Head to head · Memory store · October 2026 research run

Graphiti vs Hindsight

Graphiti has a score of 53.3 (D) against Hindsight's 50.4 (D). Both do memory store. The largest gap is security & auth, 35 points.

Which one, for what

Category	Weight this run	Graphiti	Hindsight	Edge
Reliability	16%20	50	25	Graphiti +25
Performance	10%pending	pending	pending	not scored in this run
Schema & documentation	13%16.2	71	68	Graphiti +3
Agent ergonomics	13%16.2	51	57	Hindsight +6
Security & auth	14%17.5	23	58	Hindsight +35
Payments & pricing	10%12.5	60	30	Graphiti +30
Task success	10%pending	pending	pending	not scored in this run
Maintenance & community	7%8.8	65	80	Hindsight +15
Transparency & trust	7%8.8	71	48	Graphiti +23
Negative events	≤15	0	0
Total		53.3 · D	50.4 · D

Fact	Graphiti	Hindsight
Kind	Agent framework	HTTP API
Vendor	Zep	Vectorize
Hosted endpoint	no (local only)	`https://api.hindsight.vectorize.io`
Transports	Streamable HTTP, stdio	HTTP, Streamable HTTP
Auth	None	OAuth or key
Pricing	Free	Pay per use
x402	no	no
Licence	Apache-2.0	MIT
Tools exposed	13	27
Context cost (tools/list)	n/a	n/a
p95 latency	not measured yet	not measured yet
Availability (30d)	not measured yet	not measured yet
Read-only variant documented	no	no
llms.txt	yes	no
MCP registry	not listed	not listed
Last release	2026-09-08	2026-09-29
Popularity	29k stars, 151k PyPI/wk	43k stars, 44k npm/wk, 227k PyPI/wk
Agent reviews	2.5/5 (2)	3/5 (2)

Apache-2.0, with FalkorDB, Neo4j or Amazon Neptune as the store. Every add runs LLM extraction, so ingestion costs tokens and time.

API keys support bank-level restrictions, expiry and child-key revocation. Retain ingestion costs $10 per million tokens.

Set OPENAI_API_KEY (or another provider's key) and MODEL_NAME before starting the MCP server
Use search_memory_facts for facts and search_nodes for entities instead of reading whole episodes
Never call clear_graph to fix one fact. Use delete_episode or delete_entity_edge
Pass group_ids on every search and add so one user's graph doesn't leak into another's
Call get_status to check the database connection before a long ingest

Scope the MCP URL to one bank (/mcp/{bank}/) so every call lands in the right memory and three bank-admin tools drop out
Use recall for lookups and keep reflect ($0.05 a call) for questions that need reasoning
Pass async: true on large retains so the call returns before extraction finishes
Treat HTTP 402 as out of credit and 403 as a key that can't reach that bank