Read-only by design
Brill can see everything in your workspace and change nothing: not by accident, and not if someone asks it to.
PlatformBrill
Ask your coverage anything. Get an answer you can check.
Brill is the assistant built into Brillaince. Ask what changed, why it changed and what to do about it. It answers from your own workspace data, shows the numbers behind every answer, and says so plainly when it doesn't know.
Read-onlyCan look, can never change
Your dataNever the open internet
SourcedEvery number, traceable
"Why did our coverage drop on Tuesday?"
It didn't. A naive assistant would confidently explain a decline that never happened, and you'd act on it.
1 / 6
- You ask
In plain language
No query syntax, no filters to configure first. The question is typed the way you'd ask a colleague.
- +1s
Routed to your data
Brill decides this is a workspace-data question rather than a product question, so it reaches for real rows instead of documentation.
- +2s
Pulled through fixed tools
It reads through the same queries your dashboard uses. There's no free-form access, which is why its numbers and your reports never disagree.
- +2s
It checks freshness first
Tuesday morning's press run finished late. The dip is a pipeline timing artefact, not a real fall in coverage, and Brill knows the difference.
- +3s
It corrects the premise
"It didn't drop. Tuesday's run finished late and filled in by the afternoon. Coverage is up 9% across the week." The honest answer, not the agreeable one.
- Always
With its sources attached
Press count, the window used, and the pipeline log, so you can check it rather than take it on faith.
Everything feeding this layer.
Brill reaches your workspace through a fixed set of tools, the same queries the dashboard renders from. It cannot invent a number because it cannot reach one that doesn't exist.
What it can answer from
Coverage trendsThe same rows the Overview page charts
Share of voiceYour competitors, same methodology
Crisis stateOpen alerts, severity and history
AI answersEngine results and the claim ledger
Sources & freshnessWhen each pipeline last ran, and what it found
How the product worksFeature and methodology questions, answered straight
It answers from your data, or it says it can't.
Watch it correct the premise of a question rather than agree with it. The pipeline ran late that morning; coverage never dropped.
BrillReading your workspaceWhy did our coverage drop on Tuesday?
It didn't. Tuesday morning's press run finished late, so the count filled in by the afternoon. Across the week, coverage is up 9%.
Press · 214 storiesPipeline log
Is the 42% share of voice number reliable?
Yes for press and social, which carry 1,284 and 3,420 rows this window. I'd treat the AI-answer slice carefully: it's built on 50 prompts, which is above the floor but thin enough to move week to week.
Share of voiceReliability floor
Ask Brill a question
The screen your team actually opens.
Illustrative numbers. Your workspace runs on your brand's real data from day one.
Ask Brill
Read-only · your workspace
What Brill can reach
- Coverage, mentions, sentimentRead
- Competitors and share of voiceRead
- Crisis alerts and historyRead
- Edit, delete or trigger anythingNever
Typical questions
Why did coverage dip?Who gained share?What's open in crisis?Is this number reliable?
Answer sourcingFigures with a traceable origin
100%
From raw signal to something you can act on.
- 01
Ask in plain language
"Why did coverage dip on Tuesday" or "which competitor gained share this month": no query language, no dashboard to master first.
- 02
Answered from your real numbers
Brill reaches your workspace through the same fixed tools the dashboard renders from, so what it says and what your reports say always agree.
- 03
Sourced, or refused
Every answer names what it drew from. When the data genuinely can't answer, Brill says so instead of producing something plausible.
Brill, in full.
Always sourced
Each answer names the press count, social window or report it came from.
Says when it doesn't know
A refusal rather than a guess, when the data genuinely can't answer the question.
Freshness-aware
It knows when a pipeline ran late, so it never reports a timing artefact as a real drop.
Product questions too
Ask what a feature does or how a figure is calculated and get the real answer, not marketing copy.
Injection-resistant
Hardened against attempts to talk it out of its boundaries, and tested that way rather than assumed safe.
The same layer, three different jobs.
Answer a client mid-call
Ask the question while they're still on the phone, and get a figure you can defend rather than promising to check.
Sanity-check before the board
Confirm a number and its source before it goes on a slide, including whether the data is thin.
Skip the dashboard tour
New team members get answers on day one without first learning where every chart lives.
With and without Brillaince.
Without BrillainceWith Brillaince
What it can touchAssistants that can change your dataRead-only, enforced by design
Where answers come fromThe model's general knowledgeThe same data your dashboard uses
When it doesn't knowA confident, plausible inventionIt says so, plainly
Pipeline timingReports a late run as a real dropKnows the difference and tells you
What this deliberately doesn't do
Every monitoring tool's page is wall-to-wall claims. Here is where the honest line sits.
- It cannot act, only answerNo inserts, no updates, no deletions, no triggering pipelines. Read-only is enforced in the design, not left to the model's good behaviour.
- It won't answer from the open internetIf it isn't in your workspace, Brill doesn't know it. That's a narrower assistant on purpose: a narrow one you can trust beats a broad one you can't.
Works with the rest of the platform.
Brill isn't a standalone tool. It feeds, and is fed by, every other layer Brillaince tracks.
About Brill.
Can Brill take actions in my workspace?
No. Brill is read-only by design: it can answer questions about your data and can never insert, update, delete or trigger anything.
Will its numbers match my reports?
Yes. It reaches your data through the same queries your dashboard pages use, so there is no second, disagreeing source of truth.
What if someone tries to trick it?
It's been hardened against prompt injection and tested against it, rather than assumed safe. Its boundaries are enforced by what it can reach, not just by instructions.
Does it know how the product works?
Yes. Product and methodology questions are answered from the product's own definitions, so 'how is this calculated' gets a real answer.