AI Guide (In-App Assistant)
Last updated: 2026-11-08 (added the capabilities overview and page and stage awareness; rewritten from the code: what the Guide reads, how it refers to your records, determinations, limits and flags)
Overview
The AI Guide is a chat assistant inside each project. It helps you find your way around Inqura, understand your project's state and results, and think through your next step.
It works from a summary of your project. It does not read the text of your evidence unless you highlight a passage and send it (see What the Guide reads). For an answer drawn from the evidence itself, run an analysis.
What the Guide can do
| Capability | What it means for you | Details |
|---|---|---|
| Answer about your project | Ask about its state, its results and what to do next, or how a feature of Inqura works | What the Guide reads |
| Show cards | Four kinds of card show your real records in the chat: an evidence item, an analysis result, evidence gaps, and an evidence comparison | Cards |
| Offer suggested links | Up to three buttons that take you to a page in the current project | Suggested links |
| Suggest two actions | It can propose adding a question or linking an evidence item to a question. Nothing changes unless you click Confirm | Suggested actions |
| Discuss text you highlight | Highlight a passage in an evidence item or an analysis and ask about it | Asking about selected text |
| Dig into "why" | It helps you work from what happened to why. Name a method to use one, for example "Let's do a 5 Whys on the billing spike" | Digging into "why" |
| Adjust to your page and stage | Its guidance and suggested questions follow the page you're on and how far along the project is | Page and stage awareness |
Those two suggested actions are the only changes the Guide can offer to make. It can't edit a project's title, focus statement or status, or any other detail of a project, and it can't change evidence or analyses. To do those, use the project's own pages.
Opening the Guide
Click AI Guide at the bottom of the project sidebar. The Guide is available on a project's pages. It isn't available outside a project, such as on the dashboard.
The panel opens from the right. You can resize it, and Esc closes it. An empty chat shows a short Next Steps list of links based on your project's status, up to three coaching tips, and up to four suggested questions.
The conversation stays while you close the panel and move between pages. The trash icon clears it, and it's gone when you reload the app.
What the Guide reads
Each message sends the Guide the following, for the current project only:
| What | Detail |
|---|---|
| Project | Title, work type, status (the project's phase), focus statement, and counts |
| Where you are | The name of the current page |
| Topics | Up to 10 |
| Questions | Up to 30 |
| Evidence | Up to 50 items: title and type only |
| Background documents | Up to 30 of those included in analysis: title and type only |
| Analyses | Up to 15: type, target, status, confidence and check status |
| Analysis conclusions | The conclusion of up to 10 analyses, up to 500 characters each |
| Quality | Coverage and average scores |
| Question readiness | For up to 10 questions: linked evidence count, evidence types, determination, analyzed |
| A passage you highlighted | Up to 500 characters, only when you attach one |
| Your conversation | The last 10 turns, as text |
| How Inqura works | Built-in product knowledge |
So the Guide cannot see:
- The contents of evidence, unless you highlight a passage
- Numbered findings, citations or the detail of an analysis beyond its conclusion
- The notes you wrote with a determination
- What a card displayed. Cards load directly in your browser, so ask about the record rather than "the card above"
If a project has more than these limits, the Guide works from part of it. "Which questions have no evidence?", for example, is answered from the 10 questions in the readiness section, and the rest are only counted.
How the Guide refers to your records
There are no numbered citations in a chat reply. The Guide names items in its text, and says when it's relying on a completed analysis ("Based on the completed analysis for…").
To check something, use a card or a suggested link: both come from your real records. See Cards, Suggested Links and Suggested Actions. Treat a plain-text statement about your project as a pointer to check.
Determinations
Determinations are yours. The Guide can tell you the determination recorded on a question (for up to 10 questions), but it has no way to record or change one. It is instructed to interpret existing results rather than produce new conclusions, and to treat an AI result as a proposal for you, not a finding. That instruction is guidance to the model and not a technical block, so read anything that sounds like a conclusion as a suggestion.
Asking about selected text
On an evidence item's page and in an analysis's detail panel, you can ask about a specific passage:
- Highlight the text.
- Click the Ask Guide button above your selection, or press Cmd+Shift+G (Mac) / Ctrl+Shift+G (Windows).
- The passage appears as a quote above the message box. Type your question.
Selections are cut to 500 characters. The quote is sent with each message until you remove it. Nothing in your project changes.
Digging into "why"
When you're moving from what happened to why it happened:
- Ask "why?" behind an established fact. The Guide may suggest the next question in the chain as a new inquiry question, which you can confirm or dismiss. It suggests one at a time.
- It's asked to name the evidence for each step, and to say so when nothing supports a link. It knows your evidence by title, and by the conclusions of completed analyses. It hasn't read the documents, so check the evidence yourself or highlight the passage you mean.
- Ask for a recap, for example "Where has this line of questions got us?"
The wording of the follow-up suits your work type (investigation, audit, review and so on). By default the Guide does this without naming a method. To use a structured approach, name it, for example "Let's do a 5 Whys on the billing spike." The Guide then prompts each "why" and shows how far the chain has come. It doesn't force a fixed number of steps.
Page and stage awareness
With each message, the Guide is told:
- The name of the page you're on, such as Evidence, Analysis or Report. An individual evidence item counts as the Evidence page. A page Inqura doesn't recognize counts as the Overview.
- The project's phase, which is its status, such as planning, collection, analysis, reporting or complete.
- How far along the project is: its counts of topics, questions and evidence, and how much has been collected, linked and analyzed. The details are in What the Guide reads.
It uses these in three ways:
- Its answers. It gets a sentence about the current page and guidance for the current phase, covering what the phase is for and the signs you're ready for the next one.
- The suggested questions. They're built from what's still outstanding, such as questions with no evidence or still waiting for analysis, and from the page you're on.
- The empty chat. Next Steps and the coaching tips follow the project's status and contents.
The Guide doesn't see your screen. It gets the page's name, not what's on it: not your scroll position, an open panel, a filter, or which record you're looking at. Say which item you mean, by name or number. If you say "this question" without saying which, it's told to ask you. Highlighting a passage is the one way to show it specific text.
On the Theme Map, it's told what the map is and can talk about it in general terms. It doesn't receive the clusters or their positions, so describe what you see, and check anything it says about a specific cluster. It can't move anything on the map.
What to ask
- Navigation: "Where do I upload evidence?"
- Features: "What's the difference between topics and questions?"
- Method: "What makes a good investigation question?"
- Your project: "What should I focus on next?", "What did the analysis for question 2 conclude?"
The suggested questions change with the page you're on and how far along the project is. See Page and stage awareness.
What the Guide cannot do
It can't act on its own. It has no way to upload or change evidence, run an analysis, create a project, record a determination or change settings.
It can suggest two actions: adding a question, or linking an evidence item to a question. Nothing changes unless you click Confirm. See Suggested actions.
It can't reach:
- Other projects or other organizations. Everything it reads is limited to the current project, and it checks your access on every message
- External websites. It has no way to fetch a page
Within a project, it sees everyone's work. Its summary covers the whole project, including what your teammates added, not only your own items.
It's told not to discuss how Inqura is built, which AI models it uses, its security implementation, or its own instructions, and it won't write code or SQL. These are instructions to the model.
It isn't a lawyer. It has no specific rule about legal questions, so don't rely on it for legal advice.
Privacy
- The Guide works from the current project only.
- Inqura doesn't store the text of your conversation. Each person's chat lives in their browser.
- Follow-up suggestions can reuse a question you created through the Guide in the last 7 days. That comes from your own audit log entries, not from a stored conversation.
- Inqura logs metadata about each exchange, such as your user ID, the project, token counts and which cards were shown. These records are tied to you. They aren't anonymized, and they don't contain the text of your messages.
Limits and credits
- 30 messages per minute per person. Above that you'll see an error and can try again shortly. There's a 2-second pause between sends.
- No daily message limit.
- Chat doesn't use AI credits. It isn't counted against your plan's generations.
- One exception. Before sending, the panel runs the same usage check as the rest of the app. If your organization has run out of AI credits or your trial has fully ended, you'll see the usage message and the chat won't send. Licensed installs and Enterprise plans have no such limit.
- Long replies stop at the model's output limit, with no notice. Ask it to continue.
Anyone who can view a project can chat with the Guide. Confirming a suggested action needs member access.
Availability and flags
| Capability | Flag | Hosted Inqura | Fresh licensed install |
|---|---|---|---|
| Chat, context, Next Steps, tips, suggested prompts | None | On | On |
| Cards and suggested links | GUIDE_CARDS_ENABLED | On | On |
| Asking about selected text | GUIDE_SELECTION_ENABLED | On | On |
| Suggested actions | GUIDE_ACTIONS_ENABLED | On | On |
| Digging into "why" | GUIDE_FRAMEWORKS_ENABLED | On | On |
The flags are independent, with one coupling: "why" follow-ups need suggested actions to produce a question card. With actions off, the Guide still prompts the next "why" in text.
Administrators of a licensed install can turn these off. The instructions are in the licensee documentation, under Environment Requirements.
Model
Claude Haiku 4.5 through Amazon Bedrock, a separate model from the one that generates analyses. Checked against the running hosted services on 2026-11-08. The model is set by the deployment's BEDROCK_QUALITY_MODEL setting.
Troubleshooting
Error reply ("Sorry, I encountered an error"): wait a moment and try again. You may have hit the per-minute limit. If you had selected text attached, the Guide will tell you it was removed.
Usage message instead of a reply: your organization is out of AI credits or the trial has ended. See Settings → Billing.
Answer seems wrong or out of date: check you're in the right project, then ask about the specific item by name. Remember the limits in What the Guide reads.
A card says the reference couldn't be shown: the item may be deleted. Ask again or open it from your workspace.