Skip to main content
Colosseum Copilot is a research skill that turns your AI coding assistant into a Solana startup analyst. It gives your agent access to:
  • 5,400+ hackathon project submissions with tech stack, problem/solution tags, and competitive context
  • 84,000+ archive documents across 65+ curated sources: cypherpunk literature, protocol docs, investor research, founder essays
  • Hackathon analytics across every Colosseum hackathon (Renaissance, Radar, Breakout, Cypherpunk)
  • 6,300+ products across crypto from The Grid, ecosystem metadata
  • Web search for current competitive landscape

Ecosystem data by The Grid

6,300+ products across crypto — powered by The Grid

​
What makes it different

Honest about competition. If another team is already building the exact same thing, Copilot tells you immediately — with the project name, hackathon, and what they’ve shipped. Gap classification. When incumbents exist, Copilot classifies the gap:
  • Full gap: nobody has addressed this problem
  • Partial gap: incomplete coverage (segment, UX, geographic, pricing, or integration)
  • False gap: already solved
Evidence floors. Every answer must meet minimum evidence requirements. A market assessment needs builder project data, at least one archive citation, and current landscape evidence. If Copilot can’t meet the floor, it says so rather than filling the gap with speculation. Curated archive dataset. The archive draws from 65+ hand-selected sources: the Cryptography Mailing List (1990s), Satoshi’s emails, Nick Szabo’s essays, Solana Breakpoint transcripts (2022–2025), and research from Paradigm, a16z, Multicoin, Pantera, Galaxy, and more.

​
See it in action

Ask Copilot: “I work in logistics at a Fortune 500. We spend millions on cross-border payment fees. Is anyone doing stablecoin-based trade finance on Solana?” Copilot immediately surfaces CargoBill (Colosseum Accelerator C3, 1st Place Stablecoins) as the closest predecessor, classifies the gap as Partial — Segment (CargoBill targets freight forwarders, not Fortune 500 buyers), identifies funded competitors (OpenFX at 23M,Finat23M, Fin at 17M), and recommends a specific wedge: an ERP-integrated stablecoin treasury module targeting the corporate AP side. Every claim traces to a named project, archive source, or data point. See the full unedited output and five more examples.

​
How it works

Copilot operates in two modes:
  1. Conversational (default): Answer questions with targeted API calls and inline citations. Fast, concise, evidence-backed.
  2. Deep Dive (explicit opt-in): Full 8-step research workflow: parallel data gathering across projects, archives, and web; hackathon analysis; incumbent validation; gap classification; opportunity ranking; and a structured report with revenue model, GTM strategy, and risk assessment.
Deep dive activates when you say “vet this idea”, “deep dive”, “full analysis”, “validate this”, “should I build X?”, or “is X worth building?” — or when you accept Copilot’s offer to go deeper.

​
Evaluated against real prompts

Copilot is rigorously evaluated against a suite of prompts covering DeFi lending, MEV protection, prediction markets, AI agent payments, privacy infrastructure, DePIN skepticism, cross-domain synthesis, and empty-result handling. See Examples for annotated prompt/response pairs from evaluation runs. Copilot supports Claude Code, Codex, and OpenClaw. See Getting Started to set up in under 5 minutes.