Compare the bill: an estimate

Same answer, every time, with governance built in, for a fraction of the bill.

Hosting: $44 vs $1.5K+ a month. No year of engineering. About a third of the context tokens, $204 a month less at this volume.

Estimate, based on these assumptions

51,000 context reads a month, 50,000 documents, Claude Sonnet 5.5, likely case, engineering time included at $199,300 a year per engineer.

Cheapest alternative vs ContextNest: 4.9x with engineering time, about the same without it.

In our preprint, governed selection used about a third of the context tokens of BM25 retrieval, at 97% accuracy against 90 to 93%. Preprint arXiv:2607.02116, not peer reviewed, small corpus.

$3,807

ContextNest per month, likely case. Hosting is $44 of it.

$18K–$27K

The same job on graph-and-search stacks, per month

4.9x

How much more the cheapest alternative costs

Scenario

Context reads a month

People200

AI questions per person per day that pull company context5

Working days a month21

Agents20

Runs per agent per day10

Context reads per run5

Agent days a month30

200 people × 5 questions × 21 days = 21,000
+ 20 agents × 10 runs × 5 reads × 30 days = 30,000
= 51,000 context reads a month. All assumptions.

Model and content

Model

List price. Source, checked October 2026.

Context tokens per read (RAG)3,000

Documents50,000

Engineering and cases

Engineer cost, fully loaded per year$199,300

Fully loaded engineer cost $199,300/yr = BLS median database architect wage $139,500 / 0.70 wage share (BLS ECEC June 2026).

Include engineering timeInclude answer tokens (same for every stack)

ContextNest Community (self-hosted)Ours

$3,807 / month · $137,876 over 3 years

Hosting $44Context tokens $102Engineering $1,661License $2,000

One Docker container + a disk. SQLite by default, managed PostgreSQL when your host's disk isn't durable. No graph database, no vector or search cluster, no ingestion pipeline. Time to first governed answer: Same day (install, import, approve). Assumption based on the CLI quickstart.

Line items and sources

Item	Quantity	Monthly	Basis	Source
Hosting		$44
Lightsail 4 GB instance	1 month	$24	list price	Source
Lightsail managed PostgreSQL (smallest plan)	1 month	$15	list price	Source
Snapshots/backups	1 month	$5	assumption	Source
License		$2,000
ContextNest self-hosted license 30-person minimum.	200 person-month	$2,000	list price	Internal
Context tokens		$102
1,000 context tokens per read × 51,000 reads, Claude Sonnet 5.5 input price About a third of RAG context per read (preprint arXiv:2607.02116, small corpus).	51M tokens	$102	assumption	Source · Price
Engineering time		$1,661
Engineering time running it (0.1 FTE) Time your engineers spend running plumbing instead of building your product. FTE share is an assumption. One-time build: $830.	0.1 FTE	$1,661	assumption	Source

AWS Bedrock Knowledge Bases + GraphRAG (Neptune Analytics)

$18,485 / month · $765,106 over 3 years

Hosting $1,550Context tokens $306Engineering $16,608Pipeline $21

Neptune Analytics holds the graph and vectors. Bedrock extracts entities and embeds. VPC, IAM, KMS and monitoring around it. Governance (approval, versions, read log) is yours to build. Time to first governed answer: Weeks: VPC/IAM, ingestion, graph sizing, re-sync jobs, then your own approval and audit layer. Assumption.

Line items and sources

Item	Quantity	Monthly	Basis	Source
Hosting		$1,550
Neptune Analytics graph, 64 m-NCU (memory-optimized, always on) Sizing is an ASSUMPTION: graph + 1024-dim vectors for ~500k chunks from 50k docs.	64 m-NCU-hour	$1,402	list price	Source
NAT gateways (2 AZs) for private-subnet egress Includes ~50 GB/month processed (assumption).	2 gateway-hour	$68	list price	Source
CloudWatch logs/metrics, KMS keys, S3 source bucket Typical small-footprint estimate. Not a single list price.	1 month	$80	assumption	Source
Pipeline model calls		$21
GraphRAG entity extraction (Claude Haiku 4.5), 10% docs re-processed monthly 2,000 tokens/doc and 10% monthly churn are assumptions. First full ingest ≈ $200 one-time.	10 M input tokens	$20	assumption	Source
Embeddings (Titan Text Embeddings V2) for changed docs Volume is an assumption (2,000 tokens/doc, 10% churn).	10,000 1K tokens	$1	list price	Source
Context tokens		$306
3,000 context tokens per read × 51,000 reads, Claude Sonnet 5.5 input price Context tokens per read is an assumption. GraphRAG is kept equal to plain RAG to stay conservative.	153M tokens	$306	assumption	Source · Price
Engineering time		$16,608
Engineering time running it (1 FTE) Time your engineers spend running plumbing instead of building your product. FTE share is an assumption. One-time build: $99,650.	1 FTE	$16,608	assumption	Source

AWS Context Ontology Accelerator (Neptune + OpenSearch Serverless + Bedrock)

$27,019 / month · $1,105,560 over 3 years

Hosting $1,781Context tokens $306Engineering $24,913Pipeline $20

Open-source accelerator (Apache 2.0). You run Neptune, OpenSearch Serverless and Bedrock, plus expert time approving the ontology. AWS Context itself is in preview with no public price. Time to first governed answer: Weeks to months: deploy the stack, draft the ontology, have experts approve it, wire agents. Assumption.

Line items and sources

Item	Quantity	Monthly	Basis	Source
Hosting		$1,781
Neptune Analytics graph, 64 m-NCU (memory-optimized, always on) Sizing is an ASSUMPTION: graph + 1024-dim vectors for ~500k chunks from 50k docs.	64 m-NCU-hour	$1,402	list price	Source
OpenSearch Serverless nextgen, 4 OCUs x 220 active hours NextGen collection: no minimum, scales to zero when idle. Active hours (220/month) are an assumption.	880 OCU-hour	$211	list price	Source
NAT gateways (2 AZs) for private-subnet egress Includes ~50 GB/month processed (assumption).	2 gateway-hour	$68	list price	Source
CloudWatch logs/metrics, KMS keys, S3 source bucket Typical small-footprint estimate. Not a single list price.	1 month	$100	assumption	Source
Pipeline model calls		$20
GraphRAG entity extraction (Claude Haiku 4.5), 10% docs re-processed monthly 2,000 tokens/doc and 10% monthly churn are assumptions. First full ingest ≈ $200 one-time.	10 M input tokens	$20	assumption	Source
Context tokens		$306
3,000 context tokens per read × 51,000 reads, Claude Sonnet 5.5 input price Context tokens per read is an assumption. GraphRAG is kept equal to plain RAG to stay conservative.	153M tokens	$306	assumption	Source · Price
Engineering time		$24,913
Engineering time running it (1.5 FTE) Time your engineers spend running plumbing instead of building your product. FTE share is an assumption. One-time build: $132,867.	1.5 FTE	$24,913	assumption	Source

Neo4j GraphRAG (AuraDB) + your own app and pipeline

$19,421 / month · $798,818 over 3 years

Hosting $2,486Context tokens $306Engineering $16,608Pipeline $21

Managed Neo4j for the graph and vector index, plus an app tier and an embedding/extraction pipeline you build and run. Time to first governed answer: Weeks: model the graph, build ingestion and re-sync, build the app and access layer. Assumption.

Line items and sources

Item	Quantity	Monthly	Basis	Source
Hosting		$2,486
AuraDB Business Critical, 16 GB (SLA, 30-day backups) Size is an assumption.	16 GB-hour	$2,336	list price	Source
App + pipeline host	1 month	$150	assumption	Source
Pipeline model calls		$21
GraphRAG entity extraction (Claude Haiku 4.5), 10% docs re-processed monthly 2,000 tokens/doc and 10% monthly churn are assumptions. First full ingest ≈ $200 one-time.	10 M input tokens	$20	assumption	Source
Embeddings (Titan Text Embeddings V2) for changed docs Volume is an assumption (2,000 tokens/doc, 10% churn).	10,000 1K tokens	$1	list price	Source
Context tokens		$306
3,000 context tokens per read × 51,000 reads, Claude Sonnet 5.5 input price Context tokens per read is an assumption. GraphRAG is kept equal to plain RAG to stay conservative.	153M tokens	$306	assumption	Source · Price
Engineering time		$16,608
Engineering time running it (1 FTE) Time your engineers spend running plumbing instead of building your product. FTE share is an assumption. One-time build: $99,650.	1 FTE	$16,608	assumption	Source

Priced by contract or capacity

Platforms without a simple public price

Microsoft Fabric IQ

Fabric capacity, ~$0.18 per CU-hour pay-as-you-go (F64 ≈ $8,410/month). Capacity needed for Fabric IQ not published.

Source

Snowflake (Cortex, semantic views)

Credits, typically $3–4 per credit on Enterprise on-demand, plus storage.

Source

Databricks (Unity Catalog, Genie)

DBUs, Serverless SQL ≈ $0.70–0.75/DBU on AWS, plus cloud costs.

Source

Palantir AIP / Ontology

No public price. Enterprise contract.

Source

AWS Context

Preview, no public price yet.

Context tokens = context tokens per read × reads × the model's input price. In our preprint, ContextNest's governed selection used about a third of the context tokens of BM25 retrieval. Preprint arXiv:2607.02116, not peer reviewed, small corpus. The 3,000-token RAG default is an assumption: 5 chunks × 512 tokens is 2,560 (source).

List prices as of Oct 2026, on-demand, us-east-1. Your bill will vary. Assumptions editable. Context tokens (the input that delivers context to the model) are counted per stack. Answer tokens (model output and agent work) are the same on every stack, so they're off by default and can be switched on. Enterprise discounts, support plans and Neo4j Enterprise licenses (contact sales) are excluded. Assumption, built bottom-up: people x AI questions per day that pull company context x working days, plus agents x runs per day x context reads per run x days. Engineering time is shown as the time your engineers spend running plumbing, which they could spend on your product instead.

Business case: what disconnected AI context costs your team, and what ContextNest costs

Back to all comparisons