Sessions Graph

Sessions Graph is the Context Graph component for session context and cross-session recall. It is the authority on (:Session) nodes in the Context Graph family. It stores free-form text assertions — called Memories — written explicitly by agents, and makes them searchable in future sessions.

Requires Memgraph ≥ 3.6 (text search is stable from that release).

Installation

pip install sessions-graph

To use with Agent Context Graph:

pip install sessions-graph[agent-context-graph]

To use session reconciliation (entity extraction from session content):

pip install sessions-graph[reconciliation]

Quick start

from sessions_graph import SessionsGraph

graph = SessionsGraph()  # connects via MEMGRAPH_HOST / MEMGRAPH_PORT env vars
graph.setup()  # creates constraints and the text index (run once)

# Write a memory
mem = graph.save_memory(
    user_id="alice",
    content="Prefers Python over TypeScript",
    session_id="s-abc123",  # optional — links memory to a session for provenance
)

# Retrieve all memories for a user
memories = graph.get_memories("alice")

# Search memories by content (full-text, powered by Tantivy)
results = graph.search_memories("alice", "Python")

# Update or delete
graph.update_memory(mem.memory_id, "Prefers Python, especially for data tooling")
graph.delete_memory(mem.memory_id)

Integration with Agent Context Graph

Wire the SessionsGraphConnector into an AgentLink to get automatic session provenance — the connector tracks the active session_id and user_id from SessionStartEvent so you can reference them when saving memories.

from sessions_graph import SessionsGraph
from sessions_graph.connector import SessionsGraphConnector
from agent_context_graph import AgentLink
from agent_context_graph.adapters.claude import ClaudeAdapter

graph = SessionsGraph()
graph.setup()

connector = SessionsGraphConnector(graph)
link = AgentLink()
link.add_connector(connector)

adapter = ClaudeAdapter(
    link,
    session_id="s-abc123",
    session_kwargs={"user_id": "alice"},
)

# During the session, save memories via the Python API:
graph.save_memory(
    user_id=connector.active_user_id,
    content="User works primarily in the ai-toolkit repository",
    session_id=connector.active_session_id,
)

Graph schema

(:User {user_id})
    └─[:HAS_MEMORY]─▶ (:Memory {memory_id, user_id, content, created_at})
                              ▲                        │
              [:PRODUCED_MEMORY]              [:HAS_CHUNK]
                              │                        ▼
                      (:Session {session_id,   (:Chunk {hash, text})
                                 reconciliation_status,     ▲
                                 reconcileed_at})  [:HAS_CHUNK]
                              │                        │
                        [:HAS_ACTION]                  │
                              ▼                        │
                        (:Action) ─────────────────────┘
                              │
                                              (:Entity)-[:MENTIONED_IN]->(:Chunk)

(:Action) is owned by Actions Graph; (:Chunk) and the extracted entity nodes are owned by unstructured2graph. See Session reconciliation below for how they get linked.

Text search

Sessions Graph uses Memgraph text search (powered by Tantivy) for search_memories. The text index is created on setup():

CREATE TEXT INDEX memory_content_index ON :Memory(content);

Searches run as:

CALL text_search.search_all('memory_content_index', 'Python')
YIELD node AS m, score
WHERE m.user_id = 'alice'
RETURN m.content, score
ORDER BY score DESC
LIMIT 10;

The query string follows Tantivy query syntax.

Session reconciliation

A session's Actions Graph content (Messages, ToolCalls, ToolResults) and Memories are mostly opaque text today. Session reconciliation runs that content through unstructured2graph's chunk + LightRAG entity-extraction pipeline, turning it into queryable graph entities linked back to the session that produced them — see CONTEXT.md for the Session Reconciliation / Reconcilable Content / Reconciliation Status terminology.

This requires the sessions-graph[reconciliation] extra and an LLM API key (OPENAI_API_KEY or ANTHROPIC_API_KEY) for LightRAG — see the lightrag-memgraph README.

Reconciliation never runs inside the SESSION_END hook itself. LightRAG entity extraction is LLM-backed and slow, and hook runtimes (Claude Code, Codex) enforce a timeout on hook commands. Instead:

• On SESSION_END, SessionsGraphConnector cheaply marks the session reconciliation_status = 'pending' — no LLM calls, safe inside the hook.

• The actual reconciliation run happens out-of-band, via the CLI:

  # Reconcile one session
  sessions-graph reconcile --session s-abc123
  
  # Sweep every session still marked 'pending' (e.g. from cron)
  sessions-graph reconcile --pending --limit 50
  

• Or, if you want it triggered automatically without a manual/cron step, opt in to a best-effort detached background process spawned right after a session ends: pass SessionsGraphConnector(graph, auto_reconcile=True), or set SESSIONS_GRAPH_AUTO_RECONCILE=1 in the environment the connector is constructed in (this is what hook-based runtimes read, since they don't expose a constructor kwarg for it). This is fire-and-forget — if the process dies before finishing (machine sleep, crash), the session stays pending and sessions-graph reconcile --pending is the reliable backfill.

Programmatically:

from sessions_graph import SessionsGraph
from lightrag_memgraph import MemgraphLightRAGWrapper

graph = SessionsGraph()
graph.setup()

lightrag_wrapper = MemgraphLightRAGWrapper()
await lightrag_wrapper.initialize(working_dir="./lightrag_storage")

summary = await graph.reconcile_session("s-abc123", lightrag_wrapper=lightrag_wrapper)
print(summary.status, summary.texts_considered, summary.texts_deduped)

Extracted entities land in the same LightRAG workspace as any documents ingested via unstructured2graph by default, so a person or concept mentioned both in a session and in an ingested document merges into one node. Pass entity_workspace= explicitly to reconcile_session() to isolate them instead.

Content is deduplicated by hash before ever reaching the LLM, so re-running a sweep over already-processed content never re-bills it — but session content is pulled in full, including tool output, so the first run over a chatty session can still be substantial. Consider this before enabling auto_reconcile broadly.

API reference