The available AI models depend on which harness your admin has configured. Currently, Claude models (Claude Code harness) and OpenAI GPT models (Codex harness) are supported.

• Connected tools (MCP) — partial hallucination check: Information the agent retrieves through connected tools (MCP integrations) is referenced — answers show clickable source links with readable titles — and it does reach the hallucination check, but with lower fidelity than knowledge base, web, or file sources. Confirm those results yourself.

• Page-level references depend on the file format: References to an opened file point at the exact page for PDF and Word documents. PowerPoint, Excel, HTML, and other formats are referenced at document level.

• Generated files are not covered: Referencing and the hallucination check apply only to the references added to the chat message the agent writes back to you. Facts and claims inside files the agent generates (reports, spreadsheets, charts, and other deliverables) are not citation-linked or hallucination-checked.