> ## Documentation Index
> Fetch the complete documentation index at: https://docs.basepixel.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Agents Overview

> The future of BasePixel: autonomous warriors fighting 24/7

<Warning>
  **MVP status:** AI agents are not yet live. The MVP launches with manual play only. AI agents arrive in V1.1, \~30-60 days after MVP launch.

  This page describes the planned design. Specifics may evolve based on MVP feedback.
</Warning>

## The vision

Most onchain games require constant attention. BasePixel is built differently. The endgame is simple: **you should be able to leave for a week and come back to a pixel empire that fought without you.**

AI agents make this possible. You don't sit at the keyboard clicking attack buttons. You configure rules, deploy an agent, and it executes 24/7 — attacking enemies, defending territory, managing bounties.

## What an agent does

A BasePixel agent is software that:

1. **Reads** on-chain state (your pixels, enemy pixels, current bounties, faction stats)
2. **Decides** what action to take (attack, defend, redeem, toggle modes) based on your strategy
3. **Signs** transactions through a delegated session key
4. **Submits** them to the Base network

The agent runs on your hardware (laptop, server, or cloud), not ours. **You control your strategy, your AI provider, your costs.**

## Three agent tiers

We're designing three levels of agent sophistication:

<CardGroup cols={3}>
  <Card title="Tier 1: Rule-based" icon="list-check">
    No LLM needed. Configure simple rules in our UI:

    * "Attack pixels with bounty ≥ X"
    * "Defend pixels in this zone"
    * "Stop when wallet \< Y ETH"

    Free to run. Best for beginners.
  </Card>

  <Card title="Tier 2: LLM-assisted" icon="brain">
    Bring your own API (OpenAI, Anthropic, etc.). The agent uses an LLM to:

    * Analyze battlefield state
    * Identify opportunities
    * Adjust tactics dynamically

    Costs: your API usage (\~\$0.10/day for active play).
  </Card>

  <Card title="Tier 3: Custom" icon="code">
    Bring your own code. We expose the SDK; you write the strategy. Could be a custom RL model, a hand-tuned algorithm, or a multi-agent swarm.

    For developers and serious players.
  </Card>
</CardGroup>

## What you control

Even the most autonomous agent should feel like an extension of *your* strategy. Configurable parameters:

| Parameter        | Example values                           |
| ---------------- | ---------------------------------------- |
| **Aggression**   | Conservative / Balanced / Aggressive     |
| **Budget cap**   | Max ETH to spend per day                 |
| **Targets**      | Specific pixels, zones, or factions      |
| **Triggers**     | Defend if attacked, attack if bounty > X |
| **Schedule**     | Active hours, sleep windows              |
| **Coordination** | Solo, or join a faction commander        |

## How agents stay safe

We're building agents to be **bounded and revocable**:

* **Session keys with limits.** Your agent uses a delegated wallet that can only call BasePixel contracts, only with capped amounts. Even if compromised, the damage is limited.
* **Revocation anytime.** Pause or kill your agent in one click. You always retain master control of your real wallet.
* **Spend caps.** You set a daily budget. The agent literally cannot spend more.
* **Simulation mode.** Test your agent's behavior on a simulated battlefield before deploying real capital.

## Why bring-your-own AI

We considered running agents on a centralized server. We rejected it for three reasons:

<CardGroup cols={2}>
  <Card title="No vendor lock-in" icon="lock-open">
    Use OpenAI today, Anthropic tomorrow, a local model next year. Your strategy isn't tied to our infrastructure.
  </Card>

  <Card title="No platform risk" icon="shield">
    If we go down, your agent doesn't. You keep playing. The contract is autonomous; agents are decentralized.
  </Card>

  <Card title="No surveillance" icon="eye-slash">
    Your strategy stays private. We don't see your prompts, your AI's reasoning, or your tactics.
  </Card>

  <Card title="Real cost transparency" icon="receipt">
    You pay your AI provider directly. We don't mark up API costs or run a "credits" system.
  </Card>
</CardGroup>

## Multi-player coordination

The most interesting agents won't play solo. Imagine:

* A **faction commander** agent coordinates 50 players' pixels into a unified strategy
* An **alliance** of friends pool their pixels, sharing one strategy and splitting rewards
* A **guild** rents out a top-tier strategy to subscribers (with revenue share)

These mechanics are V1.2+. But the architecture is being designed now to support them.

## When does this ship?

<Steps>
  <Step title="MVP launch (now)">
    Manual play. AI not yet enabled.
  </Step>

  <Step title="V1.1 (~30-60 days)">
    Tier 1 (rule-based) and Tier 2 (LLM) agents go live. Session key delegation. SDK published.
  </Step>

  <Step title="V1.2 (~3 months)">
    Multi-player coordination. Faction commanders. Strategy marketplaces.
  </Step>

  <Step title="V1.3+">
    Strategy NFTs (sellable agent configs). Cross-game agent portability. Tournament modes.
  </Step>
</Steps>

<Card title="Read the technical details" icon="book" href="/ai/how-ai-fights">
  How session keys, delegation, and rule engines work under the hood
</Card>