The AI Dispatch #33 - Agent Pear 🍐

Ladies, robots and gentlemen, welcome to another AI dispatch!

This week, we mainly focus on a guest piece from Pear Protocol, which has recently announced its agent, Pear, alongside our classic market alpha roundup, this time shorter, but with a focus on quality over quantity.

Let’s get straight to it.

---

Agent Pear

The particularity of Agent Pear is that this is not a separate product but a complement to Pear’s trading platform.

Pear is, in fact, a platform where users can simply open up pair trades. Pair trading is more strategic than regular trading, as it explores the relationship between tokens, thus exposing users to less risk compared to traditional perpetual trading.

While the concept of pair trading is widespread within TradFi, it is yet to be widespread in crypto.

How can users leverage the Pear Agent for pair trading?

The Pear agent is an agent for “statistical arbitrage” between two assets.

Statistical arbitrage is based on the concepts of correlation and spread.

The Pear agent works by continuously monitoring the “z” score between two assets. The z-score is a way to measure statistical deviation from the mean spread between the assets. In simple terms, this means the agent can identify when assets behave “too differently” and are uncorrelated with each other more than usual, presenting a pair trading opportunity.

Whenever the z-score falls below -2, it can be interpreted as a long signal, and vice versa.

Let’s assume two assets usually move closely together. If the spread between them goes beyond two standard deviations, then users can:

• Go long one token and short the other

• Hold until the spread goes back to the mean

• Exit with profit once the mean reverts

With the launch of Agent Pear, the barrier to entry to pair trading is lowered, as opportunities are abstracted and automated through it. Users don’t need to stay on top of charts, but they can simply set up their agent and let it do the work for them.

In the backend agent Pear possesses a full statistical library of asset correlations, z-score, beta and spread, which updates in real time.

By monitoring these relationships, the agent Pear can capitalise on opportunities that arise from movements in z-scores.

How is Agent Pear complementary to Pear Protocol?

• Able to execute on the signals

• Supporting all pairs in this supported platform

• Automated and AI-driven

The interesting aspects of Agent Pear are the customisation of the libraries in the backend, and the fact that the agent is able to contextualise the divergences according to broader market conditions: “A spread breaking down during extreme volatility is very different from one in a quiet market”.

Through this design, the agent is able to leverage two insights:

• Viable signals: Opportunities which are analysed and pass the initial filters (e.g. there is a z-score interesting enough to test).

• Live trades: viable signals trigger alerts, which Agent Pear can execute by integrating directly in DEXs like Hyperliquid.

Being the Curtains: How does Agent Pear work?

Let’s make this funnier and try to understand how Pear protocol works behind the curtains.

Here are its main components:

Core Services: basic functioning

• Configuration Service: provides credentials, settings and database connection, keeping the system flexible.

• Orchestrator: gets data, stores it, runs analytics

Automated Data Pipeline: from data to insights

Transforming data into actionable insights.

• Ingestion layer: pulls time-series price data and asset/pairs metadata from exchanges.

• Storage layer: stores data in a unified layer accessible to all services

• Orchestrator leverages Pear’s custom calculator, analysing correlation, cointegration, z-scored, volatility, etc., translating data into insights

• Results persistence: all outputs are available and verifiable in tables.

API and Integration Layer: Acting upon these insights

• API Layer: handling requests from agents or trading venues, pulling data from the database.

• Security and signals: an architecture that provides reliability, security, and real-time performance, with all requests validated using API keys.

All these work together across two phases:

• Batch Generation: This is the scouting phase where the agent finds a watchlist of opportunities and provides a small qualitative analysis.

• Continuous Monitoring: opening trades when the z-score looks favourable, managing them according to strict rules.

This week marked Agent Pear’s launch, where the agent shows key stats, such as correlation and z-score.

Make sure to give it a try, as access will be limitedly free in September. In the future, with the launch of the 2nd Phase, users will be required to either own $stPEAR or $PEAR to access the signals and get alerts on when to open and close positions.

What’s in store beyond Phase 2?

The Pear Agent deployment will continue in phases, with phase 3 planning to make the agent conversational through LLM prompts, able to connect with different APIs and source data on funding rates, liquidations, unlocks, etc.

Last but not least, Phase 4 will enable the agent to enter, monitor, and close positions autonomously. Users will be able to deposit stables in Automatic Vaults, where the Agent executes trades based on specific parameters, moving towards “AI-powered Asset Management”.

---

Weekly Alpha Roundup

Here’s your homework for the week, with a lot of interesting updates lined up for the week:

Information as the new currency? Here’s how Backroom is becoming the “community version of Echo for small-cap tokens”:

Also looks like we’re going to soon have a whole new wave of AI-based anime from giga brain chads:

We also wrote a little something on the importance of going beyond GPT wrappers if we want DeFAI agents to manage significant funds:

Giza just launched their Swarm Finance, read here to learn what it’s about:

That's it for this week! See you next Monday.