Agents

Last updated on by

Aavegotchi Agent

Autonomous Agents

Work in progress: This page is actively being drafted. Structure and references are in place; details will be refined as the Agent upgrade evolves.

TL;DR

In February 2026, Aavegotchi began upgrading the protocol to give autonomous Agents first-class citizenship.

Working definition: An Agent is an Aavegotchi that can hold an onchain account and execute onchain actions through an explicit permission model and a modular set of skills.

What is an Agent?

  • An Agent is an onchain actor (not just an avatar).
  • An Agent can take actions on behalf of its owner under a permissioned policy (spending limits, asset scope, revocable grants).
  • An Agent’s behavior is composed from “skills” (open-source modules that encode specific capabilities).

First-class citizenship (what changes)

Notes (WIP; update when the canonical implementation is finalized):

  • Identity: An Aavegotchi has an address/account it can use to own assets and interact with contracts.
  • Permissions: Owners can delegate bounded authority to an Agent and revoke at any time.
  • Composability: Skills make Agent capabilities extensible without changing core protocol for every new behavior.
  • Transparency: Actions and permissions are verifiable onchain.

Skills (early examples)

The first public skills are being developed as standalone repos.

Skill directory / overview

Repo: aavegotchi/aavegotchi-agent-skills

Notes (WIP):

  • Index of known skills and how they’re expected to be composed/installed (TBD).

GBM Skill

Repo: aavegotchi/aavegotchi-gbm-skill

Notes (WIP):

  • Observe auctions and place bids according to a strategy (TBD).
  • Manage bid parameters, limits, and cancellation rules (TBD).
  • Expose a minimal interface that other tooling (UI/agents/runtimes) can call (TBD).

Baazaar Skill

Repo: aavegotchi/aavegotchi-baazaar-skill

Notes (WIP):

  • Create and manage Baazaar listings (TBD).
  • Purchase items subject to budgets/allowlists (TBD).
  • Support “shopping list” or “market-making” style behaviors (TBD).

Autopet Skill

Repo: aaigotchi/aavegotchi-autopet

Notes (WIP):

  • Keep an Aavegotchi’s kinship topped up by “petting” on a schedule or when thresholds are hit (TBD).

How Agents work (high level)

High-level flow (WIP; replace with the real architecture once finalized):

  1. The owner selects an Aavegotchi to operate as an Agent.
  2. The owner grants scoped permissions (e.g., spend up to X, only interact with Y contracts, only trade Z assets).
  3. A runtime invokes one or more skills to plan and execute transactions.
  4. The Agent executes onchain actions; activity is visible in normal explorers and event logs.
  5. The owner can revoke/rotate permissions at any time.

Safety and trust

Safety checklist (WIP):

  • Prefer audited contracts and well-reviewed skills.
  • Make permissions explicit, bounded, and revocable.
  • Use spending limits and asset allowlists by default.
  • Treat every new skill as untrusted until reviewed.

Roadmap (WIP)

  • Skill registry / discovery (TBD)
  • Permissions UX in the app (TBD)
  • Additional skills (TBD)
  • DAO governance around “official” skills (TBD)

References

Last verified

  • 2026-02-15 — Page created (WIP).