References: How Clawless remembers (and forgets)

Source material

The lesson is anchored on one chapter from the Clawless knowledge base, the single source of truth for Clawless behavior. The KB is reachable inside the app: click the Help button above the message box and ask the knowledge base by name; answers cite the docs chapter they come from. The product itself lives at clawless.ai.

Clawless KB chapter cited:

PRIMARY (everything about the memory system, the four tiers, the panel,
the classifier, and the settings)
• Clawless Knowledge Base, "Memory System" chapter
  Covers the conversation-history vs memory distinction, the four-tier
  taxonomy (Pinned / Insights / General / Decayed), the classifier that
  watches conversations for memory candidates, the three pathways to
  add a memory, the Memory panel controls (pin, edit, delete, source
  badge, last-seen date, search), the shared-pool default across
  agents, and the Settings, Memory section (auto-extraction toggle,
  re-injection interval, capacity-pruning notification, decay rules).

Clawless documentation is open under Creative Commons Attribution-ShareAlike
(CC BY-SA 4.0). This lesson follows the KB's framing on memory architecture,
expressed in original prose for a Clawdemy reader opening the Memory
panel for the first time.

Going deeper inside Clawless

Two follow-on KB chapters that touch what memory does and does not cover.

• Clawless Knowledge Base, “Settings Reference” chapter. The full list of Settings panels. The Memory section is documented in detail, including each of the four settings this lesson summarized (auto-extraction, re-injection interval, capacity-pruning notification, decay rules). Useful if you want to tune the defaults rather than leave them.

• Clawless Knowledge Base, “What’s Under the Hood” chapter. The architectural framing for where data lives in Clawless. The “memories live on your computer, but travel to the provider with each message” rule in this lesson is one specific case of the broader local-first architecture that chapter covers in full. Read it when you start the privacy track.

Going deeper outside Clawless

Two outside sources for readers who want context on how AI memory systems work in general and on the cognitive idea of forgetting-as-fading.

• “MemGPT: Towards LLMs as Operating Systems” (Packer et al., UC Berkeley, 2023). The original research paper on a layered memory system for large language models. The four-tier shape Clawless uses (Pinned, Insights, General, Decayed) is in the same family as the multi-tier memory hierarchy MemGPT proposes. Read this if you want the technical background on why AI products are moving away from “the model remembers the whole conversation” toward layered memory with promotion and decay.

• “The Curve of Forgetting” (Hermann Ebbinghaus, 1885, ongoing research; Wikipedia summary). The classical psychological account of how human memory fades over time. The Clawless Decayed tier is engineered to mimic this curve rather than implement a hard delete. Useful background for the “forgetting without deleting” framing in the lesson body.

Adjacent lessons on Clawdemy

Where this sits in the track.

• Your first conversation and picking a model (lesson 2). Memory ships along with each message to whichever model is currently selected in the picker. The relationship between memory and the model picker becomes practical once you switch models mid-conversation; the same memories travel with both.

• API keys and the OAuth path (lesson 3). The classifier that auto-extracts memories runs server-side on whichever AI provider you have selected, which means classification quality depends on the model. Worth knowing when you switch providers and notice your Insights tier looking different.

• CostGuard and the local-first privacy posture (a later lesson). The privacy rule in this lesson (memories live on your computer but travel with messages to the provider) is one specific case of the broader local-first architecture that lesson covers in detail. Read together.