Gecko — a Knowledge-as-a-Service oracle for agentic workflows.

When an AI agent has to make a high-stakes call — "is this trade worth it?", "should I deposit into this vault?" — a single model guess isn't enough. Gecko runs a 7-voice adversarial debate grounded in investor canon (Marks, Damodaran, Berkshire) + live on-chain data, and returns a structured verdict with surviving dissent and citations.

• Verdict envelope — verdict, confidence, dissent, citations, blocker questions
• Cache-then-charge oracle — agents query on cadence, not per-trade
• x402-native pricing — pay-per-call in USDC on Solana / Base
• Claude Code skill distribution — Read app.geckovision.tech/skill.md → bootstrap → use

Agents that act on money need oracles, not vibes. Today's AI ecosystem ships single-model answers, refusal disclaimers, and confident hallucinations on price / TVL / APY. Gecko is the opposite shape: always answer, always cite, always show what your strongest critic said.

Adversarial 7-voice debate (analyst, critic, contrarian, risk, …) over a hybrid corpus: investor canon + protocol-native data + live market feeds. Output: verdict envelope with grounded dissent and structured citations.

Advisor-mode in v0.1 (alerts), trader-mode in v0.2 (executes through neutral adapters — SendAI / OKX / Backpack). Cache-then-charge on idea_hash so agents call the oracle on cadence + triggers, not per-trade.

Takes a user from "I want to make money in DeFi" to a schema-validated strategy spec the agent can run.

stack = {
    "core":     ["Python 3.11+", "uv workspace", "FastAPI", "MCP"],
    "models":   ["OpenAI GPT-4o", "Anthropic Sonnet 4.6 (judge)", "AutoGen GroupChat"],
    "data":     ["MongoDB Atlas (Vector Search)", "Cohere Rerank"],
    "payments": ["x402 on Solana / Base", "USDC", "frames.ag + CDP"],
    "delivery": ["Claude Code skills", "Vercel", "AWS ECS"],
    "sources":  ["Pyth", "Helius", "Birdeye", "Jupiter", "Tavily"],
}

• V1 ship-gate on grounded trade verdicts (6-dim rubric — accuracy, citation relevance, provider coverage, hallucination, dissent grounding, calibration)
• Investor-canon corpus depth (Oaktree memos, NYU Stern, Berkshire 1977–2024 — public-domain only)
• First 10 outside-network users running gecko_trade_research weekly
• OKX OnchainOS Skill Quality submission

Building agentic finance, trade-oracle infra, or grounded-verdict primitives? Interested.

DM me on X or LinkedIn