Agentic Memory
Temporal GraphRAG — now in beta

Memory that understands time.

Agentic Memory is the first agent memory system built on a time-aware knowledge graph. Query what was true last week. Track how claims evolve. Retrieve with confidence intervals — all exposed via MCP.

Get Started FreeExplore Features
Works with
ClaudeCursorChatGPTWindsurfCodex
terminal — zsh
$ agentic-memory init
✅ Initialized in /my-project
$ agentic-memory search "auth logic" \
--as-of 2026-04-01T00:00:00Z
Found 3 relevant code result(s):
1. authenticate (src/auth.py:authenticate)
[valid: 2026-03-15 → 2026-04-02 | confidence: 0.98]
def authenticate(username, password):
v1.2 shippedT-3d
Auth refactoredT-2d
Bug introducedT-1d
Query nowNow
as_of retrieval

Ask what was true. Not just what is true.

Most RAG systems dump documents into a vector database and hope for the best. Agentic Memory treats every relationship as a time-bounded claim with validity windows, confidence scores, and contradiction tracking.

  • Query the graph as it existed at any point in time
  • Track confidence, support, and contradictions on every edge
  • Temporal Personalized PageRank for deterministic retrieval
  • SpacetimeDB sidecar for microsecond-precision validity

Four memory domains. One unified graph.

Search across all domains in a single query — with time as a first-class dimension.

Temporal GraphRAG

Time-aware graph layer powered by SpacetimeDB. Query what was true at any point in time with deterministic temporal retrieval.

Code Memory

Structural understanding of files, entities, imports, and on-demand execution tracing — not just text similarity.

Research Memory

Ingest URLs, PDFs, and research reports as searchable findings. Schedule recurring research sessions.

Conversation Memory

Stores and retrieves past agent/user exchanges by semantic similarity. Never lose context between sessions.

From blank slate to persistent memory in minutes

1

Install

pipx install agentic-memory

2

Initialize

agentic-memory init — interactive wizard guides Neo4j and embedding setup.

3

Search

Query across all memory domains — even as they existed last week.

Hybrid GraphRAG + Temporal PPR

Structure first. Vectors second. Time always.

Agentic Memory combines durable graph relationships with semantic embeddings and a SpacetimeDB temporal sidecar. The result is retrieval that understands context and chronology.

  • Neo4j graph database with vector search
  • SpacetimeDB temporal sidecar for validity-aware edges
  • Tree-sitter parsing for language-native structure
  • JIT execution tracing on demand
  • Real-time file watcher for incremental sync
Repository
Ingestion
Neo4j Graph
Temporal Sidecar
MCP Server
AI Clients

Ready to give your agent a memory?

Join the managed hosted beta or run the full stack yourself. Either way, your agents stop starting from zero.

Start Free TrialSelf-Host Guide