# Billy, the RFP Agent

- **Event:** [The Agent Arena Hackathon](https://cerebralvalley.ai/e/vultr-the-agent-arena)
- **When:** Sep 26 at 9:00 AM – Sep 27 at 5:00 PM (PDT)
- **Where:** San Francisco, CA
- **Team:** [Yves Hughes](https://cerebralvalley.ai/u/yvesjr)
- **GitHub:** https://github.com/yveshughes/rfp-agent
- **Demo video:** https://us06web.zoom.us/clips/share/sTS61Qc8Ti2KMDl-t4yvMg
- **Gallery:** https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/7

Small businesses win government work by answering RFPs, and most never try: finding the right solicitation, reading a fifty-page packet, and writing a compliant response is a full-time job. RFP Agent gives that job to Billy.

Billy runs on a Vultr VM that is his computer and his memory: a headless browser, SQLite, every original document, every draft version and every review PDF live there and keep working while your laptop is closed. Every decision he makes is GLM 5.3 on Vultr Serverless Inference, chosen one bounded action at a time from a fixed set of tools. He cannot submit, email, sign or buy. Each fact he saves about your company quotes its source page, and the server rejects a citation that isn't there. Requirements he can't support stay labeled as gaps instead of becoming invented answers.

You start with what you already have: a past proposal, your website, a few files. Billy builds your profile with sources. Continuous Autopilot then prepares suitable RFPs one after another, each appearing in your pipeline as Researching, Drafting, then Ready for review. You open the response as one editable document, check the flagged gaps and deadline, and follow the agency's own submission link. Human review is the product, not a disclaimer.

You can talk to Billy. A call runs through Gemini Live for listening and speaking, but every question about your work is answered by the same agent on Vultr and saved in your chat. One five-dollar instance runs several companies with separate data.

Built this weekend: the agent loop, evidence provenance, discovery and shared catalog, the Autopilot queue, the response canvas, review PDFs, voice, and the deployment. The agency directory and sample proposals are inputs, not hackathon work.

## More from The Agent Arena Hackathon

- [GhostRange](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/4)
- [Market Watch Commander](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/5)
- [Replay](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/6)
- [Team JC](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/8)
- [Upgrade Chamber](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/9)
- [awesome](https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/10)

---

Markdown version of https://cerebralvalley.ai/e/vultr-the-agent-arena/hackathon/gallery/7. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt