🚀 cognee Update: August 2025

Hey there, cognee community!

August delivered a set of foundational upgrades across integrations, graph intelligence, and developer experience. Highlights below, with links and docs rolling out alongside releases.

At a Glance

Amazon Neptune integration for managed graph at scale
Community adapters: fixes merged and adapters tested
n8n integration for cogwit to orchestrate memory workflows
Time graphs for temporal relationships and "as-of" queries
Self‑improving graph logic for quality and consistency
Graph embeddings (private preview) for subgraph similarity
Memory pipelines promoted to the core SDK
Ingestion abstraction to register your own custom loaders
New benchmarks: Cognee scores 92.5% on HotPotQA

Amazon Neptune Integration

You can now run Cognee on Amazon Neptune, bringing a fully managed, scalable, and highly available graph backend to production deployments.

Supports typical Cognee graph operations with minimal config
Suitable for larger datasets and always-on workloads
Cloud-native reliability with AWS operational guarantees

Docs include guidance on VPC access, auth, and schema setup.

Community Adapters: Fixes and Tests

We refactored and stabilized multiple adapters. The latest, tested versions are now published in the community adapters set.

Clear examples and standardized interfaces
Better error handling and more consistent outputs
Easier contributions for new sources

n8n + cogwit

We shipped an n8n integration for cogwit, so you can chain ingestion, processing, querying, and notifications using a visual workflow builder.

Trigger cogwit runs from n8n
Orchestrate memory updates with external systems
Add alerts when jobs finish or data changes

Time Graphs

Graphs now capture temporal context so you can reason about how entities and relationships change over time.

Add timestamps and validity windows to nodes/edges
Run "as‑of" queries to reconstruct past states
Analyze trends and deltas across versions

Self‑Improving Graph Logic

We introduced feedback‑driven routines that continuously improve graph quality.

Automatic consolidation of duplicate entities where safe
Relationship refinement based on confidence and usage signals
Optional rules that learn from corrections and query patterns

Graph Embeddings (Private Preview)

Early access to graph embeddings for similarity over structures—not just text.

Compare subgraphs beyond keyword overlap
Improve retrieval for multi‑hop questions
Private preview with limited access while we refine APIs and performance

Memory Pipelines in the Core SDK

Memory pipelines are now part of the core SDK, simplifying setup and reducing boilerplate.

First‑class APIs for ingestion → processing → graph build → retrieval
Consistent config across local, distributed, and hosted runs
Better defaults; fewer moving parts for common flows

Benchmarks: HotPotQA

Our latest evaluation shows Cognee at 92.5% on HotPotQA. We'll publish a full methodology and comparison breakdown in the docs; early results indicate strong gains on multi‑hop reasoning with graph‑aware retrieval.