Hey everyone! We’re JJ and Mark, cofounders of Pillar 👋

TL;DR: Pillar is an embedded AI copilot that executes tasks inside your product for users and agents.

Users or agents can type what they want. The copilot carries it out client-side in the browser, using the user’s existing session, permissions, and security checks. You install the SDK via npm and register your existing frontend code as tools.

Example: in a banking app, a user types “send $200 to my cleaners.” Pillar finds the right recipient, navigates to the transfer flow, and pre-fills the form. The user still reviews and confirms. If your app requires 2FA for that acion, so does the copilot.

Demo: https://www.youtube.com/watch?v=ruNJ5OFuKsI

The Problem

Teams ship faster than ever. But when your changelog is 10x longer than it used to be, users can’t keep up.

• Discoverability breaks down: the UI changes, features move, and people forget where things live.
• Friction turns into “support”: users open tickets that aren’t really support tickets. The user wanted to do something your product already supported but couldn’t get there.
• Most products don’t have an “action layer”: AI agents are starting to use web apps the same way users do: navigate, click, fill forms. Products need to prepare for this

Bottom line: products need a reliable, permissioned way to execute actions end-to-end - not just document them.

How Pillar Works

Pillar turns a user request into action by combining planning with in-browser execution.

1. Plan: Pillar determines which steps/tools to run
2. Execute: your app runs those tools in the browser (navigation, API calls, state updates, form fills)
3. Confirm: the user stays in control for sensitive actions.

You register tools inside your existing React components:

What happens at runtime when a user types “send $200 to my cleaners”:

• Pillar calls search_recipients with “cleaners”
• Selects the right match
• Calls prefill_transfer with the recipient + amount
• User reviews and confirms in the existing flow

Because execution happens in the browser, tools run with the current user’s permissions and session. Pillar can’t do anything the user can’t do.

Pillar syncs with your help content (Zendesk, Intercom, Notion, Confluence, internal docs) so requests map to the right tools/flows. When it picks the wrong path, you flag it and the correction is captured so the same issue is less likely to repeat.

SDKs are available for React and vanilla JavaScript. Registered tools can also be reused by other copilots/agents via WebMCP (navigator.modelContext).

Try It

Install: trypillar.com

We also have live demos you can try right now - Pillar installed on open-source products:

• Grafana (trypillar.com/demos/grafana): you can have Pillar build you a monitoring dashboard and set up alerts
• Apache Superset (trypillar.com/demos/superset): you can explore the names dataset and build a dashboard

👉 Want help getting live? Reach out at founders@trypillar.com

Our Ask

• If you lead product/engineering and users open tickets for things your product already does, we’d love to talk: founders@trypillar.com
• If you know someone who fits that (founders, VP Product, Head of Support), intros are appreciated
• Share this post. Follow us on LinkedIn and X.

Backstory

We built Pillar to solve our own problem. At our last company (Double Finance), we saw a pattern: users asked for outcomes we already supported, but still opened support tickets. We wanted a way to reuse the frontend code we’d already shipped, without rebuilding flows or adding new “automation” surfaces.

Pillar is what we wish we had.