Video thumbnail

Strarifi.xyz

An Intelligent, autonomous agent for DeFi, optimising yield & liquidity

Strarifi.xyz

Created At

ETHGlobal Cannes

Project Description

We built an autonomous DeFi research agent that helps users discover the top-performing Morpho protocol markets across supported chains — Ethereum, Arbitrum, Base, and Katana — ranked by TVL, supply, borrow, and activity.

The agent is hosted on Agentverse (ASI), built using uAgents, and uses ASI:One's asi1-mini LLM to deliver concise, contextual explanations like:

“Why is this the best market?” “What are the pros and cons of this pool?”

It's designed for natural language queries (e.g. “Top 3 pools on Arbitrum”), and responds with ranked lists + reasoning, fully autonomously.

From a technical angle:

We used The Graph to query all Morpho, Aave, and Lido pools via subgraphs.

This data is passed into the agent, which feeds it to the LLM for ranking and reasoning.

The agent runs on the ASI stack: uAgents, Agentverse, ASI:One.

For Katana, we integrated it as a supported chain for Morpho (TVL, borrow, supply queries) — making this one of the first LLM-powered agents using Katana state for DeFi search.

Tracks Tackled: ASI Alliance: AI agent built with uAgents, hosted on Agentverse, integrated with ASI:One, runs fully autonomously.

Katana: Uses Katana as a supported chain for querying real-time market data from Morpho.

The Graph: Pulls all pool data using subgraphs for Morpho, Aave, and Lido — used directly by the LLM for market ranking and reasoning.

Why it fits This isn’t a toy agent. It’s production-grade infra showing how AI agents can power real DeFi research — no dashboards, no SQL, no clicking around. Just query and get answers.

We tackled cross-chain complexity, integrated real subgraph data, and deployed an actual reasoning layer (LLM) — all wrapped into a live, discoverable agent that anyone can chat with.

Try asking it:

“Top Morpho markets on Katana” “Best pool across all chains” “Give pros and cons of the #1 market”

How it's Made

We built a cross-chain DeFi research agent using the ASI Alliance stack: uAgents, Agentverse, and the ASI:One LLM. The agent autonomously ranks Morpho protocol markets based on live data, generates natural language insights, and is discoverable on asi1.ai.

Core architecture (ASI )

  • uAgents: Handles agent lifecycle, message parsing, and response logic. The agent listens for user prompts, extracts intent (chain name, “top N”, etc.), and runs the appropriate data-fetch and ranking logic.
  • Agentverse: The agent is deployed and registered here, making it discoverable on the Almanac contract and usable through asi1.ai.
  • AS-1: We used the asi1-mini LLM to interpret raw pool data and return plain-English insights. This includes “Why is this market ranked #1?” and “What are the pros/cons for users?”

Data pipeline (Graph)

  • The Graph: We query Morpho, Aave, and Lido subgraphs for pool metrics (TVL, total supply, total borrow, and activity status). These queries are executed per-chain and per-protocol.
  • This data is passed directly into the LLM for contextual ranking and summary generation.
  • Chains supported include Ethereum, Arbitrum, Base, and Katana.

Katana integration (Katana )

  • used as one of the supported chains in our agent. Its Morpho subgraph is queried live.
  • Pool data from Katana is included in global rankings and exposed through the LLM’s reasoning output.
  • This makes the agent one of the first LLM-backed interfaces interpreting Katana DeFi data via natural language.

Key Engineering Deets:

  • Intent parsing is done with regex-based checks to detect terms like “top 3”, “best”, or specific chains.
  • Agent fetches market data via GraphQL POST requests using The Graph’s API key gateway.
  • Prompts sent to ASI:One LLM are formatted using structured JSON-like strings embedding market metrics. These prompts request brief, explainable reasoning and bullet-style pros/cons.
  • Results are returned as plain text with clean formatting for chat display.

Stack Summary

Agent Infra: uAgents, Agentverse LLM: ASI:One (asi1-mini and asi1-extended) Data: The Graph (Morpho, Aave, Lido subgraphs) Chains: Ethereum, Arbitrum, Base, Katana Deployment: Hosted on Agentverse and discoverable on asi1.ai

Notable hacks

  • Built a full data-to-LLM loop using only chat prompts and structured reasoning output — no frontend, no dashboards.
  • Designed fallback aggregation across chains when no chain is specified in user input.
  • Prompt-based explainability lets users get reasons and tradeoffs for any market without knowing DeFi internals.

You said: remove the word track, stack summary, and put '—' of sub pointers

ChatGPT said: Absolutely, here’s the updated version without “track,” with cleaned structure, and using — for sub-points:

How it's made

We built a cross-chain DeFi research agent using the ASI Alliance stack: uAgents, Agentverse, and the ASI:One LLM. The agent autonomously ranks Morpho protocol markets based on live data, generates natural language insights, and is discoverable on asi1.ai.

Core architecture

— uAgents handles agent lifecycle, message parsing, and response logic. The agent listens for user prompts, extracts intent (chain name, “top N”, etc.), and runs the appropriate data-fetch and ranking logic. — Agentverse hosts and registers the agent, making it discoverable on the Almanac contract and usable through asi1.ai. — ASI:One powers the natural language layer. We use the asi1-mini LLM to interpret raw pool data and generate explainable outputs like “Why is this market ranked #1?” or “What are the pros and cons?”

Data pipeline

— We use The Graph to query Morpho, Aave, and Lido subgraphs for key pool metrics: TVL, total supply, total borrow, and activity status. — These queries are executed for all supported chains: Ethereum, Arbitrum, Base, and Katana. — Queried data is directly passed to the LLM for contextual ranking and summary generation. — Filtering, sorting, and top-N aggregation are done before sending the final dataset to the reasoning engine.

Katana integration

— Katana is a supported chain with its Morpho markets integrated via The Graph. — We query its subgraph and include Katana pools in both per-chain and global rankings. — The LLM’s output reflects Katana market data, making this one of the first autonomous agents parsing Katana state in natural language.

Engineering details

— Intent parsing is done via regex to extract signals like “top 3”, “best”, or chain names from natural language queries. — GraphQL queries are executed via POST with API key authentication to The Graph’s gateway. — LLM prompts are constructed dynamically in a structured format embedding market metrics. — ASI:One returns concise reasoning and pros/cons which we format into readable responses via uAgents chat. — The system includes a fallback mechanism: if no chain is specified, we aggregate top markets across all chains by TVL.

Technologies used

— Agent infra: uAgents, Agentverse — LLM: ASI:One (asi1-mini) — Data: The Graph (Morpho, Aave, Lido subgraphs) — Supported chains: Ethereum, Arbitrum, Base, Katana — Language: Python — Deployment: Hosted on Agentverse, discoverable on asi1.ai — UI: Text-only chat via uAgents protocol

Notable hacks

— Built a complete data-to-LLM loop using just structured prompts and output — no dashboards or frontend. — Prompted LLM for both reasoning and decision support, turning on-chain data into usable human insights. — Integrated Katana as a live DeFi data source and made it usable via natural language without requiring UI exploration.

background image mobile

Join the mailing list

Get the latest news and updates