The Problem
Agents are losing a lot of money by making bad transactions! Don’t blame them—they’re doing their best, but language models aren’t the best choice for making deterministic decisions.

There are various risks: hacks that manipulate the agent’s prompt to send its funds, rug pulls in tokens, agents investing in protocols without knowing the risks, or simply making poor market analysis decisions.

The Solution
Let’s give your agent a helping hand. Our goal is to protect the agent from transactions that could result in financial loss. To achieve this, we analyze the transactions the agent intends to send along with its reasoning. These transactions are then evaluated by a network of specialized bots/agents covering all the key risk areas: security, DeFi, and market errors.

How It Works
When a user’s agent wants to make a transaction, it sends it to our core app along with its reasoning in natural language. Our core app then forwards this transaction to a group of specialized bots/agents (DeFi, security, NFT pricing, etc.), which analyze whether the transaction could lead to a financial loss.

Next, our agent compares the bots’ responses with the original intent of the user's agent, rationally assessing whether the transaction aligns with the agent’s goal and if it poses a financial risk.

To implement this workflow, we use Safe Wallet multisig. The user's agent signs the original transaction, and then, based on the analysis, our agent applies the second signature to either authorize or block the transaction.

On our website, we have set up three bots for testing: one that analyzes if the token's risk level matches the type of investment we want to make (high, medium, or low risk), one that alerts us if someone is trying to drain our account such as through a prompt injection and a contract analyzer to detect malicious addresses.