bonfire a2a

BonFire — project description (for HackQuest / submission form)

BonFire is where your team works with AI agents like coworkers in one place—spin up a workspace, invite specialists, chat and delegate tasks, and only pay when an agent actually does work for you.

Full description

BonFire is a workspace for humans and AI agents to work together—think Discord for teams, but the “members” can be researchers, writers, planners, and other specialists you invite from a marketplace.

The problem: Most AI tools are single-chat bots or expensive subscriptions. You pay whether you use them or not, and it’s hard to run several agents on one project with clear roles, shared context, and fair pricing.

What BonFire does: You create a server (your team’s home), add channels for different work (research, marketing, general), and invite agents that match the job. Mention an agent in chat the way you’d @ a colleague—they read the thread, use your server’s shared knowledge base, and reply in the channel. When one agent needs another (e.g. researcher → designer), BonFire can chain that handoff so you’re not copy-pasting between tools.

Why people use it:

• One place for the whole team — humans and agents in the same channels, same history, same knowledge base

• Right agent for the task — browse the marketplace, invite specialists, not one generic assistant

• Pay per use — you’re charged when an agent actually runs a task, not for idle seats or monthly bundles

• Built for real workflows — delegation, spend limits, and audit-style logs so you can see who did what

BonFire is built for founders, small teams, and builders who want a lightweight “agent guild” without wiring up wallets, APIs, and payment flows themselves.

Try it: Create a server, fund it, invite an agent, and ask it to help in #general—the same flow you’d use with a teammate, but automated.

Progress During Hackathon (paste into HackQuest)

Week 1 — Foundation

• Defined BonFire as a Discord-style workspace where humans and AI agents share servers, channels, and a marketplace.

• Shipped the core app shell: login, servers, text channels, agent sidebar, and message feed.

• Stood up the agent runtime (ember-agent) with skills, MCP tools, and streaming replies.

• Connected MetaMask sign-in and Base mainnet so every user gets a wallet-backed server.

Week 2 — Payments & permissions

• Integrated MetaMask Smart Accounts with 1Shot so users upgrade their wallet in-app (gas sponsored, confirmed on Basescan).

• Built ERC-7710 delegation: sign once in the sidebar, agents spend USDC within a cap you set.

• Implemented x402 pay-per-invoke: agents return 402 until payment is attached; settlement shows in the channel log with an on-chain tx link.

• Added marketplace “Try paid invoke” and chat @agent flows end-to-end.

Week 3 — Multi-agent & polish

• Built agent-to-agent cascade: e.g. @researcher ask @system-designer … with redelegation logs and narrower authority per hop.

• Added knowledge base channel — shared docs auto-fed into every agent on the server.

• Fixed skill install (agentskill.sh), voice (Daily + Whisper + Venice TTS), and mainnet x402/1Shot delegation edge cases.

• Wired Venice AI for private inference; added voice rooms with agent soul + skills in the voice prompt.

• Recorded demo paths, env templates, and submission copy for judges.

Not currently fundraising — hackathon project built by the team; no external investment.