Beamreach AI Logo Beamreach AI

How it works

What happens between a client emailing you and you pressing send

Five stages. A message arrives on whatever channel the client chose, one agent decides what it is, the right specialists run, they write out exactly what they intend to do, and none of it happens until a person says so.

1

Something arrives

An email, a WhatsApp message, a web form. Or a scheduled check, or a case that just changed status.

2

Triage reads it

One agent works out what this is and which of your specialists should see it — or that nobody should.

3

The right agents run

Yura checks the decision against what you have actually switched on, then starts those agents in the correct order.

4

They write a plan

Every intended action, spelled out: this reply, this file in this folder, this row updated. Nothing done yet.

5

You approve

You read it, edit it, approve or reject. Only then does anything leave the building — and all of it is logged.

Stages 1 to 4 take seconds and run at 3am. Stage 5 waits for you.

1

Something arrives

Clients do not agree on a channel. One emails, one sends a WhatsApp at eleven at night, one fills in the form on your site. All three land in the same place, converted into one common shape before anything reads them.

Not everything is a message. A weekly schedule, a matter changing to retained, or one workflow finishing can all start work the same way. See which channels are live today.

2

Triage reads it

A real inbox is not a queue of tidy client enquiries. It is client enquiries mixed with spam, PR pitches, recruiters, invoices, a question about parking, and someone who emailed the wrong firm. No list of rules survives that, which is why the first thing to read a message is a model rather than a filter.

Triage is given the message and its thread, whether the sender matches a case you already act on, what is attached, your firm's practice profile, and the catalogue of agents you have switched on — each with a description of what it is for. It picks from that catalogue and nothing else.

It gives a reason for every choice, in a sentence a lawyer can disagree with. Those reasons are kept.

3

The right agents run

Triage does not start anything. It hands back a decision, and Yura checks every named agent against the ones you have actually enabled before dispatching a single run.

That separation is deliberate. An agent that could start other agents on its own say-so would be a hole in the audit trail — the chain would exist only inside a model's output, and nothing afterwards could be attributed to anything. Order is fixed too: document checks always run before the intake reply, because the reply needs to mention what the checks found.

4

They write a plan

Each agent reads what it needs — the matter, the checklist, the documents themselves — and produces a list of intended actions. Draft this reply, with this text. File this passport into this client's folder under this name. Update this row to this value.

This is where the work is done and where it stops. Nothing has been sent, filed, or written to your systems yet.

5

You approve

The plan arrives in one place with everything it intends to do. You read it, change the wording, approve the parts you want and reject the rest. Only approved actions execute, and each one is recorded — what ran, on whose authority, against which system, with what result.

Approval is per action type and it is yours to set. Filing a document into a folder is low-stakes and most firms let it run on its own within a few weeks. Sending an email to a client is not, and most never do. Corrections you make to a draft are kept and fed back, so the next one starts closer to your voice.

What happens when triage is not sure

Every message ends in exactly one of three outcomes. The third one is the one that matters, and most systems do not have it.

Route it

One or more agents should handle this, each with a reason. Work starts.

Deliberately do nothing

Marketing, a newsletter, a cold pitch, an automated notification, something plainly meant for someone else. Recorded with a category and a reason. Nothing is lost — the message is still sitting in your own inbox, because Yura reads alongside you rather than in front of you.

Admit it does not know

Anything ambiguous, anything that looks like a client in trouble, anything where guessing wrong would matter. It goes in front of a person with the message attached. Triage is told to choose this whenever it is unsure — being uncertain is never a reason to decide nothing needs doing.

A model error, a timeout, or a malformed answer also lands here. A message that could not be decided on must never look like a message that needed nothing — that is the one failure an intake system cannot have.

One message, all the way through

21:40 · Email arrives

A client sends four photographs — a passport, a birth certificate, a tax registration, a rental contract — with the message "here are the documents you asked for".

21:40 · Triage

Sender matches an open matter. Attachments present. Routes to two agents: document checks, because there are documents worth opening, and intake, because a client is waiting for an answer.

21:41 · Document checks

Opens all four and reads inside them. The passport expires in four months. The birth certificate is in Ukrainian with no certified translation. The surname on the tax registration does not match the passport.

21:41 · Intake writes the plan

Four files into the client's folder, named to your convention. The matter row updated with what arrived. A reply drafted that thanks the client, confirms what was received, and asks about the three problems.

08:15 · You

You open one screen. The findings are there with the draft. You soften a sentence about the surname, approve. The reply sends in the original thread, the files move, the row updates. Every step is in the log.

The client was answered before nine. Nobody worked at night. Nothing was sent that a lawyer had not read.

Three decisions behind this

The router cannot start work

Triage proposes, Yura dispatches, and only against agents you enabled. The most powerful agent in the system is the one with the least authority.

A message is evidence, never an instruction

Anyone can email your firm, which makes the inbox an attack surface. Text telling an agent to ignore its instructions is treated as a fact about the sender — usually a reason for more suspicion, not less.

One message, one traceable job

Everything a message caused hangs together: the decision, its reasons, every run, every proposal, every approval. You can answer "why did this happen" months later.

See it run against your own inbox

The fastest way to understand this is to watch it triage a week of your real messages, with every action still needing your approval.