Meet Parahelp 3: Build an AI support agent that securely uses all your tools

TL;DR:

• Intelligence is no longer the bottleneck for AI resolving support tickets - securely connecting your agent to the relevant context and tools is.

• With Parahelp 3, we've released automations that update your support agent whenever your product and policies change, and the ability for Parahelp to build reliable, scoped connections to any of your tools with an API.

Intelligence was the bottleneck

When Anker and I started Parahelp, AI models weren't intelligent enough to resolve complex support tickets, so companies mostly used AI to answer FAQs and triage other tickets to the right team members.

Today, frontier models are more than intelligent enough. But despite that, most AI agents still can’t resolve complex tickets.

The bottleneck moved to securely connecting context and tools

When an agent can't resolve a ticket today, we see that it’s almost always because one of two things is missing - and both are hard to provide securely.

Context. The companies we work with (like Replit and Cursor) move very fast, so their product and policies constantly change. These changes are rarely documented in a structured format, which makes it difficult to provide to an agent. And you can't just hand the agent everything, since internal context includes things that should never reach a customer.

Connections. Even if an agent has access to all up-to-date context, it still can’t resolve most tickets. If a customer was charged twice, it has to look up the subscription in Stripe and issue the refund before the ticket is resolved. And no support team should hand an agent broad access to Stripe, so each connection has to be scoped down to exactly what your policies allow.

We’ve built Parahelp 3 to solve those two bottlenecks.

What's new in Parahelp 3

Automations. Parahelp now lives in your tools like Slack, GitHub, Notion, and Linear, and automatically updates your Customer Agent whenever your product and policies change.

Coding capabilities. The Internal Agent - the agent that builds and improves your Customer Agent - can now write, run, and test code.

Connections to any tool with an API. With its coding capabilities, the Internal Agent can connect any tool with an API to your Customer Agent. It builds these connections as scoped tools with your policies written into the code, rather than just granting broad access. For example, the Internal Agent can build a tool that checks when a customer last renewed their subscription in Stripe and, if it’s within the 14-day refund window, refunds the charge.

Increased research capabilities. The Internal Agent can write scripts to effectively research large datasets across your context sources, such as historical tickets, Slack messages, and GitHub PRs.

How you build and run your agent

You set up once, and then your team works in a continuous loop - Build, Release, Resolve, and Identify gaps.

Set up: Connect your historical tickets and existing context. Parahelp analyzes the last 500-1,000 resolved tickets and creates targeted to-dos for the additional context and connections it needs to resolve them. Once your team completes them, your agent goes live in approval mode.

Build (incl. test): Prompt Parahelp in natural language, and it builds your agent's configurations - scoped tools, memory, and knowledge. As it builds, it runs test simulations that replay related historical tickets against the draft configurations.

Release: Changes ship as releases - reviewed, versioned, and revertable. Your team reviews the test simulations to see how your agent would have behaved on real tickets before setting a release live.

Resolve: Your agent monitors every incoming ticket in your ticketing system (Pylon, Plain, Intercom, Zendesk, or Front) and resolves the ones it can, across email, live chat, and Slack. Every drafted reply goes to Slack for one-click approval, or you iterate on it with Parahelp. Once your team is confident, you switch it to resolve tickets autonomously.

Identify gaps: Whenever your agent can't resolve a ticket, it asks Parahelp to close the gap across its broader connections, like the codebase through Cursor, recent GitHub PRs, or Slack. When Parahelp can't close a gap on its own, it creates a targeted to-do for your team - and if a to-do already exists for the gap, it just updates it, so closing one gap often resolves many tickets at once.

Who runs on Parahelp

Parahelp agents have already resolved millions of tickets for fast-growing tech companies like Cursor, Replit, Lovable, and Supabase.

Try Parahelp

The fastest way to understand Parahelp 3 is to try it out. Book a demo, and we'll help you get started.