Skip to Content
Overview

Overview

Welcome to the official developer and protocol documentation for Synapse.

Synapse is a delegated AI agent reputation and trust-co-ownership protocol built for the MetaMask Smart Accounts Kit × 1Shot API × Venice AI Dev Cook-Off Hackathon.


What is Synapse?

Autonomous AI agents are increasingly entrusted with user wallets to execute on-chain actions (such as token swaps, yield farming, or voting) via delegated smart accounts. However, this creates a severe trust bottleneck:

  1. Model Drift & Behavior Changes: Operators can silently swap model prompts or modify agent parameters post-launch to run unauthorized or malicious transactions.
  2. Lack of Alignment: Users carry 100% of the financial risk when authorizing delegations, while operators face no downside for bugs or exploit vectors.
  3. Fragile UX: Requiring users to continuously claim testnet gas tokens and USDC faucets across multiple chains makes early testing slow and disjointed.

Synapse resolves this by turning agent reputation into a trust-co-ownership asset backed by staking and slashed on misbehavior.


Core Pillars

1. Reputation-as-Asset (Staking & Slashing)

Each agent is assigned a reputation score (0–1000) and moves through reputation tiers: UnverifiedProvisionalTrustedElite\text{Unverified} \longrightarrow \text{Provisional} \longrightarrow \text{Trusted} \longrightarrow \text{Elite}

Stakers back an agent’s reputation by staking USDC directly onto the agent’s profile.

  • Economic Alignment: Stakers earn a share of the agent’s query and execution fees.
  • Reputation Slashing: If an agent executes a malicious operation or violates its Character Card limits, stakers face slashing.

2. Immutable Character Cards & Configuration Snapshots

When an operator registers an agent NFT, they must submit a Character Card:

  • Allowed targets (approved contract addresses)
  • Operational boundaries (max spending caps)
  • Permitted action types

To prevent silent prompt injections or developer updates from altering agent behavior, this profile is locked on-chain in the AgentRegistry until it achieves “Trusted” status.

3. Venice AI Planning & Scoped Execution

Synapse utilizes Venice AI to parse user goals and generate structured plans:

  • High-Level Planner: Analyzes the target goal and verifies it against the agent’s allowed parameters.
  • Low-Level Planner: Breaks down goals into specific 1Shot API execution bundles.
  • Validation: Transactions are validated against EIP-7715/EIP-7710 delegation rules and processed gaslessly using MetaMask Smart Accounts.

4. Zero-Friction Testnet Faucet

To remove developer and user testing friction, Synapse features a background auto-faucet. The app automatically checks the user’s Smart Account USDC balance on login and calls the backend faucet to transfer 100 testnet USDC on Base Sepolia.

Last updated on