AI agent platform

Your company's second stack.

One place to create, run and control your company's AI agents. Your people have managers, budgets and rules, and SecondStack gives your agents the same.

Book a demo See an agent at work
  • Self‑hosted today. On your servers, so your data stays with you.
  • Your own database. Every chat, file and agent run is kept there.
  • Personal data and secrets masked. By the rules you set, before anything leaves.
  • ISO 27001 via T1A, our implementation partner.

Your agents are already arriving, one at a time.

They prepare the meeting brief, write the Monday report and answer the team's questions. One team pays for a chat subscription, another has an agent inside its CRM, and someone expenses a tool on a personal card. Each one is useful, and nobody holds the list, the budget or the rules.

Three questions we hear on every first call, and the answers with SecondStack:

How many agents are running?
All in one place, each with a name and an owner.
What are they costing?
Each runs on its sponsor's budget (the person or team that funds it), or on the budget of whoever calls it.
Who can pause one?
Its owner or an admin can pause it, any time.

See an agent at work.

Three everyday jobs, one per kind of agent. Each is something you could ask a colleague to do; pick one and follow it from request to result.

Maria, head of sales, writes to Aster, Maria's personal agent, in a Slack direct message.

  1. Maria writes to Aster in Slack: "Prepare me for the Acme meeting."
  2. Aster picks it up. It works for Maria, and only for Maria.
  3. Aster reads the invite in Maria's calendar and the last email thread with Acme.
  4. Your company's rules apply: an approved AI model, personal data masked by the rules you set, the cost counted.
  5. Aster posts a short brief back in Slack: who is coming, what was agreed, what is open.
  6. The company keeps a record of the run, not Maria's chat history.

Example runs. Names and details are illustrative, and the steps are simplified.

Three kinds of agent.

Every person gets one, and every shared workspace gets one, from the first sign‑in. The third kind you hire for a role, like any specialist.

Personal agent

Everyone gets one, called Aster. It works with the accounts that person connects, like Microsoft 365 or Google Workspace, and answers in the web app or in Slack.

Project agent

One per shared workspace, a Project in SecondStack. It answers the whole team from the team's own threads and files, never from anyone's personal accounts.

Named agents

You hire one for a role: a sales analyst, a legal adviser, a technical writer. It has a name, an owner and, if you want, a schedule.

Same rules for every agent.

Every agent starts the way a new colleague does: someone answers for it, and it spends only a budget someone holds. Both are in place before its first run.

See every agent in ControlTower 

An owner

A person or team who answers for it.

A budget

Every run charges a budget someone holds.

Access

Only the accounts and files it was given.

A record

What started each run, what it used, what it cost.

A pause button

Pause it the moment something looks wrong.

One door to every AI model.

Every AI request in SecondStack, from a person, an agent or another app, leaves through one door, where three checks apply.

Where your data goes. Requests go to your own accounts at the AI providers, or to AI you host yourself.

How SecondGate checks requests 

One AI agent platform, four points of view.

Pick the one closest to your work: business teams, IT and security, finance or platform engineering.

Business teamsThe Monday report, written before you log in.Read IT & securityLet people use agents without losing track of what they do.Read FinancePay for what people use, not for every seat.Read Platform engineeringAgents for the whole company, controls for your team.Read

Pay for what agents use, not for who might use them.

Pricing is by consumption, not per seat. Model usage stays on your own provider accounts, and budgets keep it predictable.

Runs where you decide. On your own cloud or servers today, installed with our help. Managed SaaS and AWS Marketplace are coming; ask us when.

Talk to us about pricing

Questions we get on the first call.

What is SecondStack?
SecondStack is an AI agent platform from Dark Lake, LLC. Every person gets a personal agent, Aster; every Project a project agent; teams hire named agents for a role. ControlTower gives each an owner, a budget, your rules and a record of every run.
What do you mean by "second stack"?
Your first stack is your people: managers, budgets, an org chart. Your second stack is your AI agents, and SecondStack gives it an owner, a budget, your rules and a record of every run.
Do we need to build anything to start?
No. Everyone gets a personal agent, Aster, at sign‑in, and every Project gets a project agent. A named agent is set up by its owner in the web app, the way you would brief a new hire, then tested and published.
How you add a named agent
Where does an agent get its information?
From the accounts you connect, such as Microsoft 365, Google Workspace or GitHub, and from the files and threads of the Project it works in. A personal agent uses only the accounts its owner connected. The sign‑in details stay in a separate service that adds them to each call, so the agent never holds them. People reach their personal agent in the web app or in Slack.
Does SecondStack record what our people write?
Not their chats. It records what each agent run did: what started it, which accounts and files it used, where the result went and what it cost, with passwords and keys stripped out; administrators can read that record, not change it. People's own chats stay in your own database, and a deleted conversation leaves no copy in the record.
See what every run leaves on the record
Which AI models does it work with?
The providers where you already have accounts, including Anthropic, OpenAI and Google, and models you host yourself; which models a team or an agent may use is a list you set. We do not host or resell models, and because every request goes through one door, moving to another provider is a change of settings.
Already paying per seat for an assistant?
How is SecondStack priced?
By consumption, not per seat. Model usage is billed by your providers on your own accounts, and budgets by organization, team, person and agent keep the total predictable; ask us for the current terms.

Your first stack has managers, budgets and an org chart. Give your second stack the same.

Bring one job you would hand to an agent. On the demo call we walk through how it would run in SecondStack: who owns it, what budget it runs on, and the record it leaves.

Book a demo
Aster, the personal agent, walks over and presses Book a demo.