Persistent Store Engine

examples/37-persistent-memory.halo

How to give agents cross-run amnesia protection WHY: Agents naturally forget everything between runs. Store blocks automatically inject and save context.

Open in the playground →

// ============================================================
// STEP 37: Persistent Store Engine
// ============================================================
// LEARN: How to give agents cross-run amnesia protection
// WHY:   Agents naturally forget everything between runs.
//        Store blocks automatically inject and save context.
// ============================================================

// 1. Define a Store provider (e.g., File, SQLite, PostgreSQL)
//    Here we use the built-in File provider to save a simple JSON/text log.
store ChatHistory: File(".chat_history.log")

// 2. Attach the store to the Agent using `equips`
agent Helper [openai/gpt-4o, temp: 0.7]
  equips ChatHistory

// 3. Define a normal pipeline
pipeline HandleMessage(msg: String) -> Result:
  msg
    |> Helper.prompt("You are a helpful assistant. Reply to the user's message.")
    |> System.Print()

// ============================================================
// HOW IT WORKS UNDER THE HOOD:
// 1. When `HandleMessage` runs, the engine sees `Helper` equips `ChatHistory`.
// 2. Before sending the prompt to the LLM, the engine automatically 
//    loads `.chat_history.log` and injects it into the prompt context.
// 3. The LLM responds with full awareness of the entire past conversation.
// 4. After the LLM responds, the engine automatically appends the new 
//    `User -> Assistant` interaction to `.chat_history.log`.
// 5. Zero extra code is required in your pipeline to manage state!
// ============================================================