Most manufacturing teams have access to powerful AI tools, but those tools don’t know anything about their specific process. They’re general-purpose, and while useful for drafting text or answering broad questions, are disconnected from the actual system you’re trying to improve.
The AI Assistant in Threaded is different. It works inside your organization, with your Work Instructions library, your Map, your parts and tools, and your issues — and it can search all of them while you talk to it. That turns AI into a collaborator that understands what you build and how you build it, not a chatbot you have to brief from scratch every time.
#Who can use it
The AI Assistant is available to members of organizations that have it enabled. It is not part of Operator Mode — operators working on the floor do not see the chat sidebar.
#Agentic Capabilities
The assistant can give you advice and analysis, but it doesn’t stop there — it can also act. It updates work instructions, creates and assigns issues, adds parts and tools, and makes other changes to your process directly, rather than telling you where to click.
Those changes run through Threaded MCP, the connection that lets the assistant read and write your organization’s data as you, with your permissions. It connects on its own for most organizations, so there is usually nothing to set up. If it isn’t connected when the assistant needs it, you’ll get a Connect card in the conversation, and you can check or manage the connection at any time from Admin → MCP Connections. See Connecting Threaded MCP.
When the assistant proposes a change, it shows you what it intends to do and waits. Every action that modifies your data appears as a checkbox row you can review — uncheck anything you don’t want, click the pencil to Edit suggested values before it runs, then click Execute. Deletions are called out in red with an explicit warning. Read-only lookups, like searching your library, run on their own without asking.
This human-in-the-loop approach means you can hand off repetitive or time-consuming work — updating a step across several procedures, or turning an analysis into a set of assigned issues — while keeping full visibility over what actually changes.
#Auto-accept tools (bolt)
By default, every change waits for you to click Execute. For longer sequences — multi-step report authoring, bulk updates — you can turn on auto-accept so approved actions run without a click each time.
- Open the AI Assistant sidebar.
- Click the bolt icon at the top of the sidebar, to the right of the Threaded MCP status icon.
- When it is on the bolt turns green and its tooltip reads Auto-accept tools is on (click to turn off). Click again to turn it off.
Auto-accept is remembered per conversation. Questions the assistant asks you still need an answer before related work continues — those are never auto-accepted.
Use auto-accept when you trust the current thread and want speed; leave it off when you want to review every change before it runs.
#Accessing the AI Assistant
Click the AI Assistant icon in the top navigation bar to open the chat sidebar, available from anywhere in Threaded. Close it with the arrow at the top, or click View as full page to open the conversation full-width — the full page adds its own list of conversations with a Search chats field.
To start a new conversation, click the + at the top of the sidebar. The history icon next to it lists your earlier conversations.
#How Context Works
The quality of the AI’s responses depends directly on the context it has. A general-purpose tool works from whatever you type in the moment, which means you spend as much time explaining your process as improving it. Threaded supplies that context up front and gives you ways to sharpen it.
#Automatic Context
Every conversation starts with a high-level picture of your organization: your members and their roles, how many work instructions, procedures, and steps you have, your parts and tools counts, issue and media statistics, and the top-level process nodes from your published Map. From there the assistant searches — it looks up specific work instructions, nodes, parts, and issues on demand rather than carrying your whole library in every message.
The practical consequence is worth knowing: the assistant finds what you have, so the more of your process lives in Threaded, the more specific its answers get.
#Page Context
The assistant also knows where you are. Each message carries the page you’re on, what can be done there, and the relevant help documentation, which is why “What can I do on this page?” gets a real answer, and why asking about a report while you’re looking at it works without explanation.
#Adding Context with @mentions
Type @ in the chat to pin specific objects into the conversation. The assistant has direct access to anything you @mention and can analyze and reason about it specifically. You can @mention:
- Work Instructions and Map process nodes
- Process Planning groups
- Parts and tools
- Issues
- Media files
- Manufacturing Reports and report data sources
- Previous AI conversations
For a full breakdown of what can be @mentioned and where, see Search and Mentions.
#Attaching Files
Click the paperclip below the chat window to attach external files — documents, spreadsheets, images, and PDFs. This is how you bring in reference material, supplier data, or quality reports that live outside Threaded. You can also paste a screenshot straight into the message box.
Attached files are saved to your organization’s media library automatically, so they are available later without re-uploading. They also stay attached to the conversation and are re-sent with each message until you remove them with the × on the file’s chip.
#Conversation History
By default the assistant includes summaries of your previous conversations, so it can build on earlier analysis. To work without them, click the history icon to the left of the model selector — its tooltip reads Previous conversations included in context (click to disable).
Only summaries of conversations already saved are used, limited to your ten most recent in this organization. On a new organization there may be nothing there yet.
#Guided Workflows
Beyond freeform chat, the assistant can run structured workflows that walk you through a multi-step task. Ask for one by name:
- Workflow: Gather Organization Information — collect and save your Organization Info (industry, locations, priorities, challenges) so future conversations have better grounding. You can also start this from Admin → Organization Info.
- Workflow: Review Work Instructions — a conversational review of completeness and quality across one or more work instructions. This is collaborative guidance, not an automated Process CI run; for scored checks before publish, use the Work Instruction Review tab.
- Workflow: Author a Report — create or update Manufacturing Reports, planning document and dashboard together. You can also start it with Build with AI or Update with AI on a report.
#Forms and Charts in the Conversation
Some actions open a small form right in the conversation instead of asking you to fill in fields by typing — creating an object, for example. Fill it in, submit, and the assistant continues from there.
The assistant can also draw charts inline when a number is easier to see than to read, which is useful when you are comparing cycle times across stations or looking at issue counts over time.
#Model Selection
Threaded gives you access to multiple frontier AI models in the same platform. Different models handle different tasks differently, so if one isn’t giving you what you need, switching is a quick way to get another perspective. Use the model selector below the chat window; your choice is remembered for next time.
Every response carries a reminder worth taking seriously: AI models can make mistakes, so verify important information before acting on it.
#Tips for Getting the Most Out of the AI
- Add context before asking — @mention the specific procedures, nodes, or parts you want analyzed before asking your question. The assistant performs significantly better with focused context than with an org-level summary alone.
- Be specific about your goal — “Analyze my Powder Coat operation for bottlenecks” gets you further than “analyze my process.”
- Use it as a collaborator, not a search box — ask follow-up questions, push back on recommendations, ask it to explain its reasoning. The best results come from a back-and-forth.
- Let it act — if you agree with a proposed change, run it with Execute rather than making the edit by hand. It’s faster and keeps the work in context.
- Use the bolt for trusted multi-step work — turn on auto-accept when you want a long sequence to run uninterrupted; turn it off when you want to see each action.
- Start a new conversation when shifting focus — the assistant works best when a conversation stays on one goal. New problem, new chat.
#What the AI Can Help With
- Work Instructions — draft new procedures, improve existing steps, review clarity and completeness in conversation, and make bulk updates across your library. For automated scoring before publish, use Process CI on the Work Instruction Review tab.
- Process Analysis — identify constraints and bottlenecks, analyze line balance, compare Process Plan against actuals, and estimate the impact of proposed changes.
- Continuous Improvement — surface improvement opportunities, prioritize issues, and connect floor feedback to process changes.
- Parts and Tools — analyze usage, identify supply chain dependencies, and populate or clean up your parts and tools tables.
- Scenario and Project Planning — ask how to double production in a given timeframe, or how to launch a new product. The assistant will outline a plan and create the issues needed to execute it.
- Manufacturing Reports — plan a report, query its bound data sources, and build the dashboard. See Manufacturing Reports and the report prompts in AI Assistant Use Cases.
- General Manufacturing Knowledge — ask about lean principles, industry standards, technical writing, or any manufacturing concept and get answers framed by your specific process.
For detailed examples and prompts organized by use case, see AI Assistant Use Cases.