User guide

Getting started

You work inside your organisation’s Quaestra.ai workspace — usually a branded hostname provided by your administrator.

1. Open the workspace URL your organisation shared with you.
2. Register or sign in (see below).
3. Complete onboarding prompts if shown — persona and introduction routes help tailor first-run guidance.
4. Land in Ask, the default research screen.

Sign in & legal

Access is passwordless and organisation-controlled.

Your organisation may limit sign-in to approved email addresses. If access is restricted and your address is not approved, ask an administrator to add you.

Email code

1. Enter your email address on the login screen.
2. Open the one-time code sent to your inbox and enter it to continue.
3. Accept Terms and Conditions when registering by email, if prompted.

Google or Microsoft sign-in

Where your organisation has enabled them, choose Google or Microsoft on the login screen and complete the provider flow. The same approved-address rules apply.

Cookies, Terms, and Privacy

• Respond to the cookie consent prompt when required.
• Re-open Terms and Privacy from Profile whenever you need them.
• Sign out from your account menu when you finish on a shared device.

Help & onboarding

Ask

Ask is where you put natural-language questions to the curated corpus. Answers stream live and stay grounded in the library.

Ask a question and verify sources

1. Go to Ask.
2. Type your question. Paste a formatted equation or an image of a formula — equations appear as rendered blocks in the Ask box (click a block to open an editor with a live preview). You can keep typing prose around them. Attach one or more supporting files with the paperclip or by dragging them onto the Ask box if chat attachments are enabled — each file appears as its own chip (remove any you do not need). Attached files stay as background context and are not shown as math in the chat.
3. Submit and watch the answer stream in.
4. Open cited sources in the integrated viewer when your organisation shows them. Zoom or download a PDF when you need the original.
5. Continue with follow-up questions in the same conversation, or start a new conversation when the thread should stand alone.

Modes, filters, and follow-ups

• Peer Reviewed Only — limit grounding to documents marked peer-reviewed, when available.
• Go Deep — when your organisation enables it, use deeper research mode (the label is administrator-defined).
• Quest — when your organisation enables it, turn this on in the same row as Go Deep. Quest writes a sourced report from several searches, then starts a normal conversation you can follow up in Ask. It cannot be combined with Go Deep, the web, or attachments. It respects Peer-Reviewed Only and your organisation’s guiding principles (History shows which route shaped the report when that indicator is available). Questions about the most frequent topics or themes across the library are answered from the summary of every document in scope rather than a handful of passages, and note how emphasis shifts across years. When a search finishes without everything it needed, the answer notes what is still sought. History marks these answers as Quest.
• Document / author selection — narrow the corpus when your organisation exposes those controls.
• Suggested questions — pick from the Question Bank on the welcome surface when shown.
• Follow-up chips — continue from generated next questions when enabled for your audience.
• Explain selection — highlight part of an answer (including equations), choose Explain this, read the explanation in a popup, then continue in Ask as a new conversation that keeps the original turns and adds the explanation as the next turn. When you highlight a formula, the underlying math text is used so the explanation stays faithful to the equation.
• Guiding Principle indicator — see which expert lens shaped the answer.
• Context indicator — when shown, watch how much prior conversation is still in play.
• Web-aware answers — on plans that include web search, and when the model allows it, answers may draw on the open web in addition to your corpus.
• Math verification — when enabled for your audience, open Verify calculations to recompute steps with live inputs, Export to Excel for a live workbook, and — when your organisation also enables formal proof export — Export Lean proof for a machine-checkable formal artefact you verify independently. When the answer names a known formal result, a Results used section lists the statement and the hypotheses you should confirm.
• Numbered equations — when you ask about a labelled formula such as Eq. (2.19), answers prefer that equation and can cite it by number and page.
• Generated files — download any files returned with an answer.

Rate answers and retry

1. Mark an answer helpful or unhelpful.
2. Optionally choose a structured reason and leave a short comment.
3. Change or clear a previous rating if your view changes.
4. After an unhelpful rating, choose to retry with the organisation’s Powerful Model when that option is configured.

Infographic generation

When administrators enable infographics for your audience, generate a downloadable image from an answer, watch status until it is ready, then download it for presentations or notes.

History

History keeps your research threads organised and exportable.

• Search across questions and responses.
• Filter by thumbs-up or thumbs-down ratings.
• Browse with infinite scrolling; conversations group into threads.
• Open a full multi-turn conversation, rename it, or continue it in Ask.
• Delete a single question, a whole thread, or all history (typed confirmation required for delete-all).
• Rate answers from History as well as from Ask.
• Review metadata such as model, routing, source, and web-search indicators when shown. Organisation admins and testers can copy a query ID from each answer to match it with Telemetry and the MCP event log.

Share & export

My Documents & Virtual Review

Available when your plan includes personal documents (admins and testers always have access).

1. Open My Documents from navigation.
2. Upload private documents and wait until processing shows ready. If you add a file you already have in My Documents, it is skipped and you see a message naming the existing file, with a link back to My Documents.
3. Use ready personal documents alongside the shared corpus when asking questions.
4. Run Virtual Review against the organisation’s shared expert lens.
5. Export the review as PDF or Word.
6. Delete your own documents when you no longer need them.

Q&A digests

Opt into daily and/or weekly email summaries of questions you asked in the web app or on mobile, with short answer previews and links back to the conversation. Questions asked through Integrations or Alexa are not included. Equations in those questions and answers appear as formulas in the email. If you have been inactive, a re-engagement note may include suggested questions from the Question Bank.

• Manage preferences in Profile.
• Send yourself a test digest from Profile.
• Unsubscribe from any digest email with a one-click confirmation flow.

Profile & subscriptions

Account

• See your email and member-since date.
• Toggle daily and weekly digests.
• Delete your account with typed DELETE confirmation. Your user record is removed or anonymised; existing Q&A history may be retained without your email association.

Subscription

• View current plan, daily quota (or unlimited), queries used today, and reset timing.
• Compare plans and self-serve checkout, billing portal, cancel, and resume where enabled.
• For pay-as-you-go plans, review usage and estimated cost when shown.

Integrations (MCP Access)

When your plan includes MCP Access, open Profile → Integrations for the server address, copyable client configuration, and a browser-based authorisation flow so approved external tools can use your research workspace without you managing a long-lived secret by hand.

If personal documents are also available on your account, connected tools can list your ready My Documents and ask questions that appraise those files against the organisation’s shared Library — the same dual-scope research you get in Ask. Upload new PDFs under My Documents in the web app first; the integration lists and asks about them but does not upload.

When your organisation allows Quest for your role, connected tools can request Quest the same way you can in Ask.

When math verification and formal proof export are enabled, connected tools can export formal-proof source for an answer you own and look up curated formal statements with their hypotheses.

Amazon Alexa

When your organisation allows Alexa for your role (Admin → Alexa: Admin & Tester or All Users) and — for regular users — your plan includes Alexa Access, you use the shared Quaestra.ai Alexa skill — your organisation does not create its own skill. The Assistant Name from Admin → Configure App (the same name shown in chat) brands spoken replies and lets you address your organisation inside the skill. Admins and testers can link when organisation visibility includes them, without needing Alexa Access on a plan.

1. Install or open the Amazon Alexa app on your phone (same Amazon account as your Echo).
2. Enable the shared Quaestra.ai skill in the Alexa app (your organisation does not publish its own skill).
3. In Quaestra.ai, open Profile → Integrations and choose Generate pairing code.
4. In the Alexa app, open the Quaestra.ai skill and start Link Account. Type the pairing code on the Quaestra.ai screen on your phone — you do not speak the code to the Echo. The code expires in about ten minutes and can be used once.
5. Ask by voice, for example: “Alexa, ask Quaestra what is momentum theory”, or open the skill and say “ask Ricardo what is momentum theory” using your Assistant Name.
6. Say new chat to start a fresh conversation, or keep asking follow-ups to continue the current thread.
7. To use another organisation, generate a pairing code while signed into that account, then say “link organisation” and the code inside the skill. Say “list organisations” to hear your linked names.
8. Open History on the web to review, rename, continue, or export those conversations — they appear alongside your other Ask activity.

Alexa reads a short spoken summary prepared for voice when the answer is ready quickly. Longer research may take a moment: Alexa says it is researching and asks you to say continue in a few seconds — then it reads the answer. The full written answer also appears in History (and on the Alexa card when spoken). Detailed math and source lists are not read aloud as a long citation list.

To disconnect this organisation, use Unlink Alexa under Profile → Integrations, or unlink the skill in the Alexa app.

Product feedback

1. Open the feedback control (label is set by your organisation).
2. Write your message; optionally allow the organisation to contact you.
3. Attach images if helpful, then submit without leaving the workspace.

Mobile & tablet

The research workspace is responsive. Navigation adapts on smaller screens, and Ask and History are touch-friendly so you can continue research away from a desk.

Your role

You are responsible for the organisation’s research posture — what people can ask against, how answers are reasoned, and how the workspace looks and commercialises.

• Curate the shared Library — uploads, connected cloud drives, inbound email, and tags.
• Shape Ask behaviour, onboarding, and public landing pages.
• Manage users, roles (admin, user, tester), personas, model overrides, and who may sign in.
• Configure models, Guiding Principles, plans, digests, MCP, and Embed Chat.
• Review analytics, feedback, and operational snapshots.

View as User

Switch into an effective user role to preview end-user navigation, feature gating, and Ask behaviour. Your real admin role is retained so you can switch back without signing out.

Change log

Admins and testers can review dated product updates that affect what people can do in the workspace.

1. Open the Change log control in the top navigation (list icon next to Help on desktop; under the menu on smaller screens).
2. Browse entries newest first. Each item is labelled Feature, Improvement, or Fix.
3. Close the panel when finished — you stay on the same page.

Regular users do not see this control. While you use View as User, it is also hidden so the preview matches an end-user session. Tester accounts see the same Change log without needing the Admin menu.

Upload & Library

Upload

1. Open Library, then choose Upload.
2. Drag and drop PDFs or select multiple files (batches of up to 50; typically up to 50 MB each).
3. Continue working elsewhere while transfer and indexing proceed in the background.
4. Watch accepted, interrupted, already-in-library, failed, and processing states until documents are ready.
5. If a PDF is already in the Library — even under a different name — it is skipped and you see a message naming the existing document, with a link to open it in the Library.

Library

• Browse paginated shared documents; search by title, author, filename, or tag; sort columns (including A–Z by title, plus pages, size, source, status, peer-reviewed, and upload date); filter by tags, by Source (Uploaded, Google Drive, and other connected providers when available), or by Status (Ready, Pending, Processing, Error); select all matches when bulk-acting.
• Open previews, download PDFs, and edit title, description, and authors.
• Mark or clear peer-reviewed status; add or remove tags on a single document, or select several documents and use Edit tags to add or remove tags across the selection in one step.
• For documents from a connected drive: re-sync a single file, open the connector screen, or remove the Quaestra copy — the file on the source drive is never changed or deleted.
• When a connected source file is gone from the drive, Library may show a missing marker and that document is kept out of answers until it is re-synced or removed from Quaestra.
• Delete individual documents or bulk-delete a selection.
• Watch the status of any document still being read: while it is in progress the status shows a percentage that updates on its own, so you can see a long paper advancing instead of waiting on a static label. Queued documents show as pending until reading starts, and the list settles to the final status by itself when the last one finishes.
• Open document detail and extracted-content views when diagnosing readiness.
• Reprocess a document (or a selection) to re-read and re-index it — useful after a paper with heavy notation was added, or when you want an existing document to benefit from later reading improvements. Reprocessing runs in the background and the document's current text keeps answering questions until the new version is ready, so nothing drops out of search while it runs. The automatic summary is regenerated; titles and authors you edited are kept.
• Open Equation index on a document to browse numbered formulas (for example Eq. (2.19)) with a label filter. Select a labelled equation to open the PDF on that page and highlight the equation number when the paper’s text layer includes it. Then review All extracted math for every captured formula grouped by page. If a formula-heavy paper shows little or nothing here, reprocess it and check again.

Cloud Connectors

Connect a cloud drive as a read-only source of PDFs. Quaestra keeps a working copy for search and viewing so research continues even when the drive is temporarily offline. Source drives are never modified.

1. Open Connectors from Admin (or via Library → Manage connectors).
2. Choose Google Drive (additional providers may appear as Coming soon).
3. Select Connect Google Drive and complete the provider sign-in for a read-only connection.
4. Browse folders on the left; open contents on the right. Tick a folder to include all PDFs (recursive), or tick individual files. A folder shows − when only some files are included. Selecting My Drive root asks for confirmation — it syncs every PDF under My Drive now and any folders or files added later (unless you exclude specific files).
5. Choose a sync schedule (manual only, every 15 minutes, hourly, every 6 hours, or daily) and save.
6. Select Sync now for an incremental update, or Full re-sync when you need a thorough refresh. Files you excluded after an earlier sync are removed from the Quaestra library on the next sync — Drive originals are never deleted.
7. In the folder list and under Skipped & duplicates, files whose content is already in the Library stay visible and are labelled as a duplicate of the existing document, with a link to open it. You can still choose Sync anyway for a duplicate. Cancel or resume a run if storage is low, and check run history.

Inbound Email

Give trusted people a private organisation address so they can send or forward PDFs into the shared library — the same corpus as Upload — without downloading attachments and re-uploading in the browser.

1. Open Admin → Inbound Email (separate from Configure App email settings used for sign-in codes and digests).
2. Enable the feature to receive a unique organisation address; enabling is attributed to the administrator who activates it.
3. Manage the trusted-sender list so only approved addresses can contribute documents.
4. Regenerate the address if it must be rotated.
5. Review ingest activity, then confirm accepted documents appear in Library and are ready for Ask.

Configure App

Central place for branding, Ask copy, legal text, and experience toggles.

Design & theme

Tune colours, fonts, and corner style so Ask, login, landing pages, embed chat, and emails match your brand — with version history you can restore.

1. Ask your platform administrator to enable Design for your organisation if the Admin menu shows it as disabled.
2. Open Admin → Design.
3. Adjust colours and fonts, or import a starting look from your website or from screenshots of your brand. Use Top bar background and Top bar text to style the navigation bar separately from the rest of the page.
4. Review the live preview, then Save draft or Publish with a version label when you are ready for everyone to see the change.
5. Use Versions to restore an earlier published look, or Reset to default to return to the original Quaestra.ai appearance (also saved as a version).

Retrieval

Choose how Ask selects evidence from your library, including optional improved semantic matching when the platform has allowed it for your organisation.

1. Open Admin → Configure App → Retrieval.
2. Pick a Quality preset: Automatic (recommended), Precise, or Broad — so answers lean toward balanced, tighter, or wider evidence from the library.
3. When improved matching is available, set Semantic matching mode (Off, Shadow to measure without serving, or Hybrid) and choose who can see it during rollout (Off, Admins & Testers, or All).
4. Choose who can use Quest (Off, Admins & Testers, or All). Models and run limits live on the Models page.
5. Check embeddings index status (totals and whether the index is ready). Use Backfill embeddings after large library changes so improved matching can cover new documents. After changing the Embedding Model under Models, use Wipe embeddings (confirm) then Backfill until the index is ready again.
6. When the embeddings index is available, open the Embeddings map to see related library passages clustered together. Try a word or phrase to highlight nearby passages and check whether matching lands in the right place.

My Documents

Organisation settings for personal documents and Virtual Review (separate from the end-user My Documents page).

1. Open Admin → My Documents.
2. Show or hide the My Documents tab in navigation for admins and testers (regular users do not see this tab from this setting).
3. Confirm Virtual Review uses the Default Model set under Models (same default as Ask).
4. Edit the Virtual Review prompts (section critique with lens, no-lens fallback, and final synthesis) or reset them to the built-in defaults.

Users

1. Open Users to list people with email, role, persona, assigned model, registration, last login, and query count.
2. From an approved connected tool signed in as an organisation admin, you can also look up who registered or used the workspace in a date range — based on registration and activity records, not on people named in the Library.
3. Optionally restrict who may register or sign in: in Sign-in allowlist, add approved addresses or whole domains (for example @organisation.com). Leave the list empty to allow anyone. When the list has entries, only those addresses can get in — include your own address before saving so you are not locked out.
4. Change roles among admin, user, and tester.
5. Override persona or assign a per-user model when someone needs a fixed setup.
6. Enable or disable digests per person; send a test digest.
7. Inspect or export an individual’s Q&A history; export membership as CSV, Word, or PDF.
8. Delete an account, or wipe all user/query data only with the strong confirmation the screen requires.

Question Bank

Create, edit, bulk-edit, and delete suggested questions. Assign level and audience, filter the bank, and import/export CSV. The bank feeds Ask welcome suggestions and digest re-engagement prompts.

Guiding Principles

Author the expert frameworks that shape how questions are reasoned.

• Set routing model, match-score threshold, combine-all-matching-routes, and default principle behaviour.
• Create, edit, reorder, and delete hypothesis routes with prompt prefixes and tag references.
• Save labeled Guiding Principle versions (settings, default principle, and hypothesis routes as one unit) from the versions rail; restore a prior version (optionally saving live Current first); rename or delete history; and compare two or three snapshots side by side — Current is optional.
• In the comparison view, edit the Current (live) text in place; a compact Save and Revert pair appears directly beneath any field you change, so each edit is saved on its own.
• Test a route against a sample question; preview the resolved prefix.
• Review hypothesis optimisation and routing analytics (7 / 30 / 90 days), utilisation, near-misses, and recent decisions.

Models

• Refresh provider catalogues; enable or disable user-facing models; test a model.
• Set the tenant default model (used for Ask and Virtual Review), Powerful Model (unhelpful-answer retry), Math Model, Voice / Alexa Model (spoken summary only; History keeps the full written answer), PDF Extract Model (Library document text and metadata), Embedding Model (dedicated index models for improved Ask matching — uses configured Gemini or OpenRouter keys; after a change, wipe then backfill under Retrieval), and Math Equation Pipeline (Legacy or High-fidelity formula reading for Library documents). When High-fidelity is on, choose the formula reader: the PDF Extract Model, or the platform specialist formula reader.
• In the Quest section, set the report model, the planner model, optional matching rank model, how many Quest runs can happen at once, the daily cap, search rounds, and the cost ceiling.
• Store or clear organisation-owned provider credentials (Bring Your Own Key); see whether a tenant or server credential is active.
• Enable web search per supporting model.
• Tune concurrency (in-flight calls, queue depth, wait) or reset to deployment defaults; review live pool statistics.

Model Benchmark

1. Enter one question and choose the shared documents to include (or select all).
2. Optionally apply Peer Reviewed Only.
3. Run against multiple models; compare route, duration, sources, and answers as results stream.
4. Stop a run, clear results, or export the side-by-side comparison to CSV.

PDF parsing

Choose the parse model, upload a test PDF or image, run extraction, and review output for chat attachments. Library document processing uses the PDF Extract Model under Models.

Math Pipeline

Dry-run high-fidelity formula reading on a single page before you switch Library ingest.

1. Open Math Pipeline from the Admin menu.
2. Confirm availability (server tools, plus either the PDF Extract Model or the specialist formula reader, depending on Math extractor under Models).
3. Enter a Library document ID and page, or upload a PDF, choose a validation scope, and run.
4. Results are grouped by printed equation. When a fidelity check passes, the formulas shown are the ones that were read — not a rebuilt version.
5. Compare pipeline formulas and fidelity notes with the legacy equation blocks for that page.
6. When satisfied, set Math Equation Pipeline to High-fidelity under Models, choose the formula reader, and reprocess documents that should benefit.

Math Corpus

Validate how well Library documents were read for equations — without changing what Ask uses for answers.

1. Open Math Corpus from the Admin menu (or follow Review in Math Corpus on a document’s equation index).
2. Enroll a ready Library document. Optionally import the current equation index as a baseline run — you do not need to reprocess first.
3. Build ground truth model-blind: open each PDF page, add every eligible displayed equation with page, order, label, and accepted formula text, then certify completed page ranges.
4. Freeze the truth set when enumeration is complete. Assign the document to a development or test split at document level.
5. Open an extraction run (imported baseline or a later reprocess), match predictions to truth one-to-one, and score transcription, label, and page separately.
6. Review certified metrics (precision, recall, faithful transcription, and paired run comparison) and export accepted formulas for later symbolic checks.

Analytics

The tenant dashboard covers cost and usage, downloads, chat attachments, infographics, sharing funnel and social graph, answer ratings (including downvote trends by model and Guiding Principle), MCP usage, and live presence. Open downvoted answers to review the full conversation.

Feedback Inbox

• Triage general feedback with submitter, time, message, contact permission, and image attachments (lightbox).
• Review answer thumbs-up/down with structured reasons; open the surrounding conversation.
• Delete items and export feedback or ratings to CSV.

Digests administration

• Master enable/disable; default daily/weekly opt-in for new users.
• Send hour/minute (UTC) and weekly send day.
• Cap conversations per digest; set inactivity window before re-engagement. Digests include questions asked in the web app or on mobile, not through Integrations or Alexa.
• Edit subject lines, intro, and sign-off content.
• Review sent, opens, clicks, and unsubscribes.

Alexa

1. Open Admin → Alexa.
2. Set Enable Alexa for this organisation to OFF, Admin & Tester, or All Users. When OFF, nobody can link Alexa or ask by voice for this organisation. Admin & Tester limits access to those roles; All Users opens it to everyone eligible on their plan.
3. Set the Assistant Name under Admin → Configure App so voice replies use your brand and users can address this organisation inside the skill.
4. Optionally set Voice / Alexa Model under Admin → Models so spoken replies use a dedicated rewrite; History always keeps the full written Ask answer.
5. Include Alexa Access on the plans that should offer pairing when visibility is All Users (Admin → Plans/Billing). Admins and testers can use Alexa when organisation visibility includes them — open Profile in the top navigation, then Integrations, to generate your own pairing code.

How users get access: organisation visibility includes their role + (for regular users) Alexa Access on their plan + they enable the shared Quaestra.ai skill in the Amazon Alexa app, then generate a pairing code under Profile → Integrations and type it on the Link Account screen on their phone (not spoken to the Echo). Your organisation does not create an Amazon skill.

Plans & billing

• Set free-user and tester daily query limits.
• Create flat monthly or pay-as-you-go plans with prices (USD / EUR / GBP), limits, display order, student marking, and customer-facing descriptions.
• Toggle plan features: Personal Documents, MCP Access, Alexa Access, and Web Search.
• Activate, deactivate, or delete plans; resynchronise with the commercial provider; review active subscriptions and revenue breakdown.
• Reset cached checkout links after changing commercial accounts.

MCP Server

Enable or disable the tenant MCP server, set per-user connection limits, copy client configuration, toggle exposed tools and optional parameters (including Quest when your organisation allows it, plus formal-proof export and formal-statement lookup when you offer them), review analytics and the live event log, manage access keys, and reset active connections. The live event log shows the query ID for each question so it can be matched to the answer in History. A completed answer returned to a connected tool includes that same query ID. Organisation admins can also ask connected tools who registered or used the workspace — with registration and activity dates — without treating Library authors as users.

Embed Chat

Available after a platform super admin enables Embed Chat for your tenant.

1. Copy the installation snippet and create site keys with allowed origins.
2. Configure widget settings and optional debug mode.
3. Review cost, active keys, daily usage, and recent activity.
4. Let visitors ask from an external site; offer authenticated handoff into the full workspace.

Landing pages & SEO

Create, edit, reorder, and delete pre-login pages with menu label, heading, subtitle, and body. Publish under friendly URLs, control indexing, canonical URL, meta title/description, social image, and influence sitemap/robots output.

Onboarding persona

• Configure Step 2 title, subtitle, and footer copy.
• Create and reorder introduction routes with kicker, title, subtitle, and tenets.
• Map persona choices to default introduction routes.
• Edit Getting Started step titles and bodies.

Site Report

Download a PDF snapshot of tenant configuration — branding, models, plans, Guiding Principles, question bank, MCP, tags, onboarding, landing pages, and related settings — for operational review or change control.

Your role

You sit above individual organisations. Day-to-day research and Library curation happen inside each tenant; you ensure those workspaces exist, stay isolated, and remain commercially and operationally sound.

Sign in

1. Open the platform admin host for your Quaestra.ai instance.
2. Sign in with Google or Microsoft when your email is allowlisted as an active super admin — or use the break-glass credential path when configured for emergency access.
3. Confirm your session identity on the console, then proceed to Overview.
4. Sign out when finished, especially on shared workstations.

Overview

The dashboard summarises tenant count, active tenants, platform-admin count, audit-event volume, tenant status, and recent platform activity, with quick links into common tenant actions.

Tenant lifecycle

Provision

1. Open Tenants and start provisioning.
2. Provide a slug, display name, initial admin email, optionally the customer’s main website address, and a model spend cap (USD, with how often it resets, or no cap).
3. Confirm creation — the platform creates an isolated workspace and the first tenant administrator.
4. Setup site opens as soon as the organisation is ready. If you entered a website, it is scanned and drafted automatically — wait for the progress steps to finish, then review. You can also close it and run Setup site later from the tenant row. The drafting model is chosen once under Global settings.
5. Share the tenant hostname with that administrator so they can refine branding, Library, and plans.

Setup site

1. Under Global, choose the Site setup model (used for every organisation).
2. On an active tenant, choose Setup site (also offered after provision).
3. Paste the customer homepage URL.
4. Review suggested logos (or upload one), site title, tagline, and assistant name.
5. Edit the front-page intro and About page body directly as they will appear — the same styled editor used under Landing Pages.
6. Optionally apply a drafted default Guiding Principle — an existing default is snapshotted first.
7. Optionally apply a design theme proposed from the customer website (this also enables Design for the organisation so tenant admins can refine and restore versions).
8. Review organisation details and the drafted Terms and Privacy (filled from a neutral template with the organisation’s name, website, and contact details). Edit them as visitors will see them, and confirm with the customer before go-live.
9. Confirm and apply. Check the per-section results, then open the tenant or use Login as admin to verify the public site.

Operate

• List all tenants and open a tenant’s public/admin host.
• Open Settings on a tenant to reach domains, analytics, Embed Chat, Design, Retrieval, and Spend controls.
• Activate, suspend, or otherwise change status when an organisation should be paused or restored.
• Delete a tenant only with the confirmation the console requires — this removes the organisation workspace.
• Use Login as tenant admin to mint a temporary support handoff into an existing tenant admin account without sharing their credentials. The target must already be a tenant admin.
• Use Telemetry on an active organisation to open that organisation’s live model request log in a new tab. Model calls that belong to an answer show that answer’s query ID, which matches History. Organisation admins cannot open this log.
• Use Copy MCP URL to copy that organisation’s MCP address for an approved external tool (for example ChatGPT).

Spend cap

1. When provisioning, set a model spend cap in USD and how often it resets, or choose no cap. If you leave the amount blank, the platform default is used for a new organisation only.
2. On an existing organisation, open Settings then the spend tab, or use the organisation-row spend control. Remaining usage for the current period is loaded live.
3. Change the cap, reset period, or remove the cap, then save. Refresh to reload remaining usage without changing the cap.

Capacity

Open Capacity to review remaining storage on the host that runs this Quaestra.ai instance, how busy the service is, and how much storage each organisation uses. Refresh to remeasure. Use the organisation table to spot large libraries or leftover recoverable trash after a tenant deletion.

Tenant MCP (Guiding Principles)

As a platform Super Admin you can connect an approved external tool to any organisation’s MCP address and manage Guiding Principles with the same control as the tenant Guiding Principles page — including router settings, routes, the default principle, and save / list / restore versions. Organisation (tenant) admins keep the web UI; they do not receive these write tools over MCP.

1. On Tenants, choose an active organisation and select Copy MCP URL.
2. Paste the address into your external tool’s MCP / connector settings.
3. When prompted, authorise with Google or Microsoft using the same email as your platform Super Admin account (not a separate organisation-only login).
4. Connecting adds you as an organisation admin in that tenant if you are not already a member, so Ask and history are attributed to you. Guiding Principles tools appear because you are a platform operator.
5. Tune routes and settings, test sample questions, then save a version. Restore an earlier version if a change looks worse.

Custom domains (Bring Your Own Domain)

1. Open a tenant’s domain settings.
2. Add a hostname; follow the DNS setup instructions shown in the console.
3. Mark hosts active or pending; set the canonical host when multiple domains exist.
4. Remove a host that is no longer required.
5. Configure analytics separately for the tenant subdomain and each custom domain as needed.

Analytics configuration

Per tenant, review analytics status, configure an optional Clarity project ID, and enable or disable tracking on the platform tenant host and on each custom domain. This screen configures host-level session insights; it is not a cross-tenant product-usage dashboard.

Embed Chat platform gate

Enable or disable Embed Chat for a tenant from the platform console. This commercial/platform entitlement is independent of the widget keys and settings the tenant admin configures inside their workspace.

Retrieval platform gate

Open a tenant drawer and choose the Retrieval tab. Set policy to Off (legacy evidence path), Tenant (organisation admins configure quality and matching), or Force hybrid (require improved matching when the embeddings index is ready). Use the Quest switch to allow or stop Quest for that organisation. This gate is independent of the quality and visibility controls tenant admins set under Configure App → Retrieval.

Super-admin management

1. Open Admins to list platform operators.
2. Add an operator by email (and optional name). An invitation email is sent when messaging is configured.
3. Review status, who added them, and last login.
4. Suspend, reactivate, or remove another operator when access should change.

Audit log

Review time, actor, action, resource, details, and IP. Filter by action or actor, paginate, and refresh. Sensitive actions typically include break-glass and SSO sign-in, sign-out, tenant provisioning/status/deletion/impersonation, analytics, Embed Chat, and Retrieval policy changes, domain changes, platform-admin changes, instruction-template updates, and commercial setting updates.

Global settings

Configure platform-wide defaults that apply to every organisation:

• Chat thread context budget — controls the Ask context indicator and how much previous conversation is retained.
• Model request time limit — the longest Quaestra.ai will wait for a single non-streaming model response before reporting an error. Covers background work (document processing, section summaries, equation checks, Quest steps) and non-streamed answers. Streamed answers and the separate first-response wait are not affected. Applies to all organisations immediately.
• Site setup model — the model used when Super Admins run Setup site to draft branding and copy from a customer website.
• Design vision model — the model that reads screenshots when an organisation uses Design → Analyse shots and has no image-reading model of its own.
• Specialist formula reader credentials — optional platform credentials used when an organisation sets Math Equation Pipeline to High-fidelity and chooses the specialist formula reader instead of the PDF Extract Model.

Prompts

Review and update the platform instruction templates that shape Ask answers, document ingest, Virtual Review, routing classifiers, and related flows. Built-in defaults ship with Quaestra.ai; overrides apply to every tenant immediately. Reset any template to its built-in default when needed. These templates are Quaestra intellectual property and are not visible inside tenant workspaces.

1. Open Prompts in the platform console.
2. Choose a category and template from the list.
3. Edit the body and save — or reset to the built-in default.
4. Confirm the change in the audit log if required by your operating practice.

Commercial (billing) settings

Set platform-wide commercial credentials (secret key, webhook secret, and publishable key) that connect billing across tenant workspaces unless a tenant-specific arrangement overrides them. After changes, tenant admins may need to reset cached checkout links inside Plans.