Files
papers/lux-agent-sovereignty/sections/04-agent-memory.tex
zeekay cb3b572a9c papers: M/F/Z-Chain canonical naming per LP-134
Add LP-134 canonical naming banner to 46 LaTeX papers that reference
T-Chain in MPC/FHE context. Banner clarifies the M-Chain (MPC ceremonies
for external-wallet bridge custody) and F-Chain (FHE compute) split.

LP-020 quasar-consensus paper already correctly describes the M/F-Chain
split and is unmodified.
2026-05-05 16:28:50 -07:00

49 lines
2.7 KiB
TeX

%% ===================================================================
%% LP-134 canonical naming (2025-12-15):
%% Per LP-134 (Lux Chain Topology), T-Chain has been split into:
%% M-Chain — MPC ceremonies for bridge custody of EXTERNAL wallets
%% (CGGMP21, FROST, Pulsar-general)
%% F-Chain — FHE compute + TFHE bootstrap-key generation (encrypted EVM)
%% The legacy name "T-Chain" is retained ONLY for `teleportvm` (LP-6332).
%% Where this paper says "T-Chain MPC" / "T-Chain threshold governance" /
%% "T-Chain TFHE" / "T-Chain custody", read as M-Chain (MPC) or F-Chain (FHE).
%% ===================================================================
\section{Agent Memory}
\label{sec:memory}
Agent memory is the persistent, private knowledge store that distinguishes a sovereign agent from a stateless API call. The memory system has three tiers.
\subsection{Local Vault Memory}
The primary memory store is the encrypted SQLite vault (Section~\ref{sec:capsule}). The agent writes structured key-value pairs, conversation histories, learned preferences, and intermediate reasoning artifacts. All writes are append-only with content-addressed keys:
\begin{equation}\label{eq:memory-key}
k = H(\text{namespace} \| \text{content})
\end{equation}
where $H$ is SHA-256. This ensures deduplication and enables efficient retrieval by content hash.
\subsection{CRDT-Synchronized Memory}
When an agent runs across multiple instances (e.g., horizontal scaling, migration between execution environments), memory consistency is maintained via Conflict-free Replicated Data Types (CRDTs)~\cite{shapiro2011crdts}. We use two CRDT structures:
\begin{itemize}
\item \textbf{LWW-Register} (Last-Writer-Wins Register) for mutable configuration values.
\item \textbf{OR-Set} (Observed-Remove Set) for the agent's knowledge base, where entries can be independently added and removed across instances.
\end{itemize}
Synchronization occurs over encrypted channels between instances of the same capsule, authenticated by the capsule's DID. Cross-capsule memory is never synchronized---each capsule is a sovereignty boundary.
\subsection{FHE-Encrypted Memory}
For agents performing confidential reasoning---e.g., processing medical records, financial data, or proprietary strategies---the memory system supports Fully Homomorphic Encryption (FHE) using the CKKS scheme~\cite{cheon2017ckks}. The agent encrypts memory entries with its FHE public key:
\begin{equation}\label{eq:fhe-memory}
\tilde{m} = \text{CKKS.Enc}(\text{pk}_{\text{FHE}}, m)
\end{equation}
Encrypted memory can be processed by external compute nodes without decryption. The agent (or a threshold set of authorized parties via T-Chain) holds the decryption key.