prompting · recursive-summarization TEC-0017

Maintain a running summary

Periodically fold the conversation into a summary that replaces older turns.

Every few turns, have the model update a running summary with new facts and decisions, then drop the raw turns it covers. The summary stays in context; the details are archived and retrievable.

Addresses: Remembering across sessions

Contexts: Chat assistant, Customer support

Does it work?

argued, not measured2 supporting · 0 contesting sources · last moved 2026-09-04

Efficacy claims — what this technique actually moves, under which conditions, and whether that has been contested.

Needs: the unsummarised record kept retrievable

Cost: low

Fails when:

The need is exact recall of a specific earlier turn. The summary is lossy by construction, so the raw turns must stay retrievable alongside it. (helps, narrowly)

mechanismmechanism reasoning
A running summary buys continuity across a long conversation at the cost of specifics, because it is lossy compression by construction — so it helps with "what were we doing" and not with "what exactly did I say in turn 40." Keeping the raw turns retrievable alongside the summary covers the second case.
- supports Recursively Summarizing Enables Long-Term Dialogue Memory in Large Language Models (2023)
- supports LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (2024)

Code

No repository linked yet. Contribute one.

Sources

Recursively Summarizing Enables Long-Term Dialogue Memory in Large Language Models (2023)

Either way, bring evidence: a link, or what you saw.

Whether a technique works is a claim. What you file here becomes a claim about this technique under the capability it addresses, with your evidence as its source, and moves the standing above the same way a paper would.