> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trubuild.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI chat

> A context-aware assistant for your procurement team: ask anything about your documents, evaluations, and tender queries, right where you're working.

Every TruBuild package comes with an AI chat that has read access to what you've uploaded and evaluated there. Click **Chat with AI** in the toolbar, ask a question in plain English, and it answers grounded in your documents, showing you each step it took along the way.

**This feature is self-contained.** The more you've uploaded, the more the chat can answer, but even a single uploaded document is enough to start asking questions. You don't need to have run any other feature first.

<Note>
  Chat is enabled per organization. If the **Chat with AI** button shows a lock icon, hover it and you'll see: "You don't have access to this feature. Contact your TruBuild administrator to request access."
</Note>

## Why this matters

In a traditional procurement workflow, "answer a question about the tender" means:

<Steps>
  <Step title="Open several files">
    The reference BOQ Excel, a vendor bid, a Word doc with technical notes.
  </Step>

  <Step title="Search through them">
    `Ctrl+F` through each document to find the relevant section.
  </Step>

  <Step title="Cross-reference">
    Check evaluator comments, the email thread with the vendor, the technical justification.
  </Step>

  <Step title="Compose an answer">
    Summarize back to whoever asked, usually in an email.
  </Step>
</Steps>

Even for a senior procurement lead who knows their tender inside out, a single "what's Vendor C's position on HVAC commissioning support?" is easily a 10-minute answer. With TruBuild you type the question and get a grounded answer in seconds.

## Where you can chat

The chat opens as a sidebar and follows you through the evaluation workflow. Each context keeps its own conversation history, and the assistant tailors its answers to where you are:

| Where                                    | What it helps with                                                             |
| ---------------------------------------- | ------------------------------------------------------------------------------ |
| Package overview                         | General questions about the package and its documents                          |
| Technical evaluation (setup and review)  | Clarifying criteria during setup; explaining scores and evidence during review |
| Commercial evaluation (setup and review) | Clarifying setup choices; explaining pricing comparisons and results           |
| Award                                    | Questions while you prepare the award recommendation                           |
| Knowledge Hub                            | Questions about tender queries and vetted answers                              |

There is also [Agent Sam](/platform/brain), a full-page, organization-wide chat inside Brain for admins. Use the sidebar chat for questions inside one package, and Agent Sam to reason across past projects.

## What the chat can do

Beyond plain Q\&A, the chat can take on real work:

| Ability                             | What it does                                                                                        |
| ----------------------------------- | --------------------------------------------------------------------------------------------------- |
| **Answer from your documents**      | Reads and searches the package's uploaded documents to answer in plain English                      |
| **Answer about evaluation results** | In the technical and commercial review stages, explains scores, comparisons, and evidence           |
| **Check the Knowledge Hub**         | Looks up vetted tender query answers so your clarification language stays consistent                |
| **Search the web**                  | Looks up market data, standards, and regulations when external context helps                        |
| **Generate documents**              | Produces downloadable files (Word, PDF, Excel) from your data, such as a summary or comparison pack |
| **Ask you clarifying questions**    | Presents a short form when it needs direction before proceeding                                     |

Generated documents appear in the conversation with their filename; click to download. They are a strong first draft. Always review them before issuing anything externally.

## Watching it work

As it works, the assistant shows each step in plain language, such as "Checking your knowledge hub", "Searching the web", or "Created your document (summary.docx)", so you can see where the answer comes from. For longer reasoning you may see a collapsible **Thinking…** section before the final response streams in.

To double-check a claim before it goes into a decision, open the referenced document yourself and confirm it against the source.

## When the chat asks you questions

Sometimes the assistant needs direction before it can do the work well: which vendors to include, what format you want, which round to use. Instead of guessing, it presents a short form in the conversation.

* Questions arrive one page at a time, with a progress bar showing "Page X / Total".
* Required questions are marked with `*`; optional ones show "(optional)".
* Some options carry a **Recommended** badge where the assistant has a suggested default, and an **Other** field when a free-text answer is allowed.
* Use **Back** and **Next** to move between questions, then **Submit** on the last page.
* You can attach files via **Attach documents (optional)** if your answer needs supporting material.
* In a hurry? Click **Use your best judgment** to skip the form and let the assistant proceed with sensible defaults.

After you submit, the form is replaced with an "Answers submitted" summary of what you told it, so the record stays in the conversation.

## Example questions

Example questions to try (each category works standalone; you only need to have uploaded the relevant document type):

<AccordionGroup>
  <Accordion title="About a commercial evaluation (if you've uploaded priced BOQs)">
    * "What's the spread on the top 10 items by dollar impact?"
    * "Which vendor priced this package cheapest?"
    * "Summarize all lines where Vendor C is more than 30% above the average."
    * "What's the total delta between Round 1 and Round 2 for Vendor B?"
  </Accordion>

  <Accordion title="About technical proposals (if you've uploaded vendor tech docs)">
    * "Compare Vendor A and Vendor C's project management approach in 5 bullets."
    * "Which vendor has the strongest HSE track record on similar projects?"
    * "What does Vendor D say about commissioning their control system?"
  </Accordion>

  <Accordion title="About tender queries (if you've logged clarifications)">
    * "Are there any open queries that could delay award?"
    * "Summarize the clarifications we've received from Vendor B so far."
    * "Have we already answered a question about retention on provisional sums?"
  </Accordion>

  <Accordion title="About any document you've uploaded">
    * "What are the three biggest risks in this RFP?"
    * "Summarize this document in a paragraph."
    * "Find all mentions of retention across all uploaded documents."
  </Accordion>
</AccordionGroup>

## Scopes

The sidebar chat is always scoped to where you opened it, so it can't accidentally answer questions about a different tender. Within a package, your visibility follows your [package access](/get-started/roles-permissions) (a package role takes precedence over your project role), and the chat only draws on content within that scope. Organization-wide questions belong in [Agent Sam](/platform/brain), which is limited to owners and admins.

## Step-by-step

<Steps>
  <Step title="Upload something">
    Any supported document in the package: an RFP, a BOQ, a contractor submission, or a query register.
  </Step>

  <Step title="Open chat">
    Click **Chat with AI** in the toolbar. The sidebar opens scoped to your current context, such as a package overview or a commercial review.
  </Step>

  <Step title="Ask a question">
    Plain English. Be specific about which document or vendor when it matters.
  </Step>

  <Step title="Follow up">
    Chat is stateful within a conversation: "and what about Vendor B?" knows what you're asking about.
  </Step>

  <Step title="Verify against the source">
    For anything going into a decision, open the documents the assistant read and confirm the answer against them.
  </Step>
</Steps>

## Managing conversations

* **Resize:** drag the handle on the left edge of the sidebar to widen or narrow it.
* **Minimize or expand:** use the chevron icons; click **X** to close the pane.
* **History:** past conversations are listed with titles, previews, and timestamps. Click one to pick it up where you left off.
* **New conversation:** click **+** to start a fresh thread. Start a new thread when you change topic; it keeps each conversation's context clean.
* **Feedback:** rate any answer with the thumbs up or thumbs down icons. Ratings help your TruBuild team improve answer quality.

## Best practices

<AccordionGroup>
  <Accordion title="Use chat for exploratory questions, not final numbers">
    The chat is excellent for finding things and summarizing. For final scores, prices, and award positions, verify against the structured commercial, technical, query, or award screens, or against the source document itself.
  </Accordion>

  <Accordion title="Be specific about the scope">
    In a package with many documents, start with "In the RFP, ..." or "Looking at Vendor B's submission, ...". It cuts answer time and reduces wrong-scope answers.
  </Accordion>

  <Accordion title="Answer the clarifying questions when you can">
    "Use your best judgment" is fine for exploratory work. For anything you'll circulate, answering the form questions (round, vendors, format) gets a noticeably better first draft.
  </Accordion>

  <Accordion title="Use it to prep for meetings">
    Five minutes in the chat before a bid review meeting produces a much sharper read-out than 30 minutes of file scrolling.
  </Accordion>
</AccordionGroup>

## Related

<CardGroup cols={2}>
  <Card title="Brain and Agent Sam" icon="brain" href="/platform/brain">
    The organization-wide assistant that reasons across past projects.
  </Card>

  <Card title="Knowledge Hub" icon="book" href="/platform/knowledge-hub">
    The vetted tender query answers the chat can check for you.
  </Card>
</CardGroup>