Cointext

Cointext is an autonomous, crypto-native AI agent that acts like “context for every coin.” It watches wallets, tokens, and contracts across chains in real time, explains what’s happening in plain language, and can then actually take action (pay, signal, or call other agents) instead of just giving you a dashboard.

What users get:

“Should I care about this wallet?” → Cointext inspects the wallet using Blockscout/Autoscout-style onchain traces and surfaces behavior patterns (accumulation, dumping, wash volume, insider clustering, etc.) instead of raw tx spam.

“Is this token safe or just hype?” → Cointext pulls structured price/volatility/oracle info (Pyth) plus liquidity/holder distribution and returns an institutional-style risk brief, not just a chart.

“Can we automate the response?” → the agent can trigger follow-up intents: send/stream a payment in PYUSD, notify other agents via the ASI Alliance stack (Agentverse / uAgents / ASI:One chat protocol), or prepare an execution on fast finality infra (Hedera).

Who this is for:

On-chain funds / degen treasuries who need instant due diligence on a wallet or token before they ape.

Retail users who just want a yes/no answer in normal language, on OpenAI / x402 / X / Base app, instead of 5 dashboards.

Other agents. Cointext is designed to be called as an “analysis oracle” by other autonomous agents, not just by humans.

Why this matters: Right now crypto “AI bots” mostly hallucinate narratives. Cointext is different: it’s grounded in verifiable onchain data (Blockscout), trustworthy market data and randomness primitives (Pyth & Pyth Entropy), compliant stable settlement rails (PYUSD), and production-grade execution surfaces for autonomous agents (ASI Alliance + Hedera). So instead of “ask ChatGPT about a token,” you get “spin up an agent that can audit a token, price the risk, settle the fee, and ping another agent — all on-chain.”

Architecture (high-level):

Agent Brain / Reasoning Layer

Cointext runs as an autonomous agent that can be messaged via chat-style interface (human or other agents).

We integrate ASI Alliance components:

ASI:One / Chat Protocol for human ↔ agent messaging and agent ↔ agent messaging.

uAgents + Agentverse-style registry so Cointext can be discovered and invoked as a service by other agents instead of just a private bot.

Structured reasoning / knowledge graph style memory inspired by MeTTa so we can attach provenance to claims (“this wallet dumped 42% of supply on Base in the last 2h, here’s the TX set”) instead of vague guesses.

This checks the “Best use of Artificial Superintelligence Alliance” box because we’re not just a single chatbot — we’re exposing Cointext as an agent other agents can call, with explainability and cross-agent coordination. ethglobal.com

Onchain Intelligence Layer (Blockscout)

We query Blockscout/Autoscout explorer APIs to pull raw tx history, label counterparties, and cluster behavior around a wallet or token contract.

We then summarize it into plain-English risk/compliance style outputs (accumulation patterns, bridged inflows, MEV-like behavior, rug risk signals).

We also plan to surface this analysis back into an Autoscout-style “self-serve explorer launchpad” flow so projects can ship a basic transparency dashboard automatically, not manually.

This directly targets “Best use of Autoscout self-service explorer launchpad / Blockscout SDK,” which rewards projects that turn Blockscout from ‘explorer UI’ into embedded intelligence/UX. ethglobal.com

Market Data / Oracle / Randomness (Pyth)

For any token a user asks about, Cointext pulls Pyth price feeds to ground valuation, volatility, and liquidity context in live data rather than hallucination.

For simulations (“what happens if this wallet dumps 20% of supply?” “what liquidation price are we playing with?”) we also consume Pyth Entropy to get unbiased randomness for stress tests / Monte Carlo style scenario sampling.

This maps to Pyth’s “use Pyth Entropy to generate and consume randomness on-chain” requirement for the Pyth Entropy Pool Prize. ethglobal.com

Settlement & Commerce Layer (PayPal USD / PYUSD)

After analysis, Cointext can propose actions like “tip this researcher,” “unlock the premium report for this wallet,” or “escalate this alert to the fund’s main Telegram.”

Those payments / micro-access fees are denominated and settled in PYUSD (PayPal USD) so that the flow is actually usable in real commerce terms (USD), not just speculative ERC20.

This aligns with PayPal USD’s prize criteria: real-world payments and UX using PYUSD, with clear business logic and a path to monetization (pay-per-insight, pay-to-alert, pay-to-escalate). ethglobal.com

Execution Surface (Hedera)

We deploy an execution module on Hedera to prove we can do “agent-to-agent escalations” with low fees and fast finality.

Two parts:

Use Hedera’s EVM-compatible layer / Agent Kit style pattern to register Cointext as an autonomous actor that can log an “investigation result” or raise a compliance-style flag as an immutable on-chain event.

Optionally open a PYUSD-denominated bounty (or internal credit) for follow-up investigation by another agent, and settle + notarize that handoff on Hedera’s high-TPS network.

Hedera explicitly markets 10,000+ TPS, ~3s finality, aBFT security, and now agent-focused tooling (Hedera Agent Kit + Google A2A). We’re using that as the fast coordination / audit trail layer for multi-agent collaboration, which is exactly what Hedera is pushing in their ETHGlobal track (Hedera + AI agents, low-cost, production-grade infra). ethglobal.com

Frontends / surfaces

On-chain x402 / X bot style chat for humans (“Is this token a rug?” “Track this wallet for me”).

Agent-to-agent API surface via ASI:One / Agentverse listing (so other autonomous agents can call analyzeWallet(address) or priceRisk(token) and get a structured answer). ethglobal.com

Lightweight dashboard powered by Blockscout data + our summaries: basically “Autoscout but opinionated,” meant for founders, funds, and KOLs to show credibility.