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

# When to use StateBase

> Identifying the anti-patterns of stateless agents.

Reliability in AI agents is not about making the model smarter; it's about making the system more resilient. You need StateBase when your agent moves beyond simple Q\&A into complex, multi-turn workflows.

## The Signs You Need StateBase

### 1. The "Memory Leak" Prompt

**Anti-pattern**: You are manually appending history to a string that grows until it hits the context limit.
**The Fix**: Use **StateBase Sessions**. We manage the sliding window and semantic pruning so you only ever feed the model relevant context.

### 2. The "Goldfish" Effect

**Anti-pattern**: Your agent forgets the user's name or a key constraint halfway through a task.
**The Fix**: **Durable State**. Define specific variables that must persist (e.g., `user_authenticated: true`) regardless of chat history.

### 3. The "Infinite Loop" Hallucination

**Anti-pattern**: An agent calls the same tool 10 times with slightly different parameters, wasting \$1.00 in tokens.
**The Fix**: **Decision Traces & Rollbacks**. Detect a loop and instantly revert the agent to its last "sane" state.

### 4. The "Black Box" Problem

**Anti-pattern**: A customer says "the agent broke," and you have no way to reproduce exactly what happened.
**The Fix**: **Replayability**. Every turn is a checkpoint. You can replay any session step-by-step in your local terminal.

## When NOT to use StateBase

* Single-turn completions (e.g., translation, summarization).
* Simple "search-and-answer" bots with no ongoing state.
* Prototypes where reliability is secondary to speed.

StateBase is for **Production Agents** where failure has a cost.