Grafana can mean two very different operating models: self-hosted LGTM that your team runs, or Grafana Cloud with usage-based pricing. Kloudfuse runs inside your cloud and prices in tiers: pick the band that covers your volume, and the price is fixed inside it.
The biggest differences come down to how the platform is deployed, how usage is priced, and what happens as data volume and complexity grow.
Grafana, Loki, Mimir, Tempo and Pyroscope are AGPLv3 and cost nothing to download. Grafana’s own documentation sets out the rest: "You pay for servers, database, and staff time."
Metrics are billed per 1,000 active series at $6.50 on Pro. Cardinality is not a surcharge on top of the bill; it is the bill.
Grafana Cloud bills logs at $0.05 per GB processed, $0.40 per GB written and $0.10 per GB per additional 30 days retained. Queries beyond 100× your written volume are billable as well.
Kloudfuse prices the telemetry footprint within one fixed ingestion tier. Your volume decides the band; the platform price holds inside it.
Each dimension is billed on its own, so more than one can move at a time.
Your platform price stays fixed anywhere inside the tier; only infra moves with volume.
One volume. One tier. One price.
Explore the Kloudfuse pricing calculator →Where the data plane runs, what each meter charges for, and what each platform will and will not do. Where a vendor does not publish something, the table says so.
| Grafana | Kloudfuse | |
|---|---|---|
| Where it runs | ||
| Data plane location | Three answers. Self-hosted LGTM runs anywhere, and you operate it. Grafana Cloud is multi-tenant SaaS. Grafana Cloud BYOC runs a managed region inside your own AWS or GCP account (Azure case by case). | Inside your VPC. Ingest, storage, query engine and the AI runtime all deploy into your own cloud account, writing to your object storage under your keys. Only metadata reaches Kloudfuse. |
| AWS, GCP and Azure | Partly. BYOC on AWS and GCP; Azure case by case | Yes. AWS · GCP · Azure |
| Drop-in migration from their agent | Not published. Their own agents | Yes. Datadog and OpenTelemetry agents accepted |
| What it costs | ||
| What you are billed on | Grafana Cloud meters active series, log and trace gigabytes twice over (processed and written), retention, and then seats on top — visualisation, IRM and AI users are separate lines. Self-hosted LGTM has no licence cost and runs on your infrastructure and staff time instead. | Ingestion tiers. Pick the band that covers your TB/day and the price is fixed inside it. Storage and compute land on your own cloud bill because the platform runs in your account. |
| High cardinality | Cardinality is the billing unit — active series is the meter. Adaptive Metrics is on by default and recommends aggregations to bring it down, which works, but the work is ongoing rather than structural. | No cardinality premium, no active-series cap, no per-custom-metric charge. Adding a label does not change the bill. |
| Retention | Cloud Pro: 13 months for metrics, 30 days for logs, traces and profiles. Extending logs costs $0.10 per GB per additional 30 days and is not retroactive. | Your bucket, your lifecycle policy. Long retention is a storage decision rather than a pricing tier. |
| Your data | ||
| Sampling | Adaptive Traces (tail sampling, GA October 2025) is available and configurable. No default sampling rate is published for Cloud. | No forced sampling. Full fidelity across metrics, logs, traces, events, RUM and profiles. |
| Every signal in one store | Partly. Separate LGTM components | Yes. One data lake |
| Log pattern reduction at ingest | Not published. Not published as an ingest control | Yes. Fingerprinting at ingest |
| Works with existing Grafana dashboards | Yes. It is Grafana | Yes. Kfuse datasource |
| Portability | The strongest of the six on portability — it is open source, and Kloudfuse is deliberately compatible with it. Kfuse ships a Grafana datasource so existing dashboards and TraceQL queries keep working. | Built on OpenTelemetry, with Grafana-compatible dashboards through the Kfuse datasource and support for TraceQL and PromQL. Your data stays in your object storage. Your instrumentation and dashboards remain portable and independent of Kloudfuse. |
| AI | ||
| Agents and assistants | Grafana Assistant at $20 per active AI user, Investigations and Agent Observability at $2 per million tokens. Note that Investigations and Agent Observability are unmetered only until 1 October 2026. | Dexter runs inside your boundary, on the model you choose, including a local one. Prompts and completions stay inside the boundary. A human stays in control. |
| Inference inside your own boundary | No. Assistant is Grafana-hosted | Yes. Dexter runs in your VPC |
| Security and compliance | ||
| Government and regulated | Grafana Federal Cloud is marketed as FedRAMP High and DoD IL5 — but the authorisation is inherited through Palantir FedStart (listing PFCS-SS, FR2315464863), not a standalone Grafana package. HIPAA is not listed on Grafana’s trust centre. | The data plane runs inside whatever boundary you have already had authorised. FIPS 140-3 validated cryptography (CMVP #5186, #5209); FedRAMP-targeted installations run with EBS encryption. |
| SOC 2 Type II | Yes. Type II | Yes. Type II |
| HIPAA | No. Not listed on their trust centre | Partly. In progress |
| FIPS-validated cryptography | Not published. Not published | Yes. 140-3 · CMVP #5186, #5209 |
| Data residency | Partly. Region choice, or your account on BYOC | Yes. Structural — your own VPC |
YesPartlyNoNot published
No. Kloudfuse ships a Grafana datasource, so you keep Grafana as the front end and change what runs underneath it. What changes is the storage and query layer — operating Mimir and Loki at scale, or the Grafana Cloud bill.
At small scale, often yes — the software is free and a single-binary Loki on one box is genuinely low effort. At the scale where cardinality and ingest volume start to hurt, the cost moves into headcount and incident load rather than disappearing. Grafana’s own docs say it plainly: you pay for servers, database and staff time.
They are structurally similar: a managed deployment inside your own cloud account, where you own the infrastructure. The differences are in the billing model underneath (active series versus terabytes ingested) and in whether the AI layer runs inside the same boundary.
Grafana Federal Cloud is marketed that way and the underlying authorisation is real, but it is inherited through Palantir’s FedStart programme rather than held by Grafana Labs directly — the marketplace listing is Palantir’s. Whether that satisfies your agency is a question for your authorising official, not for a comparison table.
Grafana, Loki, Mimir, Tempo and Pyroscope are trademarks of Grafana Labs. Kloudfuse is not affiliated with, endorsed by or sponsored by Grafana.
Run Kloudfuse against your own telemetry, in your own cloud, and see what a single correlated view does to your mean time to resolution.