Introducing Selena
Selena AI is the AI sales engineer for security integrators — it reads your drawings, resolves your parts, and writes estimates inside the system your team already quotes from.
Selena AI is an AI sales engineer for security integrators. It does the work that sits between a plan set and a priced quote: counting devices off a drawing, matching them to real parts, building the estimate, and writing the documents that go around it.
Selena works inside your system of record. Parts resolve against your live catalog, estimates are written back as real line items, and Selena acts as the connected user — so permissions and audit trails stay exactly as they already are. Nothing comes back as text you have to re-key.
See what shipped recently in the release notes.
Already signed up? Start here
Open Selena
Sign in and start a chat. You will need an active organization.
Connect an integration
Nothing useful happens until Selena can see your catalog and write estimates. WeSuite is the integration available today.
What is Selena good at?
Turning drawings and BOMs into estimate line items
- Importing a CSV or Excel bill of materials into an estimate, with
/import-bom - Reviewing an estimate's existing BOM for gaps, incompatibilities, and uncertain
items, with
/analyze-bom - Reading devices off a floor plan image or PDF, with
/analyze-floor-plan— a counted, grouped device list you read in chat - Full drawing-set takeoff, driven from the connected system's own front end rather than from chat — every page classified, then counted and grouped into estimate line items. With WeSuite, that surface is QuoteAnywhere
Parts that come in through /import-bom and /analyze-bom are resolved against your
live catalog — including your organization's favorites — and an ambiguous match comes
back as a choice, not a guess.
/analyze-floor-plan reads the drawing; it does not yet write to an estimate. The
detected devices are not resolved against your catalog, so the panel below them
reports that estimate comparison is unavailable and leaves import disabled — a
deliberate stop, so a detection cannot silently duplicate parts already on the
estimate. To get devices from a drawing into an estimate today, use the takeoff in
your connected system (QuoteAnywhere, for WeSuite), or list them in a spreadsheet and
run /import-bom.
The paperwork around a quote
- Scopes of work, generated or tightened up from what you already have
- Proposal cover letters
- Notes: summarize, rephrase, clean up, or answer questions about them
Questions about live data
Ask in plain language and Selena reads your connected system directly:
- Estimates and quote packages — line items, categories, totals, scope
- Customers, their locations, and their contacts
- Projects, opportunities, and pipeline rolled up by stage
Selena can also write back. How much stands between your request and the write depends on which write it is:
| What you ask for | What happens before it is written |
|---|---|
Import a BOM with /import-bom, or accept the missing parts /analyze-bom found | You approve it in a panel. Nothing is written until you accept the resolved parts. |
| Create or update a customer or contact, add a location or account note, create a project, log an opportunity, move a project stage | Selena is instructed to ask first — to state what it is about to change and wait for your go-ahead. |
| Add a note to an estimate, add a catalog part or a one-time part, create an estimate | It is written immediately. No confirmation, no panel. |
Ambiguity is handled the same way everywhere: if a part matches several catalog entries, or a required folder or customer is missing, Selena stops and asks rather than guessing.
Only the review panel is a hard gate. The middle row is instructed behavior, not a block — if Selena skips the question, nothing in the system stops the write. Two cases are worth knowing about: "add a note to estimate 137581" writes that note immediately, and moving a contact to a different account happens with no confirmation step at all.
Selena never retries a failed write on its own — it reports the error and stops. That leaves the decision to you, and whether repeating a request is safe depends on the write:
- Updates are safe to repeat. Changing a customer, contact, or project overwrites the same fields with the same values, so asking twice lands the same record — though a wrong value is not recoverable by asking again.
- Most creates can be guarded, but only briefly. Creating a customer, contact,
location, account note, project, or opportunity, and adding a line item — catalog
part or one-time part alike — can carry a duplicate guard: a repeat comes back as
duplicate_ignoredrather than writing twice. It is a safety net, not a guarantee. It applies only when the repeat carries the same key as the original, and it lives in memory for about two minutes inside the process that handled the first call — so a repeat with no key, one after the guard expires, or one that lands in another process writes again. - A duplicate contact is refused by the connected system itself. In WeSuite, creating a contact is the one create protected independently of that guard: a repeat matching an existing contact's name, company, and email comes back as an error instead of a second record. Change any of those three and it is a new contact, so this is not a substitute for looking the contact up.
- Two creates have no guard at all. A new estimate and an estimate note have no duplicate protection on either side. Ask twice and you get two records.
When a write fails or its outcome is unclear, check the record in the connected system before asking again. That is the only reliable way to tell a write that never landed from one that landed and failed to report.
Sales analysis
Upload an Excel workbook and ask what is in it: patterns, trends, and outliers, with follow-up questions answered from the same analysis rather than a fresh upload.
To get the best results from Selena
- Attach the file with the request. A floor plan, a BOM spreadsheet, or a workbook in the same message is what the specialist agents work from.
- Say which estimate you mean. Type
@in the composer to search your connected system's estimates and attach one as context — that is how estimate, BOM, and import requests land on the right quote package. - Reach for a
/command when you already know the workflow. The/menu routes straight to a specialist and skips the guesswork. Without one, Selena routes your request itself. - Review before accepting. A BOM import comes back in a panel where you approve what gets written. Read the resolved parts and the add-versus-update decisions before you accept.
- Keep one job per conversation. Follow-ups reuse the session, so a chat that stays on one estimate answers faster and more accurately than one that switches subjects.
The workflows that work best
- An estimator drops a floor plan into chat and reads back a counted device list — a check against their own takeoff before the quote is priced.
- A subcontractor's BOM spreadsheet arrives by email and goes straight into the estimate it belongs to, with unmatched rows added as one-time parts.
- Before a quote goes out,
/analyze-bomreads the estimate back and lists what is missing or incompatible. - The scope of work and cover letter are drafted from the estimate that already exists, instead of from the last job's copy.
General product features
Chat
The main surface. One assistant routes each request to the specialist that handles
it, so you do not have to pick an agent. / commands and @ estimate mentions live
in the composer.
Artifacts
Generated work opens in a side panel — documents, code, spreadsheets, and a purpose-built estimate view with financial summary, BOM rows, and category breakdown. Artifacts are versioned, so you can diff an edit against what came before.
Action panels
A BOM import is reviewed before anything is written: resolved parts, the folder each one lands in, and whether it is an add or an update. It is the only write path with a hard gate — everything else (an estimate note, a single line item, a customer or contact change) is written without one. A floor-plan detection shows the same panel with import disabled, because those parts are not resolved.
Dashboard
Integrations, OAuth applications, automations, usage analytics, and billing.
Documents
Scopes of work, cover letters, and notes are produced as editable artifacts rather than chat text you have to copy out.
Usage and billing
Every agent session and interaction is tracked against your organization, so token usage and cost are attributable per user and per workflow.
Getting access
Sign in
Open Selena and sign in. Selena scopes everything by organization, so you need an active organization selected — not just an account.
Have an admin connect an integration
An admin configures the system your team estimates in, once, in
Dashboard → Integrations. Until that is done, every user's connection status reads
not_available.
WeSuite is the integration available today — see the WeSuite guide for what each field expects.
Connect your own account
Each user then authenticates with their own credentials for that system. Selena acts as that user, so it can never see or change anything that user could not already.
See Integrations for how connection works and what is available, or the WeSuite guide for the full setup of the integration most teams start with.
Feedback
Selena AI is developed alongside the integrators who use it, and what you run into shapes what gets built next. Ask questions, report what broke, and tell us what is missing in the Selena AI Discord.
When something looks wrong, the three facts that narrow it down fastest are: the workflow you were in, whether it affects one user or everyone in the organization, and the estimate or file involved.