Use case

One document. A panel of reviewers.

Contracts, proposals and filings get reviewed in parallel by specialized agents: technical, business, compliance. The findings merge into one report.

Deploy LM-Kit One Build with LM-Kit.NET
Parallel reviewers One lens per agent Merged findings Runs fully local

Expert review does not scale. Skipping it costs more.

Every contract and filing deserves technical, commercial and compliance eyes. What it gets is whoever had time, whenever that was.

  • The queue at legal. High-stakes documents wait days for a reviewer whose calendar is the real SLA.
  • Single-pass misses. One reader, one lens: the cross-cutting issues fall between specialties.
  • Inconsistent depth. The same document gets a different review on a busy week.
  • Unsupported verdicts. "Looks fine" carries no evidence anyone can check later.

Three lenses, one report.

The panel is yours to define; the mechanics are the platform's to run.

01

Assign

Each agent gets one lens: technical, business, compliance, yours to define.

02

Review

Agents read the same document in parallel, grounded in its text.

03

Merge

Findings collect into one report with each agent's evidence.

04

Decide

A person reads one report instead of chairing three reviews.

Why a panel beats one broad prompt.

The same reason human review works that way: focus finds what breadth skims over.

Lenses

Specialists beat a generalist

A focused agent with one concern catches what a single broad prompt misses; you define the panel.

Parallel

Wall-clock of one review

Reviewers run concurrently on local hardware, so a three-lens panel costs the time of the slowest lens.

Evidence

Findings carry their source

Each finding cites the passage that triggered it, so disagreement is settled by the document.

Define the panel, run it in parallel.

Each reviewer is an agent with a persona; the orchestrator runs them together and collects the results.

ReviewPanel.cs 
using LMKit.Agents;
using LMKit.Model;

var model = LM.LoadFromModelID("qwen3.5:9b");

var technical = Agent.CreateBuilder(model)
    .WithPersona("Review for technical gaps and undefined terms.")
    .Build();
var business = Agent.CreateBuilder(model)
    .WithPersona("Review commercial terms against our policies.")
    .Build();
var compliance = Agent.CreateBuilder(model)
    .WithPersona("Review for regulatory and residency issues.")
    .Build();

var orchestrator = new ParallelOrchestrator()
    .AddAgent("Technical", technical)
    .AddAgent("Business", business)
    .AddAgent("Compliance", compliance);

var results = await orchestrator.ExecuteAsync(reviewPrompt);

The technology underneath.

The panel is orchestration over documented building blocks, each usable on its own.

Panel

Multi-agent workflows

Parallel, pipeline and supervisor orchestrators for agents that work together.

How orchestration works

Routing

Delegation

Supervisors that hand work to the right specialist and collect the results.

How delegation works

Shape

Graph orchestration

Review flows beyond a straight line: branches, joins and conditions.

How graphs work

Evidence

Observability

Every agent's reads, findings and reasoning on the record, so the merged report can cite its sources.

How observability works

Frequently asked questions.

Why several agents instead of one strong prompt?

Because attention is a budget. A reviewer told to check everything checks nothing thoroughly; three specialists each reading with one mandate, technical, commercial, regulatory, surface findings a generalist pass smooths over.

Do the reviewers run at the same time?

Yes. The parallel orchestrator fans the document out to every specialist at once and collects their findings, so panel review takes one specialist's time, not the sum of all of them. On your hardware, the extra passes cost no per-token fees.

How do the findings become one report?

A merge step collects each agent's findings with their evidence, and observability keeps every read and conclusion on the record, so the combined report cites which reviewer found what, where.

Does privileged content stay inside?

Yes. Every reviewer runs on your infrastructure, so contracts and filings never reach a third party. Embed the panel with LM-Kit.NET, or serve it from LM-Kit One when several teams share it.

Private Document Intelligence

Put a panel on your next contract.

Deploy LM-Kit One Multi-agent workflows in depth