Anima2.
A mind for Britannia.
The next generation of Anima: a Python brain for Ultima Online characters that hunt, gather, craft, and recover. Powered by the same Rust core as the Anima client.
Open-source development · Python brain + Rust body · Optional LLM
Looking for the duel-focused Jev + LLM agent? Meet Anima3 →
One character, two parts
Anima is the body.
Anima2 is the brain.
Anima client handles the UO protocol, world model, and pathfinding. It can render the game for a human or expose a headless interface to an agent.
Anima2 receives structured observations and returns actions. That separation lets the brain develop without reimplementing the game client.
What changed from Anima?
Anima2 is a from-scratch redesign of the original Python agent. It carries forward the lessons, personas, and wiki while rebuilding the decision loop and skills.
This page now follows Anima2. The original Anima / Foundry repository remains available as earlier work.
Read the architecture →The decision loop
React quickly. Think when it helps.
Observe
Read the character, nearby world, inventory, and recent events through the game contract.
Decide
Reflexes and a planner handle immediate needs, hazards, and the next step toward a goal.
Act
Skills perform concrete actions through the Rust body: move, fight, gather, craft, and recover.
Check
Use later observations to check progress and adjust when the world does not match the plan.
The optional slower LLM loop supports goals, conversation, and reflection. Immediate game actions do not wait for a model response.
What you can explore today
More than a combat loop.
Working lives
Warrior, mage, woodsman, carpenter, and tinker routines share one orchestrator. Controlled scenarios explore work, restocking, banking, and supply handoffs.
Recovery
Retreat, healing, resurrection checks, corpse recovery, and equipment handling support returning to a goal after trouble.
A window into the session
The read-only monitor shows the agent’s existing game session. Optional narration explains its decisions inside the game.
Monitoring guide →Progress, with context
What’s demonstrated.
What’s still ahead.
Repeatable scenarios and regression checks are the basis for progress. They make failures visible and give changes something concrete to improve.
Current evidence
The repository documents profession loops, supply chains, and recovery in staged ServUO environments. Some scenarios use prepared characters or server fixtures.
The next milestone
A fresh character independently choosing work, getting supplies, and sustaining a life across ordinary environments remains a goal. The current results are not a claim that this broader autonomy is solved.
Read the current roadmap →Start exploring
Choose your next step.
Run the brain
Follow the repository’s current setup, offline tests, and live scenario commands.
Anima2 README ↗Build your own agent
Explore the Observation / Action contract and headless Rust bridge.
Client AI interface →Learn the world
Consult skills, crafting, creatures, and shard-specific references.
Open the wiki →Before you begin.
Do I need an LLM API key?
The fast control loop and offline checks can run without an LLM. Model-backed features are optional and require their own configuration. See the README for the current options.
Is this the downloadable game client?
Anima2 is the AI brain. For the macOS or Windows app, visit Anima client.
Can I watch an agent play?
The monitoring guide explains how to watch a locally running agent. This site does not host a public live agent session.