Attack on Memory

Memory governance for multi-agent systems.

English | 简体中文

Why this project

As agents move from single-turn chat to long-running collaboration, the bottleneck shifts from model size to memory quality:

• Trustworthy memory (verifiable, traceable)
• Controllable memory (role-aware disclosure, revocable)
• Evolvable memory (inheritance, rollback, branch comparison)

Attack on Memory is a protocol-oriented memory layer for multi-agent systems: Graph Retrieval + Time-window Retrieval + Governance + BranchWorldModel.

3-minute quickstart

# 1) unit tests
PYTHONPATH=src python3 -m unittest discover -s tests -v

# 2) scenario validation
PYTHONPATH=src python3 examples/validate_scenarios.py

# 3) simulation demo
PYTHONPATH=src python3 examples/simulation_runner.py

What you get in v0.1

• Domain model and runtime memory packet design
• Governance policies (selective disclosure, emergency controls)
• Scenario-driven simulation and validation
• OpenClaw adapter for context injection + writeback
• CI + Scorecard + Dependabot + contribution governance

Reproducible benchmark snapshots

• Baseline report: docs/benchmarks/v0.1-baseline.md
• Snapshot report: docs/benchmarks/v0.1-benchmark-snapshot.md
• Raw latest results: docs/benchmarks/latest-results.json
• Snapshot generator: scripts/generate_benchmark_snapshot.py

Architecture & docs

• Architecture: docs/architecture.md
• Scenario spec: docs/scenario-spec.md
• Experiments and value: docs/experiments-and-value.md
• OpenClaw integration: docs/openclaw-integration.md
• Threat model: docs/threat-model.md
• FAQ: docs/FAQ.md
• Starter contribution walkthrough: docs/STARTER_PR_WALKTHROUGH.md

Project structure

• src/attack_on_memory/domain/ – core models
• src/attack_on_memory/application/ – retrieval and orchestration
• src/attack_on_memory/governance/ – policy layer
• src/attack_on_memory/runtime/ – runtime adapters and context
• src/attack_on_memory/infrastructure/ – in-memory and backend adapters
• examples/ – scenarios, validation, simulation runner
• tests/ – unit tests and adapter wiring tests

Security and governance

• Security policy: SECURITY.md
• Code of conduct: CODE_OF_CONDUCT.md
• Contribution guide: CONTRIBUTING.md

Roadmap (next)

• Replay-derived benchmark (beyond scenario-derived baseline)
• Pluggable vector index adapter
• Pluggable graph backend adapters
• Stronger anti-poisoning governance and observability

License

MIT