yulliwasameur/linera

How yulliwasameur/linera works

LinEra is both a lineage-aware cryptographic-erasure library for LLM-agent memory and a research workflow evaluating memory writing, replay, forgetting, and QA. The main path builds conversation memories, replays them through storage backends with lineage, then measures retrieval and post-forgetting answers; a mem0 adapter is a separate integration branch. Source-index references and README statements support the connections shown; unsampled implementation details are not inferred.

Main entry points: store.py Encrypted SQLite Store · keydag.py Lineage Key DAG · memory.py Memory Backends · build_log.py LLM Memory Writer

Components 20

Memory Experiments

LLM Memory Writer agent/build_log.py
Window Writer agent/build_window.py
LongMemEval Writer agent/build_lme.py
LLM and Embedder agent/common.py
Memory Replay agent/replay.py

Memory Backends

Memory Backends agent/memory.py
Baseline Stores baselines/stores.py
Lineage Capture lf/lineage.py
Encrypted SQLite Store lf/store.py
Encrypted Vector Index lf/vindex.py
mem0 Adapter integrations/mem0_linera.py

Erasure Primitives

Lineage Key DAG lf/keydag.py
Puncturable PRF lf/pprf.py

Evaluation

Post-Forget QA agent/qa.py
mem0 Case Study eval/mem0_e2e.py
Canary Evaluation eval/canary.py
Scale and Radius eval/scale.py
Paper Tables and Figures eval/make_tables.py

Other components

Researcher
Conversation Data

Connections 33

Researcher → LLM Memory Writer: runs phase A README.md
Conversation Data → LLM Memory Writer: provides sessions README.md
LLM Memory Writer → LLM and Embedder: uses models agent/build_log.py
Window Writer → LLM and Embedder: uses models agent/build_window.py
LongMemEval Writer → LLM and Embedder: uses models agent/build_lme.py
LongMemEval Writer → LLM Memory Writer: reuses writer agent/build_lme.py
Memory Replay → LLM and Embedder: uses embeddings agent/replay.py
Memory Replay → Memory Backends: writes and forgets agent/replay.py
Memory Backends → Encrypted SQLite Store: stores objects agent/memory.py
Memory Backends → Encrypted Vector Index: indexes vectors agent/memory.py
Memory Backends → Lineage Key DAG: uses lineage keys agent/memory.py
Memory Backends → Puncturable PRF: uses puncturing agent/memory.py
Baseline Stores → Puncturable PRF: uses puncturing baselines/stores.py
Encrypted SQLite Store → Lineage Key DAG: derives object keys lf/store.py
Encrypted SQLite Store → Puncturable PRF: initializes PRF lf/store.py
Lineage Key DAG → Puncturable PRF: evaluates and punctures lf/keydag.py
Encrypted Vector Index → Lineage Key DAG: derives shard keys lf/vindex.py
mem0 Adapter → Encrypted SQLite Store: stores memories integrations/mem0_linera.py
mem0 Adapter → Encrypted Vector Index: indexes memories integrations/mem0_linera.py
mem0 Adapter → Lineage Key DAG: uses lineage keys integrations/mem0_linera.py
Post-Forget QA → LLM and Embedder: uses models agent/qa.py
Post-Forget QA → Memory Replay: uses replay records agent/qa.py
Post-Forget QA → LLM Memory Writer: reuses writer formats agent/qa.py
mem0 Case Study → mem0 Adapter: evaluates adapter eval/mem0_e2e.py
Canary Evaluation → Encrypted SQLite Store: tests forgetting eval/canary.py
Canary Evaluation → Lineage Capture: tests lineage eval/canary.py
Canary Evaluation → Baseline Stores: compares baselines eval/canary.py
Scale and Radius → Encrypted SQLite Store: measures storage eval/scale.py
Scale and Radius → Encrypted Vector Index: measures indexing eval/scale.py
Scale and Radius → Lineage Key DAG: measures derivation eval/scale.py
Researcher → Post-Forget QA: runs QA README.md
Researcher → mem0 Case Study: runs case study README.md
Researcher → Paper Tables and Figures: generates paper results README.md

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·