Agent Memory

Instead of treating every interaction as isolated, memory helps agents:

• Stay consistent

• Avoid repeating questions

• Improve decision quality

• Handle multi-step tasks

Memory turns agents from responders into context-aware workers.

What Agent Memory Is (and Isn’t)

Agent memory is:

• Structured context stored across runs

• Scoped to the agent or workflow

• Controlled and configurable

Agent memory is not:

• A full conversation log

• Unlimited recall

• A replacement for your knowledge base

Memory complements instructions and knowledge — it doesn’t replace them.

Types of Memory in Sprig

Sprig supports multiple memory types depending on the use case.

1. Session Memory

Temporary memory for a single interaction or workflow run.

Used for:

• Multi-step conversations

• Short-lived tasks

• Guided flows

Cleared automatically when the session ends.

2. Persistent Agent Memory

Longer-term memory attached to an agent.

Used for:

• Remembering user preferences

• Tracking past decisions

• Maintaining continuity across runs

Example:

{
  "memory": {
    "type": "persistent",
    "scope": "agent",
    "retention": "30d"
  }
}

How Memory Is Used at Runtime

When an agent runs, Sprig:

1. Loads relevant memory

2. Applies instructions and guardrails

3. Retrieves external knowledge (if needed)

4. Produces an output

5. Updates memory (if configured)

Memory updates can be automatic or conditional.