Agents
The routine runs. You keep the decision.
Build an agent by describing the job, or install one from the Marketplace. It runs on its schedule, reports what it did, and waits for a named lawyer wherever your firm's rules say it must.
can build one: the person who owns the work describes the job, with no developer and no development cycle
agents run on the schedule you set, on your own servers
run is written to the activity log, with what was done, who approved it and when
Agents run on the schedule your organisation sets and stop where its rules say.
31 new documents arrived across 12 active matters this week. Answer periods run out within ten days on three of them, and two have no one assigned to the draft yet.
- Read new documents in active matters
- Write a short summary per matter
- Match upcoming deadlines to the people responsible
- Send the summary for a lawyer's approval
Described in a sentence, built by the person who owns the work
Anyone who can describe the job can create an agent. It is built from a sentence or from a department template, not from a development cycle.
An agent reaches only the systems and data the person who built it is already cleared for.
- Runs on its ownSummaries, classification, internal reminders, task preparation
- Runs once you enable itNarrow notifications with a set recipient and content
- Waits for a lawyerPetitions, appeals, client messages, material deadlines, anything filed or sent out
Where an agent stops
Your organisation decides what an agent may do alone. Low-risk, clearly defined actions run directly.
Anything sensitive or outbound, such as a reply to a client, is prepared and left for an authorised person to send.
Agents a firm actually runs
Weekly matter summary
Reads the week's new documents across open matters and writes one summary per matter.
Deadline candidates
Proposes dates with the trigger and rule behind them, for a lawyer to confirm.
Document intake
Files what arrives to the right matter and flags anything that reads like a demand.
Contract triage
Sorts the review queue and surfaces the clauses that need a person first.
Renewal watch
Surfaces notice periods and renewal dates before they bite, with the clause that sets them.
Client update drafts
Prepares the update in the firm's format and waits for the lawyer whose name goes on it.
One run, start to finish
A Monday morning summary of everything that moved in the firm's open matters.
- 01
You describe the job
In a sentence. Sage fills in what it should read, what it may do, when it runs and when it must ask.
- 02
You set what it may do alone
Allowed from the start, allowed once you enable it, or never on its own. The firm draws that line.
- 03
It runs on its schedule
Overnight, weekly, or on an event. It keeps working when nobody is watching.
- 04
It stops where your rules say
Petitions, client messages and anything filed wait for a named lawyer. The work is prepared; the decision is not taken.
- 05
It reports what it did
What ran, what it read, what it produced and who approved it, recorded as it happens.
Running more than one
How agents are kept in order once a department has several of them.
An owner
Every agent belongs to the person whose work it does, and reports back to them.
A scope
It reaches only the systems and data that person is already cleared for, set once by an administrator.
A register
Every agent is listed with its owner and its scope, so two never act on the same record.
The questions people ask
How agents are built, what they may touch, and where they stop. If yours is not here, ask us directly.
Who can build one?
The person who owns the work. An agent is built by describing the job in a sentence, or installed ready-made from the Marketplace, and neither route needs a developer.
What can an agent do on its own?
Whatever your firm allows and nothing more. Actions sit in three groups: allowed from the start, allowed once you enable them, and never alone. Petitions, client messages and filings are always in the last group.
What is recorded?
Every action an agent takes: what ran, what it read, what it produced and who approved it. The record is designed to be added to rather than edited. It is not independently verified as tamper-evident, and we do not describe it that way.
What if it gets something wrong?
The lawyer who approves the work is responsible for it. Sage researches and prepares; it does not decide. An agent that is unsure notes the gap rather than guessing.
Can we try one before it touches anything?
Yes. An agent can run as a rehearsal, producing its output without performing any outbound action, so the firm sees what it would have done first.
A two-day demo, built around your needs
We agree your priorities together and, over two days, show how Sage works on an example workflow we choose with you.
