Skip to content

Product

Three planes, one platform.

WeBelieve is built around an idea most ops software has skipped: if you own the full model of how a firm works, every other capability gets dramatically cheaper to build well — including the AI.

PLANE 01

One model of the firm's work.

Clients, engagements, projects, tasks, documents, calendar, email, notes, time, billing, intake — same data plane. No integration layer pretending five tools are one.

The schema isn't generic. Each vertical pack extends it with industry-correct types: a CPA engagement has filings; a construction project has submittals and change orders; a real-estate transaction has contingencies and disclosures.

// Core entities (excerpt)

Client → Engagement → Tasks ↓
                       Documents ↓
                       Deadlines ↓
                       TimeEntries ↓
                       Invoices

Communication: Email, Note, CalendarEvent
Intake: Lead, Application, Disclosure
People: User, Role, Permission

// Vertical pack extension: CPA

Engagement.kind: 'tax_prep' | 'audit' | 'advisory' | ...
Engagement.filings: Filing[]
Filing.jurisdiction, .form_type, .due_date, .extension

RULES (excerpt — CPA pack)

  • Filing deadline approaches

    If a Filing's due_date is within configured threshold and required documents aren't received → surface to lead partner.

  • Vendor reconciliation stalls

    If a reconciliation needs client docs and no response for 5+ business days → surface to engagement owner.

  • RFP fits win pattern

    If an inbound RFP matches features of past won engagements (size, industry, scope) → surface to BD lead.

PLANE 02

Ambient awareness on top of that model.

Background workers continuously evaluate the model. When something deserves a human, it shows up on the morning brief — ranked, grounded, with one-click drill-through to the underlying engagement, task, or thread.

The signals aren't reactive notifications. They're proactive — generated by rules in your vertical pack, tuned against your firm's history.

PLANE 03

An AI runtime with full context.

The model is the system prompt. Ask what's overdue across every engagement and get a real answer, grounded in your data, with provenance you can click through.

The runtime calls typed planners that traverse the model, executors that act on it, and validators that keep it honest. Every claim cites the engagement, task, or document it's drawn from.

Watch it think →

RUNTIME PHASES

  1. 01

    Planner

    Reads the question, asks: what entities does this question touch? Builds a typed traversal plan.

  2. 02

    Executor

    Walks the plan over your data. Pulls only the relevant rows. Cites every one of them.

  3. 03

    Validator

    Checks the answer against the cited evidence. Refuses to claim what the data doesn't support.

Story

Why three planes instead of one big AI.

Plenty of teams are bolting AI onto existing SaaS. It works for narrow tasks. It falls apart when the question crosses tools, because the model never sees the full picture — only what the integration layer remembered to send.

The shortcut to a useful AI assistant isn't a smarter model. It's a unified model of the firm's actual work, with the assistant living inside it.

That's WeBelieve. The data plane is the moat. The ambient layer is the daily product. The runtime is the part you talk to.