Atrium Request a conversation

An AI operating layer for investment offices.

One governed interface that coordinates specialist agents across your firm's knowledge, secure from day one, built around how your office actually works.

Founded 2026 · Built with Lua AI

Request a conversation See the architecture
Built on the frontier models
OpenAI Anthropic Google Meta Perplexity
Market news Portfolios Stock movements FX rates CRM Emails Custodians Research AI agent AI agent AI agent AI agent AI agent AI agent INFORMATION AGENTS
The problem

Scattered tools and no system.

Investment offices are stretched thin. A lot of time goes into routine report creation, tracking follow-ups, rushed meeting prep, formatting presentations, trying to find the right data point or previous interaction across sources, and other non-judgment work.

But the market's answer is a fragmented mess of point solutions: an AI tool for the CRM, an AI tool for portfolio analysis, a non-customized tool for reporting. Individuals are playing with different AI tools themselves, but there is rarely control over how anything works together within an org or appropriate security measures in place.

AI ToolCRM AI ToolPortfolio Management Research Market Data Inbox Reporting
AI tools, old systems, scattered information.

Ungoverned by default

No permissioning

No audit trail

Atrium solves this with a governed operating layer that applies AI agents to your firm's specific needs and workflows.

What changes

Today

Reports and routine prep eat hours every week.
Knowledge sits scattered across memos, inboxes, and custodial portals.
More capacity means another hire or another tool.
Each new AI tool is one more login and thing to learn.

With Atrium

Agents draft the reports and prep. Your people lead the judgment.
One governed interface reaches across the office's knowledge.
More capacity means deploying an agent this week.
Every agent runs under the same permissions and audit trail framework.
What Atrium does

Build

Atrium builds the custom agents on your own data and workflows.

Integrate

Atrium wires into the systems you already use.

Govern

Atrium sets up the governance protocols to permission and audit agents and information.

You own the agents, the IP, the architecture. No lock-in, no vendor dependency, no starting from scratch.

Why Atrium

Secure by default. Model-flexible. Yours to own.

Secure and governed by default.

  • Every query passes through permissioning, audit logs, and approval workflows before anything is retrieved or written. Nothing is opt-in.
  • Each agent is scoped to its role. The data analyst agent cannot read tax documents; the relationship agent cannot read private investment memos.
  • Source-aware outputs: every answer shows where it came from, so the principal can trust it and the office can defend it.

Model-flexible, so it never goes stale.

  • Each task routes to the model best suited for it: reasoning, extraction, research, drafting, across OpenAI, Anthropic, Google, Meta, and Perplexity. No bet on a single lab.
  • New frontier models go live in days, on your timeline. When the next leader ships, your agents use it that week. No contract change, no migration project.
  • You're buying a capability that compounds, not a tool that ages. The deployment you run in 2 years is sharper than the one you start with.

You own it. It all stays governed.

  • You own the agents and the IP. Build them, change them, take them with you.
  • Anyone on the team can add an agent. The moment it goes live, it runs under the same privacy controls as everything else in the firm.
  • Every agent inherits the firm's permissions and audit trail at deploy time. No security review to reopen, no gaps to chase down later.
Architecture & governance

One governed conversation.

The principal should not have to know which system, folder, advisor, or agent holds the answer. Atrium is one controlled interface that coordinates specialist agents behind the scenes, and a permissioned space for individuals to build their own.

Atrium · architecture & governance
"Using the last 3 years of interaction with this fund, write me a full investment memo."
External · public information
News
WSJNYTFT
Market data
BloombergPitchBook
Regulatory
SEC EDGARForm ADV
Macro · public
FREDBLSIRS 990
The office · perimeter
Internal · your firm
CRM
Relationships
Investment
MemosIC notes
Board
DecksMinutes
Internal financials
P&LCap tableTax
PersonalOAuth
EmailCalendar
Research
CustodiansData rooms
Governance layer
Input controls
Permissioning
Source-aware
Audit logs
Select any agent to see its sample prompts
Firm agentsShared · customized
Investment Officer
Portfolio Manager
Data Analyst
Deal Sourcer
Report Creator
Relationships
Knowledge
Personal agentsOAuth-scoped
Travel prep
Meeting prep
Personal Assistant
Daily Market Update
Build your own

via HeyLua MCP · OAuth-scoped

Manager view: sees which agents ran, who owns them, what data they touched, and what governance flags fired, but not the contents, unless they have independent permission to that data.

What we won't do

Lines we don't cross.

What an AI system will not do matters as much as what it will. Four things Atrium is built not to do.

Atrium does not make investment decisions.

The agents prepare, summarize, draft, and surface. They build the case file. The decision sits with the principal, the IC, or the CIO. That line does not move.

Atrium does not move capital without human approval.

No wire, no transfer, no allocation, no fund commitment is executed without explicit human signoff. Capital movement is a hard boundary, not a default.

Atrium does not contact outsiders on its own.

Outbound emails, calendar invites, messages to counsel, GPs, bankers, family members. None of it happens unless you authorize that channel for that agent.

Atrium does not expose all office knowledge to every user.

Office knowledge is not broadcast to whoever asks. Each agent respects the same access rules as your file system, your CRM, and your inbox. Compartmented by default.

How we engage

A diagnostic, a build, then an ongoing relationship.

01

The diagnostic

3 weeks · $20,000 fixed fee

We map your workflows, your data, your existing stack, and the places your team is losing hours. You walk away with a written 12-month AI roadmap, a vendor stack audit, a security and privacy assessment, and one working prototype agent built against your real data.

If you don't continue, the diagnostic is still useful on its own.

02

The build

6 to 12 weeks

We deploy Lua AI inside your environment and integrate it with your existing platforms (Addepar, Masttro, Eton, your CRM, your inbox). We configure each agent to your firm's history, your investment thesis, your principal's voice. We train your team on how to use the agents and how to extend them.

By the end, you have agents running inside the tools your team already uses, plus one or two people on your team who can build new ones.

03

The relationship

Annual + optional retainer

Annual platform and support. Includes Lua AI platform fees, agent updates, new additions to the library, and continuous monitoring.

  • Flat fees only
  • No hourly billing
  • No add-on upsells
10hr
Our guarantee

Ten hours per week back to your team. Every week. Or you don't pay.

Within 90 days of build kickoff, your office reclaims at least 10 hours per week of professional time. We set the baseline together during the diagnostic and measure against it after deployment. If we miss the bar, we refund the build fee. The promise is structural: we sized the engagement so this is the floor, not the ceiling.

Who we are

Built by a practitioner. Powered by Lua AI.

Connor Dickson
Connor Dickson
Founder & Managing Partner

Currently Head of AI at Jackfruit Network, where he builds and deploys agentic systems across the firm. Previously Investment Officer at the University of Washington endowment.

LinkedIn
YY Combinator · Fall 2025
Lua AI
Technology Partner · The Agent OS

Every Atrium deployment runs on Lua AI's Agent OS, a production platform purpose-built for enterprise AI agents. Lua provides the underlying infrastructure, security, and multi-model orchestration so we focus on building the agents specific to your firm. Built on OpenAI, Anthropic, Google Gemini, Meta, and Perplexity, with SOC 2 and GDPR compliance built in.

heylua.ai
No lock-in SOC 2 compliant GDPR ready
Get in touch

A 45-minute conversation. No deck.

You, your principal or CIO, and us. We listen, we ask, we leave you with one written observation about what we'd attack first. If the fit is right, the diagnostic follows. If it isn't, you have a free outside read on your operation.

Request the conversation