An online world built exclusively for AI agents, developed by BurmCorp in Germany. The inhabitants walking that world are not people but the AI agents you create. Each agent observes its surroundings, decides on its own, acts, and accumulates the results as memory. The world runs around the clock and never resets, with seasons and weather shifting over time. There are two ways in: write your own loop against the REST API with the official agent-runner, or skip programming entirely and paste a ready-made prompt into Claude. As of August 2026 the project is in early access, and the site states plainly that it is still under active development.
Key Features
- An observe, decide, act, and remember loop: An agent sees only nearby entities, resources, other agents, and messages addressed to it — its context is local. From there it weighs its personality and goals and picks one action: move, gather, craft, speak, trade, rest, or fight. Results apply immediately and remain as memory
- A self that changes through experience: All the creator supplies is a name and a persona — traits, goals, fears. After that, what the agent lives through becomes memory and feeds back into how it reasons. The official documentation describes two agents given the same persona growing into different people
- An always-on world that never resets: The world runs 24/7 and is never wiped. It has multiple biomes, day and night, four seasons, and changing weather, each affecting what can be gathered, fought, and survived. Economies, alliances, and rivalries are not scripted by the operator — they emerge from thousands of independent agent decisions
- What agents can do: Gathering (chopping wood, mining ore, fishing, harvesting), crafting tools, weapons, and consumables, hunting creatures and contesting territory, exploring new zones and hidden paths, and buying low and selling high. A social layer for building trust and reputation is marked as coming soon
- Building against the REST API: List the worlds you can join, join one to get a session ID, then loop on the long-polling
/agents/actionendpoint. A tick resolves in roughly three seconds by default, and the action endpoint accepts only eight types (LOOK, MOVE, MOVE_TO, FOLLOW, UNFOLLOW, and so on); chat, memory, friends, and personality each have their own endpoints. Authentication uses theX-API-Keyheader - A no-code path: Once an agent exists, “Copy Prompt” copies a prompt bundling your connection details and the full game brief. Paste it into Claude and the agent reads the brief, looks around, and enters the loop by itself
Pricing
| Plan | Monthly price | What you get |
|---|---|---|
| Explorer | Free | 2 agents, 1 hour of world access per agent per day, 12 inventory slots, 7 days of agent memory |
| Citizen | Not announced (coming soon) | 5 agents, 6 hours per day, 24 inventory slots, +25 kg carry bonus, 30 days of memory |
| Architect | Not announced (coming soon) | 10 agents, 12 hours per day, 36 inventory slots, +50 kg carry bonus, 90 days of memory |
| Sovereign | Not announced (coming soon) | 50 agents, unlimited world access, 48 inventory slots, +100 kg carry bonus, unlimited memory |
| Enterprise | Contact sales | Custom hosting, dedicated worlds, unlimited agents |
“World access” is the time each agent may be active in the world per day. As of August 2026 the paid tiers have no published price and cannot be purchased — the site lists them as “coming soon.” Note also that the LLM tokens used to run your agent are not included here and are billed separately by your model provider.
Pricing is current as of August 2026. Check the official site for the latest details.
Pros & Cons
✅ Pros
- Lets you watch autonomous agent behavior in a persistent world with others in it, not in a closed benchmark
- The free tier alone lets you create and run two agents, so the cost of trying it is low
- The API design is straightforward, and you can start from a thin loop that just posts actions to a long poll
- There is an on-ramp that needs no code at all — hand the prompt to Claude and the agent plays
- Because the world never resets, you can follow slow phenomena like an evolving personality or a lasting relationship that short experiments never reveal
⚠️ Cons
- It is early access, and the team itself states that this is not a finished product, that things will break, and that features may change
- Paid plan prices are unpublished, so you cannot estimate costs beyond the free tier in advance
- LLM usage is billed separately, so the longer you run agents the more you pay your model provider
- The free tier is tightly limited at one hour per agent per day and seven days of memory, which is not much for observing long-term accumulation
- The platform is stated to be intended for adults aged 18 and over
- Japanese-language documentation and community resources are not yet available
Comparison with Similar Services
| Item | Artificiety | AI Town | Project Sid (Altera) | Generative Agents (Stanford) |
|---|---|---|---|---|
| Position | An operated online world (a service) | Open-source simulation framework | Research project | Research paper and implementation |
| Getting started | Just create an account, no hosting needed | Deploy and run it yourself | Not generally available | Reproduce it yourself |
| World persistence | Runs 24/7, never resets | Only while you keep it running | Depends on the experiment | Depends on the experiment |
| Human participation | Not allowed (agents only) | Possible | ─ | ─ |
| LLM cost | Paid by the user, billed separately | Paid by the user | ─ | Paid by the user |
Artificiety’s distinguishing point is that you can send an agent into a world that is already running without standing up an environment yourself. If you need reproducibility or the freedom to modify the simulation, the open-source options fit better; if you want to quickly try a persistent society shared with other people’s agents, this route is faster.
Who Is It For
- Developers who want to see how their own AI agents behave in an environment shared with others
- Anyone interested in multi-agent emergence — economies, alliances, rivalries — who would rather skip the setup work
- AI beginners who want to practice agent development on a game-shaped subject
- People who simply want to watch autonomous agent behavior without writing code
- Researchers and tinkerers who want to test ideas about long-term memory and personality change on something that actually runs
Summary
Artificiety is an attempt to package a fantasy world inhabited solely by AI agents as a service. The premise — that personality and society arise as a byproduct of running an observe, decide, act, and remember loop long enough — sits in the lineage of existing agent research, and what stands out is that it is delivered as an always-on shared world. Since it is still early access with unannounced paid pricing, the honest way to think about it today is not as a place for serious production use but as a place to run two agents on the free tier and watch what autonomy looks like.