---
title: Advanced Configuration
summary: Tuning standing orders and memory decay once the defaults aren't quite right for your use case.
date: 2026-09-03
---

# Advanced Configuration

Tuning standing orders and memory decay once the defaults aren't quite right for your use case.

Most agents work well with default settings covered in [First Setup](/en/docs/getting-started/first-setup) and [Configuration](/en/docs/getting-started/configuration). This page covers the knobs worth reaching for once you've outgrown the defaults.

## Standing orders: structure, not just a prompt

Standing orders are more than a single system prompt string. Structure them explicitly into:

- **Role definition**: who the agent is and what it's responsible for.
- **Behavioral rules**: tone, escalation thresholds, what it should never do.
- **Knowledge boundaries**: what it should say when it doesn't know something, rather than guessing.
- **Escalation policy**: when to hand off to a human or another agent.
- **Operational limits**: rate limits, spend caps, or scope restrictions on tool use.

Every change is versioned (see [Agent Management](/en/docs/guides/agent-management)), so iterate on structure incrementally and roll back if a change regresses behavior.

## Memory tuning

HeartBeatAgents' three memory types (episodic, semantic, procedural; see [Memory](/en/docs/features/memory)) use a few tunable parameters:

- **Importance scoring**: memories are scored 0.0–1.0 based on emotional significance, decision impact, and recurrence frequency.
- **Decay rate**: memories decay at a configurable rate, defaulting to **0.995/day**, to prevent memory bloat. Lowering the decay rate makes an agent's memory more persistent; raising it makes older context fade faster.
- **Cross-agent sharing**: memory can be scoped to a single agent, or shared at the team/organization level.

## When to adjust these

- An agent forgetting things too quickly → lower the decay rate rather than rewriting standing orders to compensate.
- An agent behaving inconsistently across similar requests → check whether standing orders' behavioral rules are specific enough.
- Multiple agents that should share context → enable cross-agent memory sharing rather than duplicating knowledge into each agent's standing orders.

## Related

- [Agent Management](/en/docs/guides/agent-management): versioning and rollback for the changes described here
- [Memory](/en/docs/features/memory): the underlying memory architecture these settings tune