AgentMemory.Abstractions 1.0.0

Agent Memory for .NET

Independent community project. Not affiliated with, endorsed by, or supported by Neo4j, Inc.

Persistent graph-native memory for AI agents in .NET, backed by Neo4j and built for Microsoft Agent Framework, Semantic Kernel, direct .NET usage, and MCP clients.

Start here:

• CONTINUE-HERE.md - 30-second current-state landing.
• docs/ROADMAP.md - current release status, shipped capabilities, and next work.
• docs/core/ - canonical project philosophy, requirements, design, specification, ADRs, and summaries.
• docs/getting-started.md - install and first-run guide.

Status

1.0.0 is the first stable release: the public API surface is locked under Semantic Versioning. The library is feature-complete — three-tier graph memory (short-term / long-term / reasoning), bitemporal recall with non-destructive decay, multi-tenant owner isolation, GraphRAG retrieval, consolidation/hygiene, and first-class adapters for the Microsoft Agent Framework, Semantic Kernel, MCP, and direct .NET use. It passes the upstream agent-memory compatibility kit (TCK) through the Gold tier.

The library ships from a warning-free Release build with a full unit + Semantic Kernel + live-Neo4j integration suite green on main.

What It Provides

Agent Memory for .NET provides three memory layers:

• Short-term memory: conversations, messages, ordering, sessions, roles, timestamps, and embeddings.
• Long-term memory: entities, facts, preferences, relationships, provenance, owner scope, confidence, and temporal state.
• Reasoning memory: traces, steps, tool calls, tool aggregate nodes, task embeddings, and prior execution patterns.

It supports:

• Neo4j-backed graph persistence.
• Vector, fulltext, hybrid, and graph traversal retrieval.
• Optional GraphRAG context retrieval from AgentMemory.Neo4j.
• Microsoft Agent Framework context, chat-store, tools, and trace-recording integration.
• Semantic Kernel plugin integration.
• Model Context Protocol server surface.
• LLM and Azure Language extraction backends.
• OpenTelemetry observability, geocoding/entity enrichment, optional GDS analytics, schema bootstrap/migration, and CLI maintenance workflows.

Package Topology

Quick Start

dotnet add package AgentMemory

using AgentMemory;
using AgentMemory.Abstractions.Services;
using AgentMemory.Neo4j.Infrastructure;
using Microsoft.Extensions.DependencyInjection;

var services = new ServiceCollection();

services.AddNeo4jAgentMemory(
    memory => { },
    neo4j =>
    {
        neo4j.Uri = "bolt://localhost:7687";
        neo4j.Username = "neo4j";
        neo4j.Password = "password";
    },
    configureLlm: null);

var provider = services.BuildServiceProvider();

var bootstrapper = provider.GetRequiredService<ISchemaBootstrapper>();
await bootstrapper.BootstrapAsync();

var memory = provider.GetRequiredService<IMemoryService>();
await memory.AddMessageAsync("session-01", "conversation-01", "user", "I prefer dark mode.");

For production semantic search, register a real IEmbeddingGenerator<string, Embedding<float>> from Microsoft.Extensions.AI; the core stubs are safe defaults, not production embeddings.

Isolation Model

Memory is scoped as store → owner → session.

• Store: ApplicationId, default shared database, optional database per application for Enterprise/AuraDB.
• Owner: owner_id, owner_key, and MemoryScope; null owner means shared/global.
• Session: session_id and conversation IDs for run-local context.

Owner and store scope are enforced in repository, recall, GraphRAG, reasoning, maintenance, and temporal paths rather than only in adapters.

Relationship to Upstream Projects

This project is inspired by neo4j-labs/agent-memory and preserves compatible graph concepts where useful. It is implemented independently in .NET and is not an official Neo4j repository or product.

GraphRAG retrieval is implemented inside AgentMemory.Neo4j and registered explicitly with AddGraphRagAdapter(...); there is no separate current GraphRAG adapter package.

Documentation

Full documentation lives in docs/. The canonical core set is docs/core/:

• Philosophy
• Requirements and Constraints
• Design Document
• Specification
• ADRs
• Summaries

Contributing

See CONTRIBUTING.md for build, test, and contribution guidance.

License

This project is licensed under the MIT License.