AI built into every Delalify product

Beryl answers questions, drafts messages, flags the accounts that need attention, and completes multi-step work from the contacts, conversations, files, and team records that your organization keeps in Delalify.

Answers built from your own records

You can use Beryl inside People, Channels, OneCloud, Teams, Tables, Builder, and Wallet. When you ask Beryl about a contact, it searches the contact's profile, your conversation history with them, the files that you share, and the access records in Teams at the same time. Beryl checks your role in Teams before it answers, so each answer contains only the records that your role grants you access to. Beryl answers a question that covers several products in 10 seconds, and each part of the answer links to the record that it comes from.

Use the records in every Delalify product for answers, alerts, drafts, and workflows

Beryl works from the contacts, conversations, files, and team records in your workspace whenever it answers a question, flags an account, drafts a message, or performs a workflow step.

What Beryl does across your Delalify workspace

Your team stores what it knows about one customer in several products, such as the profile in People, the threads in Channels, the signed documents in OneCloud, and the shared workspace in Teams. Anyone who needs all of it before a call will have to open each product and search it separately.

When you type a question the way that you ask a colleague, Beryl searches every product that you have access to. If you ask for everything about one client, Beryl returns the primary contact, your threads with them from the last 60 days, the signed statement of work, the pipeline record from Tables, the latest invoice from Wallet, and the colleagues who can open the shared drive. Each result opens the record behind it in its own product. Beryl includes only the records that your account has permission to open, because it applies the access rules from Teams to every search.

Between two account reviews, you can miss several changes: a top account that goes six weeks without a reply, a partner who opens the same shared file fourteen times in two days, or a customer who stops answering while an invoice passes 30 days overdue. Channels, OneCloud, and Wallet each record one of those changes, so whoever prepares the review will have to search each of them.

Beryl measures how quickly each contact replies in Channels, how often people open your shared files in OneCloud, who on your team opens the shared workspaces in Teams, which details change in People, and whether customers pay their Wallet invoices on time. When a contact goes quiet, Beryl flags their profile and notifies the account owner with the signals that change. Before each meeting on your calendar, Beryl writes a one-page brief that lists the reopened files and the current renewal risk score for the account that the meeting concerns.

A general-purpose AI tool writes its draft from its training data and from whatever you paste into the prompt. Before its draft can mention your five earlier conversations with a contact, your last three proposals to them, or the objection from your last meeting, you will have to paste each of them in yourself.

Beryl retrieves every message, file, and note that your team holds for a contact before it drafts a reply, a summary, or a follow-up. Its draft names the last meeting, the objection that the contact raises there, and the deliverable that you commit to, so you can start the message by editing a draft.

When your team onboards a new client, someone will have to create the contact in People, send the welcome messages from Channels, set up a shared drive in OneCloud, assign a colleague in Teams, and book the kickoff meeting, opening a different product for each step.

Once you build the workflow, Beryl performs each of its steps in order every time it starts. When someone adds a new client in People, Beryl starts the welcome series in Channels, creates a shared drive in OneCloud, sets up a client portal in Builder, adds the right colleagues in Teams, issues the first invoice from Wallet, and books the kickoff meeting on the calendar. Beryl stops at any step that needs approval and continues once the approver accepts it.

When your staff use AI tools without shared rules, an employee can paste a customer's card number into a prompt, while a support agent can ask a chatbot about a patient's records. Your compliance team will hear about each of those requests only after the AI provider has the data.

Beryl Guard masks personal data, financial identifiers, and health information in each request before Beryl sends it to a model. Beryl stops sending requests for a team, a project, or a user once they spend the budget cap that you set for them. Beryl refuses any request on a topic that you restrict and records each of these decisions in an audit log that your compliance team can query.

To prepare a quarterly review of your healthcare partnerships, someone will have to gather the partner contacts, summarize each conversation, collect the shared documents, and check who on your team has access, all before they can start writing the analysis.

When you give Beryl a goal, a Beryl agent plans the steps and works through them until it delivers the result. For the partnership review, the agent collects the partner contacts from People, summarizes the last quarter of threads in Channels, gathers the shared documents from OneCloud, checks access in Teams, and writes the brief. You can follow each step as the agent completes it and answer whenever it stops to ask you a question. The agent can open only the records that you have permission to open.

1

Question to search every product

0

Unmasked personal data sent to a model

24hr

Longest agent session

What changes for your team with Beryl

One answer from every product

Beryl searches your contacts, conversations, files, and access records for each question, so a single answer can include a contact from People, a thread from Channels, and a contract from OneCloud.

Your access rules apply to every answer

Beryl answers each colleague from the files and records that their role in Teams permits them to open, so the roles that you already set in Teams apply to Beryl as well.

Compliance rules that you set once

Once your compliance team sets the masking rules, budget caps, and restricted topics in Beryl Guard, Beryl checks every request against those rules before sending it to a provider and logs the result of each check.

Beryl inside the app that you are using

You can call up Beryl from whichever Delalify screen you have open, so you can ask about a contact from that contact's profile.

How organizations can use Beryl

A cardiologist reads one brief before each referral appointment

Before a cardiology referral, Beryl retrieves the patient record from People, the last six months of messages with the referring physician from Channels, and the imaging results and lab files from OneCloud, then writes a brief for the specialist to read before the appointment. When the specialist copies part of the brief out of the clinical system, Beryl masks the personal health data in the copied text.

Result

Beryl puts the patient's history, the open questions, and the most recent imaging result on one page, so the specialist reviews that page before each appointment. Because the brief also lists the referring physician's questions, the specialist can answer that physician with the brief open.

A cardiologist reads one brief before each referral appointment

A registrar's office answers 4,000 transfer-credit questions every summer

Students ask the registrar about transfer credit, prerequisite equivalencies, and graduation audits. For each question, Beryl retrieves the student profile from People, the email thread with the advisor from Channels, the transcripts stored in OneCloud, and the enrollment records, then drafts an answer that cites each of those records.

Result

Advisors review Beryl's draft for each question and send it, so they write only the corrections. Each answer links to the transcript or enrollment record that it quotes, so a student can check the decision against their own records.

A registrar's office answers 4,000 transfer-credit questions every summer

A payments operations team investigates accounts that break their usual pattern

Beryl compares each customer's transactions with that customer's usual pattern and flags any account whose activity differs from it. The operations dashboard lists each flagged account with its three most recent support threads, its last wire transfer, and its linked compliance notes. Beryl Guard masks account numbers and other sensitive identifiers before any model receives them, while Beryl writes every investigation to an audit log that your compliance team can query.

Result

Each fraud analyst starts an investigation from the flagged account's screen, which already lists the support threads, the last transfer, and the compliance notes. When a regulator asks about a case, the compliance team exports that investigation from the audit log.

A wholesale distributor wants its dormant retail accounts back

A regional food distributor has 200 retail buyers who no longer place orders. Beryl retrieves each buyer's profile from People, along with the buyer's order history and their last messages in Channels. For every buyer, Beryl drafts a personal note that names the product that the buyer orders most and a new product line that suits their shelves. Channels sends each note at the hour of day when that buyer usually opens email.

Result

The sales team reviews each of the 200 drafts before Channels sends it, so the team spends its own time calling the buyers who reply. Because each note names products from that buyer's order history, every buyer receives a note written for their store.

A wholesale distributor wants its dormant retail accounts back

An IT service desk receives one weekly brief on its 80 managed accounts

Each week, before the service desk's planning meeting, a Beryl agent reviews all 80 managed accounts. For each account, the agent retrieves the week's support threads from Channels, the runbooks and incident reports that engineers open in OneCloud, and the engineers' access activity in Teams. The agent then writes a one-page brief for each client, ranks the briefs by ticket volume, and names the open incident, the service-level agreement at risk, and the technician best placed to follow up.

Result

The service desk lead opens the planning meeting with the 80 briefs already ranked, so the team spends the meeting assigning work. Because Beryl marks each account that is heading toward a service-level breach, the team can fix the problem before the client escalates it.

Questions about Beryl

Put Beryl to work on the records that your team already keeps

Ask for early access to use Beryl on the contact profiles, threads, and files in your Delalify workspace.