Why CognoDB
The differences, stated without hand-waving.
No unsourced multiples. Four concrete differences, each verifiable, and a footnote for every number — including the comparative ones, which trace to a public harness rather than to us.
- 1Read-only throughput: 40 concurrent clients replaying a seeded workload for 60 s. The mixed run (10% writes) sustained 49,777/s, and the write-only run 25,668/s. Measured 2026-08-25 on engine 0.9.15 (commit 3106d98). SNAP soc-Pokec loaded to exactly 3,000,000 edges over 663,230 nodes, on one n2-standard-8 (8 vCPU / 32 GiB, Intel Cascade Lake), engine on defaults, median of 3 reps after a discarded warm-up.
- 2p95 latency for a two-hop, friends-of-friends traversal against the loaded graph (p50 0.25 ms). Measured 2026-08-25 on engine 0.9.15 (commit 3106d98). SNAP soc-Pokec loaded to exactly 3,000,000 edges over 663,230 nodes, on one n2-standard-8 (8 vCPU / 32 GiB, Intel Cascade Lake), engine on defaults, median of 3 reps after a discarded warm-up.
- 3Edges merged per second, sustained while loading the graph over Bolt in batches, alongside 224,519 nodes/s. Measured 2026-08-25 on engine 0.9.15 (commit 3106d98). SNAP soc-Pokec loaded to exactly 3,000,000 edges over 663,230 nodes, on one n2-standard-8 (8 vCPU / 32 GiB, Intel Cascade Lake), engine on defaults, median of 3 reps after a discarded warm-up.
- 4On-disk store size divided by edges loaded (304 MB for 3,000,000 edges), read from the container's cgroup counters rather than self-reported. A compact store keeps the working set cached. Measured 2026-08-25 on engine 0.9.15 (commit 3106d98). SNAP soc-Pokec loaded to exactly 3,000,000 edges over 663,230 nodes, on one n2-standard-8 (8 vCPU / 32 GiB, Intel Cascade Lake), engine on defaults, median of 3 reps after a discarded warm-up.
Fast reads, low latency, compact store.
Your drivers already work.
One price per machine. No itemized surprises.
Each machine type has a single published monthly price — compute, memory, disk, network and daily backups all included — billed by the second. The only add-on is additional storage, at a flat $1.00/GB. No list price with a discount behind a sales call; the estimate on the pricing page is the same figure the billing engine charges from.
A paused instance costs only its storage. A quiet tenant is close to free.
Usage-based billing. 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.
Vendor-versus-vendor numbers are on /benchmarks. The comparison that decides whether you want a graph at all is a different one: against the relational model these workloads usually start in.
| What an agent needs | Similarity search | Rows & joins | Context graph |
|---|---|---|---|
| Find what's connected, not just similar | Cosine distance only, no concept of an edge | A join per relationship type | Typed edges are the data model |
| “What depends on X?” at unknown depth | No traversal at all | Recursive CTEs, tuned per query | [*1..n]: depth is a parameter |
| Provenance on every fact | Metadata blob, invisible to search | Another audit table to maintain | An edge property, returned with the path |
| A private graph per agent or tenant | Namespaces in a shared index | A schema per tenant, migrations forever | An instance each, a credential boundary |
| Explain why this answer | A score, unexplained | Reconstruct the join path by hand | The 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 customer logos beyond the one customer who's really in production. This list used to include benchmark wins over named competitors; it no longer does, because /benchmarks now publishes a cross-engine comparison with the harness, the raw output and the disqualifications all public. That is the change this section promised it would say plainly.
Start now
~98.7%
token efficiency at 3,700 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
LiveCreate a free instance
No card. Ready in about a minute.
Connect your driver
bolt+s:// URI into the driver you already use.
Write two MERGEs
That's the entire shape of agent memory.
Point an agent at it
One MCP config block. No integration code.