Cheatsheet: How Clawless remembers (and forgets)

The distinction worth keeping clean

Layer	What it is	Where it lives	Travels?
Conversation history	The literal back-and-forth of one chat	Inside that one conversation	No
Memory	Distilled facts pulled out of conversations over time	The Memory panel	Yes, across every conversation and every agent

The four tiers (most-important to least)

Tier	What it is	How memories get here	Behavior
Pinned	Facts you explicitly told Clawless to always remember	You Pin them	In every prompt; hard size cap around 2,200 characters total
Insights	Classifier inferences confident enough to surface routinely	Classifier auto-extracts	Surfaced regularly; promotable to Pinned with a click
General	Everything else the classifier noticed but is unsure about	Classifier auto-extracts	Included only when relevant to the current message
Decayed	Old memories that have aged out of routine use	Auto-aging from disuse	Surfaces only if current context makes them look very relevant again

Decayed in one phrase

Forgetting without deleting. Memory still exists. Memory still visible in the panel. Memory does not get fed into new conversations unless something current makes it look very relevant. Mimics human forgetting (stop accessing) instead of computer deletion (data gone).

Three pathways memories get in

Pathway	What you do	Where the memory lands
Classifier auto	Nothing. You just talk.	Insights (high confidence) or General (lower confidence)
Ask explicitly	Tell any agent “remember that I prefer…” then confirm the candidate	Wherever the engine sorts it; you can pin after
Add manually	Memory panel, click Add Memory, type the fact, save	Unpinned by default; pin it if load-bearing

The Memory panel (where you control what is kept)

Control	What it does
Pin or unpin	Move into or out of the Pinned tier
Edit	Rewrite the text (useful when the classifier saved something slightly wrong)
Delete	Remove permanently
Source badge	Shows which conversation the memory was extracted from
Last-seen date	The last time the memory was used or referenced (old dates flag Decayed candidates; new dates confirm something is in use)
Search box	Filter across all tiers by keyword

Open the panel from the Memory icon on the navigation rail on the far left.

Sharing across agents (the default)

All agents on a single Clawless install share one memory pool. Tell the Assistant your name, the Writer knows it. The Researcher knows it.

Per-agent memory isolation (one agent forgets while another remembers) is on the roadmap. At launch, all agents on the install share the same pool.

When you switch agents mid-task, the new agent receives a short briefing of what was happening in the other agent’s conversation. Like handing off to a colleague who has not been in the room but gets a one-line summary.

Privacy in one rule

Question	Answer
Where do memories live?	On your computer, in a local Clawless data file. Not on any Clawless server.
Where do memories travel?	To the AI provider with each message they are relevant to.
The practical rule	If there is a fact you never want leaving your computer, do not save it as a memory.

Memory is for things that are useful for the agent to know; “useful for the agent to know” means “shared with the AI provider every time it might be relevant.”

Settings worth knowing about

Setting	Default	What it does
Auto-extraction	On	Whether the classifier scans new conversations for memory candidates. Turning off does not delete existing memories; just stops new auto-additions.
Re-injection interval	Every ten messages	How often Insights are re-fed into a long conversation to keep the agent grounded. Lower = grounded but costlier; higher = cheaper but risk of drift.
Capacity-pruning notification	Off	Whether to notify you when General memories are pruned. Turn on for visibility.
Decay rules	Reasonable defaults	How long an unused memory takes to age down toward Decayed. Tighten or loosen if needed.

Open Settings, gear icon at the bottom of the navigation rail, then Memory section in the sidebar.

Pitfalls to dodge

• Treating memory and conversation history as the same thing (they are not; history stays inside a chat, memory travels)
• Saving secrets as memories (they travel to the provider with messages; not local-only)
• Cramming everything into Pinned (the cap is intentional; consolidate or move to Insights)
• Expecting per-agent isolation today (shared pool at launch; isolation on roadmap)
• Editing a memory and expecting yesterday’s replies to update (they were generated with yesterday’s state)
• Ignoring the Insights tier (review once early; the inferences may need cleanup or promotion)

Worth opening once

The Memory panel, in the first week. Read the Insights tier, clean up one or two that landed slightly off, pin the one or two facts that should always travel. After that, memory mostly runs itself.

What lands next

A later lesson covers Channels and integrations: how Clawless reaches outside the chat (email, Slack, calendars). A later lesson covers CostGuard and the local-first privacy posture in depth: the cap that prevents a runaway BYOK spend, and the architectural reason your data does not pass through Clawless. Everything else in the track assumes you have a working memory pool. That is what this lesson left you with.