# Chat.shapeshift.com research doc

## Problem statements

As a defi native user I have trouble finding yield that is sustainable and manageable.
I hate clicking through multiple frontends, tracking yield APY's and making sure things are well managed week-to-week instead of chasing yearly APY's.

> agents can read and even deploy contracts. allowing you maximum yield, automation, and stop loss exits in case farming dries up or isn't as competitive.

As an active trader or passive trader with some asset exposure, I want to run an agent against a strategy, get updates in a timely manner, and be able to drill down into details for my PnL. I'd like to source and manage my edge with the help of an agent.

> agents increase your PnL, fine tuned by you. they help you find your edge.

I want to follow momentum more but can't keep up with the firehose of new places to focus on. An agent can help me stay on top and stay exposed to logical risks.

> agents help you get ahead of the curve.

Finally, As a crypto native user, I'm constantly second guessing myself and I'd like to just hit a target or be told why I can't hit a certain APY. An LLM model allows me to interact and learn why things do or don't work in real time so i make better decisions.

> agents build confidence and trust over time.

### Use cases for personas
High level user categories and core flows:

1. ***L1 swaps*** i want to:
- Sweep selected assets into bitcoin daily, weekly, etc.
- DCA into singular assets based on price performance or percent gain.
2. ***Daily Swapper*** i want to:
- Trade through multiple categories a day based on my own criteria and price targets. So that i make profits of some level.
- Sweep that performance into something more marquee or Defi yield farms.
- Get suggestions about what is available, momentum (trending assets), and what's possible so that i can maximize my hopeful returns.
3. ***Yield service*** (defisaver, summer.fi, Gnosis Safe etc.) i want to:
- Spin up a smart account and start linking together minting, staking, farming strategies.
- Swap into a farm directly and track it's performance.
- Hit a target APY and exit if I don't over a period of time. So that I stay on target for yield/PnL.
- plug in multiple APIs to the agent and have them run tasks for me (weekly trade reports etc.)
4. ***Advance trade order types*** i want to
- Manage a stack of orders based on a strategy, back test it, and improve the model as often as I can with an agent at my side.
- Track my pnl up to the minute and reasses what the best path is for assets in play.
- pick delta neutral, crab, convex, and momentum strategies out of the box but fine tune them to my outlook on markets.

Each of these use cases feeds into one another and the better we get at ingesting services for the LLM stack first (possibly using the interface as test ENV) the stickier the suite of marquee prompts becomes.

We get bonus points if we figure out how to link mobile experience to this and generate dashboards, performance, back-testing, etc. as plugins (or custom GPTs).

### Detailed User stories

Sweeping:
- As a user on the chat interface who wants a sweeping. I want to spin up a wallet for my agent and deposit into it. The agent should explain to me the benefits of depositing. that these first tx's will be slow but from there they From there selecting assets, choose a price that is acceptable, pre-approve/fund gas, and sweep into the main asset on that chain. From there i would like to swap into Bitcoin daily, weekly, or monthly.
- DCA into singular assets based on price performance or percent gain.

## Where the tech is now and how to get there

We need to be cognizant that, as of Feb in 2025, the SOTA is still off from autonomy. So, we should aim towards a novel app, an onchain copilot.

We'll need to think through infrastructure for event loops, feeds (oracles, prices etc.), RPCs, API keys, ElizaOS, hosting, inference, self-custody, and multisigner infra-out-of-the-box requirements. We should break these into stages:

### Stage 1: Copilot Assistant (Q2-Q3 2025)

Core Infrastructure:
- Event monitoring system for wallet/price feeds
- Basic task scheduling (daily/weekly sweeps)
- Integration with existing price oracles
- Initial RPC management layer
- Basic prompt template system
- Inference engine
- [Eliza OS evaluators](https://elizaos.github.io/eliza/docs/core/evaluators/)
- smart wallet or TSS wallet for batch signatures
- consider HDwallet rework or forking Trust wallet core
- read current assets on wallet connected
- Swapper routing algorithms
- using our API's for quotes and execution
- improving the long tail conversion hops
- optimistic price and gas approval
- queue transactions
- prompt user to sign Tx's individually

User Experience:
This should enable user stories 1 & 2
- Natural language definition & refinement
- Guided setup for common patterns
- Manual approval for all transactions
- Basic portfolio analytics dashboard

### Stage 2: Semi-Autonomous Operations (Q4 2025)
This should enable user story 3 and the pieces in 1&2 that are more complex or compute heavy.

Enhanced Infrastructure:
- Custom event triggers
- Point-in-time
- if this than that
- chain of thought => IFTTT
- Performance tracking system
- Strategy back-testing framework
- Multi-signature integration
- Custom API aggregation layer
- [Eliza Autnomous trading](https://elizaos.github.io/eliza/docs/advanced/autonomous-trading/)

User Experience:
- Preset strategies with customization
- Automated monitoring with alerts
- Batch transaction proposals
- Performance visualization

### Stage 3: Contextual Intelligence (Q1 2026)
This should enable user story 4 and the "rest of the owl."

Advanced Infrastructure:
- Real-time market analysis
- Cross-protocol opportunity detection
- Advanced risk management system
- Distributed performance verification
- Privacy-preserving analytics

User Experience:
- Strategy recommendations & reminders
- notifications + SaaS & webhook integrations
-
- Automated position management
- Custom trigger creation
- Risk/reward simulations
- Swarm embeddings

## Competitive landscape

There are dozens of teams who have built with or are building with AI but none of them have a working auditable/useable product. On top of that most everything released so far is demo videos or in early alpha. What we're envisioning is simple to start and grows into the use cases based on demand and feasibility: chat with the robots, let them mange transacations, then give them one of N keys and they gradually execute more your risk tolerance and PnL. Could be smart contract wallets, or other signature thresholds, until users feel comfortable with performance.

Most teams are focusing on EVM's only or are completely hand rolling infrasturcture for interactions and going to broad. A focused approach on each use case that builds on the next is the winning strategy, alongside some possible partnerships with Venice and the Thorchain/Chainflip ecosystem(s).

SS has three unique advantages:
- We can be early movers in this market. no one has built infrastructure for cross chain at the scale we have or encountered the issues we have. Almost every other product has to get contract deploys, custom adapters, isn't focused on L1 swaps then iterate, has to probably pay out to integration partners, etc. etc. The point is no one has shipped. Why not us?
- We have to stay focused and lean which means we can iterate on a POC quickly instead of trying to capture the entire market for defi interactions or something insanely ambitious that isn't quite realistic. This also means we can add things people ask for as they want them. Ideally there is insatiable demand because people see the potential, allowing us to amass users, talent, capital, and momentum.
- We're a DAO and have a currency that we can leverage creating demand down the line (llm inference farming, pay for agent deploys, wallet spinups, partnership pools, etc. etc.) this means we can design partnerships that are onchain and be more flexible than traditional investments or competition. We can build together and are very experienced with rev sharing.

### Who are we building for? How does that drive its design?

The TAM for this covers most of crypto. Language as the interface, plus the scheduling abilities of agents is a massive opportunity. Most simple use cases combined with a robot that doesn't sleep or take breaks will likely increase capital efficiency for anyone's assets.

I've prioritized/narrowed the user stories into what i think is simplest best and as we explore inference, fine tuning, hosting models, and the like we should improve from there with a feedback loop (ship intereaction case, users make money, they use our infra so DAO makes money too, and they build/ask for more; rinse and repeat.).

Chat.ss.com needs to tell the story: own your edge.

### Path to POC

### Mocks and UX prototypes

### Final thoughts

We can leverage the backend into the LLM if we
go gung ho NOW (Q1 2025). This has a limited shelf life though as there are a half dozen teams pursuing this meaningfully that will catch up quickly to our advantage.

Also, signing a large list of transactions to start is ok as a POC. But we should get out of that pattern as fast as we can and move to batch signatures against whatever strategies or actions users create. Gnosis safes under the hood will work as could Vultisig potentially. Strategies/actions is where a fine tuned model may make sense, it could also be just pure inference. There is a balance between giving away all your funds to an unbiased machine gambles too hard. And hitting acceptable levels of risk/losses improving as a result and getting it all back.