Working with the AI assistant

Overview

The dashboard assistant on Home is your natural-language interface to Tendersight. Ask in plain language — it searches tenders, opens pursuits, edits requirements, updates team settings, runs quotation workflows, and surfaces document gaps. Sensitive writes pause for your approval before anything changes in your workspace.

This guide covers when to use the assistant vs the UI, approval flows, capabilities and limits, effective prompts, and troubleshooting. For composer controls, queued messages, and session management, see Dashboard assistant.

Tip: Before your first multi-step request, confirm:

Correct active team selected in the sidebar
Web search toggled appropriately (on for market questions, off for internal-only audits)
Intelligence level matches task complexity — Maximum for multi-module bid workflows
You know how to read What I am doing steps and Confirmation required cards
You will Approve or Reject deliberately — never rush past amber summaries

Important: The assistant acts on behalf of your signed-in user within team permissions. It cannot bypass subscription limits, cross-team data boundaries, or approval gates. Destructive actions always require explicit confirmation.

Assistant vs traditional UI

Use the assistant when…	Use the UI when…
You know intent but not navigation	You need precise field-by-field data entry
Exploring — "find electrical tenders this week"	Bulk-editing hundreds of questionnaire rows in a table
Multi-step orchestration across modules	Visual review of PDF layout and pagination
Quick status on open pursuits	Kanban drag-and-drop for pipeline status
Auditing empty team settings fields	Uploading large files with progress bars
Summarizing missing requirements	Configuring alert filter dialogs visually

The assistant complements the UI — it does not replace structured modules like Financial offer pricing grids or Attribute mappings matrices.

Where the assistant lives

Location	What you get
Home	Full composer, starter chips, Recents, credits badge
Assistant session URL	Deep link to a conversation — Assistant session
New chat	Fresh thread for unrelated tasks — New chat

Placeholder in the composer:

"Search tenders, manage pursuits, edit requirements, quotations, documents, or team settings"

Attachments: drag-and-drop or Add photos & files from the + menu. Upload tender PDFs when you want document context in the thread.

Starting conversations

New chat vs continuing

Action	When
New chat	Unrelated task — keeps Recents titles meaningful
Continue Recents row	Follow-up on same pursuit or audit
Starter chip	One-click common workflows on empty composer

First message in a new thread creates a session and updates the URL.

Starter suggestions (empty chat)

Click to send immediately with current intelligence and web search settings:

"Search recent tenders relevant to electrical equipment."
"Open the latest pursuit and summarize the missing requirements."
"Create quotations for unmapped requirements and show the best matches."
"Review our team legal representative and banking details."

Follow-up chips

After replies, up to four context-aware suggestions may appear (hidden while streaming, during approvals, or on read-only chats).

Examples after a tender search:

"Open the most relevant tender"
"Create a project from this tender"

Composer settings

Open the + action menu on the composer.

Setting	Default	Guidance
Web search	On	Disable for internal-only questions (team settings audit, pursuit status)
Intelligence	Auto	Raise to Maximum for complex multi-step bid workflows
Attachments	—	Tender PDFs, screenshots, spreadsheets for context

Intelligence levels

Level	Trade-off
Auto	Everyday balance of cost and quality
Fast	Lower latency; higher error risk on long chains
Balanced	Middle ground
Maximum	Best for orchestration across questionnaire, offers, and documents

Levels consume team credits. Heavy features elsewhere in the app may also show credit estimates before confirmation.

Approval gates (human-in-the-loop)

When the assistant needs you before continuing, a card appears above the composer. The composer and follow-up chips disable until you resolve the card.

Cancel run aborts the paused operation. Do not use New chat to cancel — use Stop (when streaming) or Cancel run on the approval card.

After you approve or reject, a resolution bubble records "You approved this action" or "You rejected this action".

Input required

Element	Purpose
Badge	Input required
Fields	Text areas, single-line inputs, checkboxes
Actions	Continue · Cancel run

Use when the assistant needs parameters you did not provide — open questions, missing pursuit name, ambiguous dates.

Decision required

Element	Purpose
Badge	Decision required
Controls	Radio (single choice) or checkboxes (multiple) with optional descriptions
Actions	Submit decision · Cancel run

Use when multiple valid paths exist — which tender to open, which team scope, which requirement batch to import.

Confirmation required

Element	Purpose
Card style	Amber confirmation (distinct from generic input card)
Summary	Plain-language description of the sensitive action
Actions	Approve (primary) · Reject (outline)

Always read the summary line by line before approving.

Examples that require confirmation:

Action category	Example summary
Delete requirement	"Delete '{requirement}' from project '{project}'"
Create pursuit	"Create project from tender '{title}'"
Import requirements	Batch import from extraction results
Quotation generation	Start quotation task for unmapped lines
Team settings	Update company profile, legal representative, or banking
Documents	Create or AI-fill pursuit documents

Non-exhaustive — treat every amber card as binding.

What the assistant can do

Reads and acts across the platform within your permissions:

Domain	Examples
Notices / tenders	Search, filter, open overview, summarize key facts
Pursuits / projects	Create, open, summarize status, list gaps
Requirements	List, create, update, delete (delete → confirmation)
Quotations & financial offer	Start generation, summarize unmapped lines
Team settings	Read and update company profile, banking, legal representative
Analytics	Company and authority statistics summaries
Alerts	Discuss and guide alert configuration
Web search	External context when toggle is on

Actions respect:

Team membership — only data for teams you belong to
Subscription limits — pursuit caps, credit balance, integration gates
Approval gates — writes pause until you confirm

What the assistant cannot do

Limit	Why
Edit notices	Notices are read-only market imports
Submit bids to authority portals	Submission is external to Tendersight
Bypass Confirmation required	By design — you must approve sensitive writes
Access another team's pursuits or settings	Team boundary enforced
Force notice sync	Updates arrive on Tendersight's import schedule
Replace bulk table editing	Large matrix edits are faster in module UI
Guarantee legal correctness	You review all compliance answers and PDFs
Run when credits exhausted	Upgrade or wait per Billing
Write in read-only teammate sessions	Open New chat for your own writable thread

Caution: Do not paste passwords, API secrets, or personal data unrelated to the bid into chat. The assistant is for procurement workflow — not a secrets vault.

Tool steps — "What I am doing"

Expand What I am doing on assistant messages for transparency.

Header example	Meaning
"Searching tenders"	Catalog query with filter detail
"Creating 2 projects."	Pursuit creation in progress
"Added 1 requirement"	Questionnaire write

Each step shows:

Each step shows an icon and a plain-language label (for example Update company profile, not technical jargon).

- Status: In progress, Queued, Completed, or Could not complete this step

- Optional detail lines — search query, project name, result counts

- Source links opening in new tabs

Panel auto-expands while steps run; collapse manually if you prefer a cleaner thread.

Effective prompt patterns

Be specific about team, pursuit title, country, CPV, and deadlines — especially when you belong to multiple teams.

Tender discovery

Search tenders in Romania for CPV 45310000 published in the last 14 days with estimated value above 500000 EUR.

Pursuit status, settings, documents

Open pursuit "{exact title}" and list requirements still missing an owner.

Read our company profile and banking details — flag empty fields required for Romanian public tenders.

In the active pursuit, list required documents not yet uploaded.

Be specific about team, pursuit title, country, and CPV. Use Maximum intelligence for multi-step chains; expect multiple approval cards.

Queued messages

Sending while the assistant is still replying queues your message (FIFO) instead of interrupting.

Control	Behavior
Send during stream	Message enters Queued messages panel
Send next (↑)	Jump queued item to front; may stop current run first
Edit	Load queued item back into composer
Remove	Delete from queue
Composer empty + streaming	Stop button cancels current run

When the assistant finishes, queued messages send automatically in order unless you stopped the run.

Read-only teammate sessions

Recents includes teammates' chats. Opening another member's session is read-only:

Composer disabled
Cannot send, queue, stop, or approve
Rename and delete disabled in session menu

Start New chat for your own writable session.

Use read-only mode to learn prompt patterns — not to execute actions on their behalf.

Credits and subscription limits

Signal	Where
Credit balance	Header credits badge on Home
Heavy AI estimate	Dialogs before some in-app AI features
Pursuit limit reached	Creation blocked with upgrade message
Integration gates	Assistant surfaces plan messages

The assistant reads limits; it cannot override them. Team admins handle billing — Billing.

Safety checklist

Read Confirmation required summaries before Approve
Verify active team in sidebar before settings or pursuit writes
Regenerate documents after assistant updates company profile or banking
Reject if the summary does not match your intent
Use Stop or Cancel run instead of abandoning via New chat mid-approval
Do not paste secrets into chat
Re-check PDFs and compliance answers — assistant output is a draft until you validate

Troubleshooting

Symptom	Likely cause	Action
Assistant cannot update settings	Wrong team or missing membership	Switch team; confirm role
Wrong pursuit opened	Ambiguous title	Name pursuit exactly in prompt
Run stuck or looping	Prompt too large or ambiguous	Stop; retry shorter phased prompt
Missing tender in results	Filters too narrow or notice not synced	Broaden search; enable web search; check Alerts
Composer disabled	Approval card pending or read-only session	Resolve card or start New chat
"Could not complete this step"	Permission, limit, or missing data	Read step detail; fix data; retry
Placeholders still in PDFs	Settings updated but documents not regenerated	Regenerate in documents module
Credit error	Balance exhausted	Billing upgrade or wait for reset
Teammate's chat won't send	Read-only session by design	New chat under your identity

After a failed run: Stop → read the last failed step in What I am doing → fix data in UI → retry with a narrower prompt (one module, one pursuit).

During Getting started and First pursuit, the assistant audits settings, summarizes gaps, and starts quotations — kanban moves and external submission stay in the UI.

FAQ

Does the assistant automatically apply changes?

No. Sensitive writes show Confirmation required (or input/decision cards). Nothing changes until you Approve or Continue.

Can I trust requirement text the assistant adds?

Treat it as a draft. Always verify against authority PDFs in Questionnaire and In-depth analysis sources.

Why did my message queue instead of sending?

The assistant was still streaming a reply. Queuing preserves order. Use Stop if you need to cancel the current run first.

Will the assistant delete data without asking?

Destructive actions require Confirmation required. If you Reject, the run cancels with no delete.

How do I cancel mid-run?

Stop when composer is empty and assistant is streaming. Cancel run on approval cards. Do not rely on navigating away without stopping.

Dashboard (Home)
Dashboard assistant
Assistant session
New chat
Search chats
Getting started workflow
First pursuit workflow
Team settings
Scheduled tasks
Billing