This is NOT a simple collection of tools.
SharedMemoryServer demonstrates a production-grade infrastructure designed to solve the two biggest bottlenecks in Agentic Workflows: Latency and Knowledge Fragmentation.
If you are evaluating for roles like AI Architect, Tech Lead, or LLM System Designer, this project serves as verified proof of:
- Systematic Architecture Design: Decoupling compute from transactions.
- Data Integrity & Consistency: Multi-agent atomic operations.
- Actionable Value Quantification: Second-Gen Insight Engine (Knowledge Maturity).
Important
"Compute-then-Write" Pattern
We solved the SQL lock contention problem by moving expensive LLM operations outside the database transaction.
graph TD
subgraph "Parallel AI Compute (Data Plane)"
A[Agent Request] --> B1[Gemini Embeddings]
A --> B2[Conflict Detection]
end
B1 & B2 --> C{Orchestrator}
subgraph "Atomic Sync (Control Plane)"
C --> D[SQLite Transaction]
C --> E[Memory Bank MD]
end
D & E --> H[Response Success]
- Lock Contention: Reduced DB lock duration from ~2000ms to <50ms by computing embeddings outside transactions.
- Agent Density: Verified to support 3-5 simultaneous agents performing complex read/write operations in ~1.36 seconds total.
- Atomic Mirroring: Ensures Knowledge Graph (DB) and Memory Bank (Markdown) are always in sync.
๐ Deep Dive into Architecture (Ayato Studio Portal)
Unlike typical RAG systems, SharedMemoryServer measures Knowledge Maturity. We don't guess ROI; we observe the physical transfer of knowledge across sessions.
- Knowledge Age Transfer:
Long-term (24h+) Assetsare reused across session boundaries, proving long-term value. - Search Precision: Average similarity scores of
0.85+, ensuring agents never hallucinate on core documentation. - Reuse Multiplier: Every byte of knowledge is utilized 4.2x on average across different tasks.
What you can evaluate from this specific codebase:
- Concurrency Design: See how
AsyncSQLiteConnectionandGlobal File Lockprevent data corruption in multi-agent environments. - Layered Decoupling: Observe the separation between
Agent Core(Runtime) andAdmin Server(Maintenance). - Professional Lifecycle: Look at the 3-tier testing suite (Unit, Integration, System) ensuring 100% reliability of the logic layer.
We strictly separate Agent reasoning from System administration to ensure safety and prevent cognitive overload.
The primary tools used by AI agents during task execution.
read_memory: Hybrid semantic + keyword search across the Graph and Bank.save_memory: Atomic update for both structured entities and markdown documentation.synthesize_entity: Aggregates distributed information into a coherent master summary.sequential_thinking: Context-aware reflective problem-solving tool.
Infrastructure tools for system integrity (Separated from standard agent access).
admin_get_audit_history: Audit logs for all memory changes.admin_rollback_memory: Revert specific changes via Audit ID.admin_create_snapshot: Create point-in-time database backups.admin_repair: Reconstruct physical workspace files from DB mirroring.
uv pip install -e .uv run shared-memory # Start Agent Server
uv run shared-memory-admin # Start Admin Serveruv run shared-memory-register # Register with Cursor/Claude- Local-First Architecture: Your IP never leaves your system.
- Principle of Least Privilege: Agent tools cannot invoke destructive admin rollbacks.
Licensed under the PolyForm Shield License 1.0.0. For commercial SaaS use, please contact Ayato Studio.
Built to elevate AI Agents from "Simple Assistants" to "Systematic Thinking Assets".