Custom AI agents for innovation teams. A small team of specialists, each with one trade.

Next to the CEO and the analyst, add the specialists your portfolio needs: a pricing agent, a claims expert, a reviewer for your sector. You say what each one does, when it answers and which projects it reaches. FounderFold keeps the team small and warns you before two agents overlap.

Illustrative data — Fictional company, ideas and figures, shown for illustration only.

New agent editor filled in for “Pricing”: one-line description, role, the keywords price*, plans and discount, answers by default in projects, scope limited to P-0107 and P-0112, and the colour of its robot, with a preview of the robot.

An agent in a minute: what it does, when it answers and which projects it reaches. Illustrative data.

What is a custom agent in FounderFold?

A custom agent in FounderFold is a specialist you create for your portfolio, with a name, a one-line description, its role, keywords, the areas where it answers by default, a scope of projects and the colour of its robot. It always works under the base rules shared by every agent — sources, citations and caps — and every change is saved as a version you can go back to.

Create one by hand in a minute

In the agent room, “+ New agent” opens the editor. Fill in the name, “What it does, in one line” and “Its role: what it does and how it answers”, add keywords and choose where it answers by default: documents, projects, experiments, ideas or the portfolio.

  • Keywords decide when FounderFold picks it: with “price*, plans, discount”, a question about prices goes to it. A trailing asterisk covers every word that starts that way.
  • “Which projects it covers”: the whole portfolio or only the projects you tick, so it never slips into questions about the others.
  • Pick the colour of its robot, and it appears in the agent room.

An agent is a trade, not a task

FounderFold keeps the team bounded: seven base agents and up to five of your own. An agent covers a trade — pricing, claims, regulation — and lasts as long as the portfolio; a goal such as “first B2B customers” belongs on the board, in steps or experiments, and the editor says so.

  • The overlap check warns you, strongly or partially, when a new agent shares keywords or areas with another: “B2B Expert” gets an amber warning because it shares “price” with Finance.
  • Adjust instead of duplicating: if an agent almost fits, add your own instructions to it.

Illustrative data — Fictional company, ideas and figures, shown for illustration only.

Two warnings from the overlap check: “B2B Expert” with an amber warning for sharing the keyword “price*” with Pricing, and “First B2B customers” rejected because it looks like a task, not a role.

The team stays small: overlaps and tasks disguised as agents are flagged before saving. Illustrative data.

Every change is a version

Each change is saved with “Save version” and a note in “What you changed”. The “Versions” tab lists them with their origin — from the system, proposed by the CEO or created by a person — and “Use this version” brings an earlier one back.

  • Agents are not deleted: they are archived and can be recovered from “Archived”.
  • “Back to the base configuration” resets a system agent to how it came.

Illustrative data — Fictional company, ideas and figures, shown for illustration only.

Versions tab of the Pricing agent with three versions, each with its note on what changed and the button “Use this version”, and below, the archived agents with “Restore”.

Every change is a version; agents are archived, never deleted. Illustrative data.

Tune the agents that come with FounderFold

Each system agent can be adjusted for your organization: your own instructions — sector, tone, rules it must always follow — extra keywords, the areas where it answers by default, and whether it is active. The CEO and the analyst are always active.

  • Your instructions add to the agent’s role; they do not replace it.
  • A deactivated agent appears faded in the room, and FounderFold does not pick it.
  • From an agent’s card in the room, “Add a critic” gives it a reviewer that checks every answer with a second AI call; critics count towards your five agents.

A pricing agent for Halden Mutual

A fictional organization with fictional agents.

  1. 1The need. Two projects keep asking about premiums and broker discounts, and Finance answers in general terms. Ines creates “Pricing” with the keywords “price*, plans, discount”.
  2. 2The scope. She limits it to Working-dog insurance and Landlord rent guarantee, the two lines with a price on the table, and gives its robot an orange colour.
  3. 3The warning. A colleague tries to add “First B2B customers”. The editor flags it as a task, not a trade, and suggests putting it on the board as steps or an experiment.
  4. 4The version. A month later she adds rules on broker commissions and saves version 2 with the note “Added discount rules for brokers”. Version 1 is still one click away.

FAQ

How many agents can a team have?

Seven base agents and up to five of your own. The limit keeps each agent on one trade and makes it clear who answers what.

Who can create or change agents?

Owners and editors. Readers see the team in the agent room but cannot change it.

Can a custom agent change my data?

No. Like every FounderFold agent, it reads, explains and proposes under the same base rules on sources, citations and caps. Every change to a record goes through your confirmation.

Can I delete an agent?

Agents are archived, not deleted, and every change stays as a version. You can recover an archived agent from “Archived” in the agent room.

Build the team your portfolio needs.

Early access is by invitation.