Agentic Memory

Memory that understands time.

Agentic Memory is the first agent memory system built on a time-aware knowledge graph. Search across code, research, conversations, and git history — with temporal validity as a first-class citizen.

Temporal GraphRAG

Ask what was true. Not just what is true.

Every relationship in Agentic Memory carries a validity window. Query the past. Track contradictions. Retrieve with confidence.

search_all_memory(query, as_of="2026-04-01")
Temporal GraphRAG

Temporal GraphRAG

Agentic Memory is built on a time-aware knowledge graph. Every relationship carries validity windows, confidence scores, and contradiction tracking — enabling deterministic retrieval at any point in time.

  • Query the graph as it existed at any moment with as_of
  • Track valid_from, valid_to, confidence, and support_count on every edge
  • Temporal Personalized PageRank for time-bounded retrieval
  • SpacetimeDB sidecar for microsecond-precision validity
  • Flag contradictions as claims evolve over time
# Query what was true on April 1st
agentic-memory search "auth logic"   --as-of 2026-04-01T00:00:00Z

# Returns results valid at that exact moment
# with confidence and validity metadata
Code Memory

Code Memory

Agentic Memory builds a structural graph of your codebase — not just text embeddings. It understands files, functions, classes, and imports.

  • Semantic search with contextual prefixing
  • File dependency analysis
  • Blast radius impact detection
  • JIT execution tracing on demand
  • Real-time incremental sync via file watcher
agentic-memory index
agentic-memory search "auth logic"
agentic-memory trace-execution src/app.py:run_checkout
Research Memory

Research Memory

Ingest URLs, PDFs, and research reports into a searchable knowledge base. Schedule recurring research sessions to stay up to date.

  • Ad-hoc URL and PDF ingestion
  • Scheduled research sessions
  • Project-based organization
  • Semantic search across findings
agentic-memory web-ingest https://example.com/paper.pdf
agentic-memory web-search "transformer attention"
agentic-memory web-schedule --project llm-memory --interval 24h
Conversation Memory

Conversation Memory

Store and retrieve past agent/user exchanges. Your AI assistant can finally remember what you decided three conversations ago.

  • Ingest conversation logs
  • Semantic search over past turns
  • Retrieve full session context
  • Explicit memory writes via MCP
agentic-memory chat-ingest conversation.json
agentic-memory chat-search "auth flow decision"
# Or via MCP: add_message(role, content, session_id)
Git Memory

Git Memory

Opt-in git graph adds commit history, author signals, and file-version tracking to the same Neo4j database.

  • File-level commit history
  • Commit context and diff stats
  • Recent risky change detection
  • Hybrid code + git signals
agentic-memory git-init --repo /path/to/repo --mode local
agentic-memory git-sync --repo /path/to/repo --incremental
agentic-memory git-status --repo /path/to/repo --json

search_all_memory(query, as_of=...)

The unified search tool spans all domains in a single MCP call — and it respects time. No more context switching between code docs, chat history, and research notes.

Explore the API