Kaelia — Operating Memory
Every session inherits everything that came before it.
Operating Memory is the structured context layer at the core of Kaelia. It is not a transcript or a document store — it is an organized, persistent model of your business designed specifically to be reasoned from.
Structure vs. Storage
Many AI tools offer document upload or conversation history. These are retrieval systems — they find content when asked. Operating Memory is a reasoning system — it maintains context in a form that can be reasoned from without a specific query.
The distinction is critical. A retrieval system surfaces the document you asked for. A reasoning system identifies the assumption that underlies three decisions made across six weeks, without being asked to look for it.
Operating Memory is organized into four structured categories — not arbitrary files, not raw conversation logs, not a search index. Each category is designed to hold a specific type of operational context that the Decision Layer uses differently.
Memory Categories
Four structured categories. Each reasoned from differently.
Decisions
Definition
A directional choice made by the operator, with its stated reasoning, the options that were considered, and the expected outcome at the time of the decision.
How It's Used
Decisions are the primary anchor for reasoning. When the Decision Layer evaluates current context, it cross-references against prior decisions to identify conflicts, confirming signals, or outdated positions.
Decay & Weight
Decisions do not expire, but their weight in reasoning adjusts as their outcomes become known or their underlying assumptions are invalidated.
Signals
Definition
Quantitative or qualitative inputs observed from the business environment — client feedback, market movement, team observations, competitive intelligence, financial data.
How It's Used
Signals are analyzed for pattern and trend over time. A single signal is a data point. Repeated signals in the same direction are an indicator. Signals that conflict with active assumptions are flagged.
Decay & Weight
Signal weight decreases over time unless reinforced by subsequent signals. Old signals that have not been validated remain in memory but are noted as aged.
Assumptions
Definition
A belief that underlies one or more active decisions — something the operator is treating as true but has not fully validated. Assumptions are the hidden input in most decision-making.
How It's Used
Assumptions are tracked continuously. When new signals arrive, they are cross-referenced against active assumptions to identify validation or contradiction. Unvalidated assumptions are surfaced in the Intelligence Brief.
Decay & Weight
Assumptions that have not been reviewed in over 30 days are flagged as stale. Assumptions that underpin high-weight decisions are surfaced more aggressively.
Workstreams
Definition
An active initiative — a project, process, or area of focus with defined intent, current state, open questions, and responsible ownership.
How It's Used
Workstreams provide the current operational context. The Decision Layer uses workstream state to assess how recommendations interact with active initiatives and where conflicts or dependencies exist.
Decay & Weight
Stalled workstreams are flagged when no progress signals are received within an operator-defined time window. Completed workstreams are archived but remain accessible for reasoning.
Operator Access
Operating Memory is not a black box. Operators can view, edit, and delete any entry in their operating model at any time. Every piece of context is operator-owned and operator-controlled. Kaelia provides the structure and the reasoning capability — the data itself belongs entirely to the operator.
Operators can also export their full operating model in structured format — a complete representation of every decision, signal, assumption, and workstream logged during their account lifetime.
Part of a System
Operating Memory is the foundation that makes the other two Kaelia components work. The Decision Layer reasons from it. The Intelligence Brief is generated from it. Without accumulated, structured memory, both produce generic outputs. With it, they produce outputs that are specific to your business and its history.