Pactix

The Agentic Negotiator for Freelancers

Comment

Inspiration

I noticed that freelancers leave a ton of money on the table simply because negotiating is awkward. We tend to under-price ourselves and reply to client emails at midnight out of panic. I wanted to fix this by building a 24/7 "deal desk" for independent creators, giving them the same negotiation leverage that big corporate agencies have.

What it does

Pactix is basically your personal AI negotiation team. When a client emails you, it triggers a "Council of Six" AI agents. Instead of just using one prompt, these agents actually debate each other, one researches the client's budget, one flags scope risks, and two argue over the best price. Once they reach a verdict, Pactix drafts a reply that sounds exactly like your writing style. If the client agrees, it automatically generates a PDF contract and sends a Stripe payment link to close the deal.

How I built it

I built the full stack using Next.js 16 and React 19. For the UI, I created a custom design system with Tailwind CSS and used Framer Motion for the interactive animations. The backend uses Next.js API Routes and Server-Sent Events (SSE) to stream the agents' thoughts live to the screen. For the database, I went with Supabase and Prisma. The core AI engine is powered by Google Gemini 2.0 Flash, which is fast enough to run all six agents in just a few seconds.

Challenges I ran into

Getting six different AI agents to talk to each other without hallucinating or taking way too long was really tough. I had to design a strict pipeline where the "Judge" agent specifically waits for the "Prosecutor" and "Defense" to finish their debate before deciding on a final price. Also, making the AI sound like a real human rather than a robot took a lot of prompt engineering and style-transfer work.

Accomplishments that I'm proud of

I'm really proud of the "Council Live" UI on the landing page. It's a fully draggable, physics-based graph that makes the AI deliberation feel tangible and interactive. I'm also super happy with the "Deal Replay" feature, which lets you scrub through the agents' thought process just like a video player.

What I learned

The biggest takeaway for me was that multi-agent systems completely outperform standard chatbots. Breaking a complex task (like pricing strategy + copywriting) into specialized, debating agents yields much better results. I also learned a ton about streaming real-time data with SSE and building premium, high-trust UIs.

What's next for Pactix

Next, I want to integrate Pactix directly into Gmail so it can drop negotiated replies straight into your drafts folder. I'm also planning to build an autonomous redlining feature, where a client can upload their own contract and the AI will automatically strike out predatory clauses.

Built With

Share this project:

Updates