Why CognoDB

The differences, stated without hand-waving.

No unsourced multiples, no “fastest” anything. Four concrete differences, each verifiable — and a footnote for every number.

74,314/s
Read queries1
0.27 ms
2-hop p952
45,029/s
Ingest edges3
80 B
On-disk / edge4
  1. 1Read-only throughput — 40 concurrent clients replaying a seeded workload for 60 s. The mixed run (10% writes) sustained 62,189/s. July 2026 graph benchmark: SNAP soc-Pokec (1.63 M nodes, up to 30.6 M edges) on one n2-standard-8 (8 vCPU / 32 GiB), CognoDB on defaults, median of 3 reps after warm-up.
  2. 2p95 latency for a two-hop, friends-of-friends traversal against the loaded graph. July 2026 graph benchmark: SNAP soc-Pokec (1.63 M nodes, up to 30.6 M edges) on one n2-standard-8 (8 vCPU / 32 GiB), CognoDB on defaults, median of 3 reps after warm-up.
  3. 3Edges merged per second over a fixed 60 s window (CognoDB wrote 2.70 M edges in its minute). July 2026 graph benchmark: SNAP soc-Pokec (1.63 M nodes, up to 30.6 M edges) on one n2-standard-8 (8 vCPU / 32 GiB), CognoDB on defaults, median of 3 reps after warm-up.
  4. 4On-disk store size divided by edges loaded, read from the container's cgroup counters — a compact store keeps the working set cached. July 2026 graph benchmark: SNAP soc-Pokec (1.63 M nodes, up to 30.6 M edges) on one n2-standard-8 (8 vCPU / 32 GiB), CognoDB on defaults, median of 3 reps after warm-up.

Fast reads, low latency, compact store.

On the 2026 graph benchmark CognoDB answered a two-hop traversal in 0.27 ms and served 74,314 read queries a second, storing each edge in about 80 bytes on disk. The compact, disk-backed store keeps the working set cached and serves graphs larger than RAM — so a single box scales into the hundreds of millions of edges without an in-memory wall.

Your drivers already work.

CognoDB speaks Bolt 5.0–5.4 and understands Cypher, verified by a compatibility suite of 2,582 tests on every release build. The official Neo4j drivers for Python, JavaScript, Go, Java and .NET connect unmodified — evaluation costs a URI change, and so does leaving. We think that's how it should be.

The price is the infrastructure cost.

Published rates, itemized to the line, billed by the second: $26.04 per vCPU-month, $3.49 per GB of memory, $0.113 per GiB of storage. There is no list price with a discount behind a sales call — the estimate on the pricing page is generated by the same cost model the billing engine charges from.

A paused instance costs only its storage. A quiet tenant is close to free.

g82 vCPU · 8 GB · 16 GiB
Compute$52.08/mo
Memory$27.92/mo
Storage$1.81/mo
Daily backups$0.91/mo
Network$1.50/mo
Monthly$84.22/mo

Billed by the second. A paused instance is charged for storage only.

Isolation as the default posture.

Every instance is a real database: its own credentials, its own storage, its own resource limits, reachable only through its own encrypted endpoint. Console identity and data-path credentials are deliberately separate layers, every administrative change is audit-logged, and backups restore to a new instance so recovery can't destroy evidence.

When a security review asks how tenants are separated, “each has their own database” is a complete sentence.

The data model

Why a graph at all.

The honest comparison isn't against another vendor — it's against the relational model these workloads usually start in.

What an agent needsSimilarity searchRows & joinsContext graph
Find what's connected, not just similarCosine distance only — no concept of an edgeA join per relationship typeTyped edges are the data model
“What depends on X?” at unknown depthNo traversal at allRecursive CTEs, tuned per query[*1..n] — depth is a parameter
Provenance on every factMetadata blob, invisible to searchAnother audit table to maintainAn edge property — returned with the path
A private graph per agent or tenantNamespaces in a shared indexA schema per tenant, migrations foreverAn instance each — a credential boundary
Explain why this answerA score, unexplainedReconstruct the join path by handThe path IS the answer — return it

What we won't claim

The absences are deliberate.

No compliance badges we haven't earned, no SLA we haven't published, no benchmark wins over named competitors, no customer logos beyond the one customer who's really in production. When those change, this site will say so — plainly.

Start now

~98.7%

token efficiency at 2,000 entities — see the footnotes above

Judge it against your own workload.

The free tier exists so the evaluation is yours, not ours.

First-graph path

Live
  1. Create a free instance

    No card. Ready in about a minute.

  2. Connect your driver

    bolt+ssc:// URI into the driver you already use.

  3. Write two MERGEs

    That's the entire shape of agent memory.

  4. Point an agent at it

    One MCP config block. No integration code.