Reentrant Session Checkpointing & Episodic Vector Memory Stores in INTHON

Long-running AI agent workflows face frequent interruptions: API rate limits, network disconnects, human approval delays, and process restarts. Today we take a deep dive into INTHON's native Session Checkpointing and Vector Memory Store capabilities.

1. Reentrant Session Checkpointing

When an agent plan encounters an approval gate or long-running subtask, INTHON automatically serializes the active call stack frame into a reentrant checkpoint file. Upon resuming, execution picks up at the exact opcode instruction pointer where it paused.

2. Episodic Memory Statements (`remember`, `recall`, `forget`)

Memory management in INTHON is part of the core language syntax:

// Save semantic fact to persistent memory
remember "User prefers dark mode UI and concise code blocks" in user_store

// Query memory via vector embeddings
let results = recall "UI preferences" from user_store

// Clean up memory
forget "dark mode" from user_store

3. SQLite & In-Memory Storage Backends

INTHON memory stores support both lightweight zero-dependency in-memory mode and persistent SQLite disk databases (`.inthon/memory.db`):

# Run script with SQLite-backed persistent memory
inthon run --persist-memory workflow.inth