---
title: "ADR-005: TYPO3 Caching Framework Integration"
manual: "TYPO3 LLM Extension"
version: "0.35"
permalink: "https://docs.typo3.org/permalink/netresearch/nr-llm:adr-005@0.35"
source: "Adr/Adr005Typo3CachingFrameworkIntegration.rst"
modified: "2026-09-16T22:09:16+00:00"
---

# ADR-005: TYPO3 Caching Framework Integration

-   *Status:* Accepted
-   *Date:* 2024-03
-   *Authors:* Netresearch DTT GmbH

## Context

LLM API calls are:

-   Expensive (cost per token).
-   Relatively slow (network latency).
-   Often deterministic (embeddings, some completions).

## Decision

Integrate with **TYPO3's caching framework**:

-   Cache identifier: `nrllm_responses`.
-   **No backend specified** — TYPO3 uses the instance's
    default cache backend (respects Redis/Valkey/Memcached).
-   Cache keys based on: provider + model + input hash.
-   TTL: 3600s default (configurable).
-   Cache group: `nrllm` (flush via
    `cache:flush --group=nrllm`).

Caching strategy:

-   **Always cache**: Embeddings (deterministic).
-   **Optional cache**: Completions with temperature=0.
-   **Never cache**: Streaming, tool calls, high temperature.

## Consequences

**Positive:**

-   ●● Reduced API costs.
-   ●● Faster responses for cached content.
-   ● Follows TYPO3 patterns.
-   ◐ Configurable per deployment.

**Negative:**

-   ✕ Cache invalidation complexity.
-   ◑ Storage requirements.
-   ✕ Stale responses if TTL too long.

**Net Score:** +4.5 (Positive impact - significant
cost/performance gains with manageable cache
complexity)