Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Grafana

The open-source dashboards and alerting most teams already run, with an MCP server, an LLM plugin and paid AI help in Grafana Cloud.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

4.4/5

Grafana is the safe default for seeing what your LLM app is doing in production, as long as you accept that its smartest AI features live in Grafana Cloud rather than the open-source build.

What is Grafana?

Grafana queries metrics, logs and traces wherever they are stored, Prometheus and Loki among many data sources, and draws them on shared dashboards with template variables, ad-hoc exploration and alert rules that notify Slack, PagerDuty, OpsGenie and similar tools. It is licensed AGPL-3.0-only, with some parts under Apache-2.0, and version 13.3.1 shipped on 10 November 2026.

For teams running LLM apps, the AI pieces sit around that core. The open-source MCP server (grafana/mcp-grafana, Apache-2.0) lets Claude Code, Cursor and other clients query telemetry and manage dashboards and alert rules. The open-source LLM plugin lets you bring your own model provider and keys to add explain-in-place features. Grafana Assistant, the chat agent that writes queries, builds dashboards and investigates incidents, is generally available in Grafana Cloud and in public preview for self-managed Grafana connected to Cloud. Grafana Cloud's AI Observability, built on OpenLIT and OpenTelemetry, tracks LLM latency, tokens and cost alongside vector databases, GPUs and MCP servers.

Is Grafana worth it?

Grafana is where most engineering teams already look when something breaks, and that is its strongest argument for AI workloads: token counts, latency and cost per model can sit on the same dashboards and alert rules as the database and the queue. With 77,989 GitHub stars and patch releases for five branches, from 12.4 to 13.3, on a single day in November 2026, it is about as established as open-source infrastructure gets.

The AI layer splits into open and paid parts. Open: the MCP server, which gives any MCP client dashboards, Prometheus and Loki queries, alert rules and incidents through a service-account token, and the LLM plugin, which centralises model keys for features such as explaining a panel. Paid: Grafana Assistant, which writes queries, builds dashboards and runs incident investigations from chat, Slack or Microsoft Teams. Grafana Cloud's Free tier includes 3 active Assistant users with 40M tokens each; on Pro each user starts at $20 a month and extra tokens cost $2 per 1M.

For watching your own models, Grafana Cloud's AI Observability uses OpenLIT and OpenTelemetry to record response times, tokens, spend, evaluations for hallucination and toxicity, vector database calls, GPU health and MCP tool use. Agent Observability includes 30k generations a month, then $1.50 per 1k on Pro. These are Cloud products; on self-hosted Grafana you assemble a similar picture from your own OpenTelemetry data, which is more work but keeps everything inside your network.

The catch: the licence and the bill. AGPL-3.0 is fine for internal use, but anyone who offers a modified Grafana to users over a network must make that source available, and some legal teams refuse AGPL outright. Grafana Cloud Pro is $19 a month plus usage, with metrics from $6.50 per 1k series, logs at $0.40 per GB written and users from $8 each, so an app that logs every prompt can climb quickly; Enterprise starts at a $25,000 yearly commitment. Grafana is also not a prompt-debugging tool: datasets, prompt versions and per-trace evals go deeper in Langfuse or Phoenix.

Our take: Grafana is the room where the whole team already meets, and Langfuse is the microscope for a single prompt; most teams running LLMs in production end up wanting both.

Where it earns its place

  • A six-person SRE team running a customer-facing RAG chatbot

    Get paged when answer latency or token spend jumps

    Send OpenTelemetry metrics from the app, chart tokens, cost and slow responses per model next to the vector database, and alert to Slack or PagerDuty. Self-hosted Grafana costs only servers; Cloud Free covers 10k metric series and 50 GB of logs a month.

  • A backend developer on call who works in Claude Code

    Ask an agent what changed before an incident without clicking through dashboards

    Run mcp-grafana with a tightly scoped service-account token, or point the client at Grafana Cloud's hosted MCP server, which uses OAuth 2.1. Keep write permissions off unless needed, since the server can also edit dashboards and alert rules.

  • A head of engineering at a 40-person SaaS company

    Decide whether paid AI help with incidents is worth it

    Trial Grafana Assistant on the Free tier's 3 AI users before paying. On Pro, the $19 monthly platform fee plus $20 per active AI user includes 40M tokens each, and extra tokens are $2 per 1M.

Before you commit

Setup
Install Grafana from the official packages or Docker image and add an existing Prometheus or Loki data source, or sign up for Grafana Cloud Free without a credit card. mcp-grafana runs with uvx mcp-grafana, GRAFANA_URL and a service-account token, and needs Grafana 9.0 or later.
Integrations
Prometheus, Loki and many other data sources; Slack, PagerDuty, VictorOps and OpsGenie for alerts; Claude Code, Cursor and other MCP clients; OpenLIT and OpenTelemetry for AI telemetry; Assistant in Slack and Microsoft Teams.
Data and privacy
Self-hosted Grafana keeps data on your servers. Grafana Cloud lists SOC 2, PCI DSS and FedRAMP. Grafana's AI pages do not say which model providers Assistant uses or whether inputs are used for training, so ask before sending prompts that contain personal data.
Cost at scale
Cloud Pro: $19 a month base, metrics from $6.50 per 1k series, logs $0.40 per GB written, users from $8 and Assistant users from $20. Enterprise starts at $25,000 a year. Self-hosting has no licence fee, but you run and store Prometheus and Loki yourself.
Limits
Cloud Free keeps data 14 days and allows 3 active users; Pro keeps metrics 13 months and logs 30 days. The main repository has 3,349 open issues and pull requests. AGPL-3.0 obligations apply if you offer a modified Grafana as a service.

What’s good

  • Puts LLM cost and latency beside the rest of your infrastructure
  • Open-source MCP server and LLM plugin under Apache-2.0
  • Cloud Free tier includes 3 Grafana Assistant users
  • Mature alerting to Slack, PagerDuty and OpsGenie

What’s not

  • Grafana Assistant is Cloud-first; self-managed access is a public preview
  • AGPL-3.0 worries some legal teams
  • Usage pricing grows with prompt logs and metric series
  • No prompt-level tracing or eval workflow in the open-source core

Best for

Teams that already run Grafana, or want one place for infrastructure and LLM metrics, and can live with AGPL or pay for Grafana Cloud.

Grafana alternatives

Grafana: common questions

Is Grafana worth it?
Grafana is the safe default for seeing what your LLM app is doing in production, as long as you accept that its smartest AI features live in Grafana Cloud rather than the open-source build. We rate it 4.4 out of 5.
How much does Grafana cost?
Free and open source under AGPL-3.0. Grafana Cloud: Free at $0; Pro $19/month plus usage; Enterprise from $25,000/year. Grafana Assistant: 3 active AI users free, then from $20 per active AI user on Pro.
What are the downsides of Grafana?
Grafana Assistant is Cloud-first; self-managed access is a public preview; aGPL-3.0 worries some legal teams; usage pricing grows with prompt logs and metric series; no prompt-level tracing or eval workflow in the open-source core.
Who is Grafana best for?
Teams that already run Grafana, or want one place for infrastructure and LLM metrics, and can live with AGPL or pay for Grafana Cloud.
How much does Grafana cost for a team or in production?
Cloud Pro: $19 a month base, metrics from $6.50 per 1k series, logs $0.40 per GB written, users from $8 and Assistant users from $20. Enterprise starts at $25,000 a year. Self-hosting has no licence fee, but you run and store Prometheus and Loki yourself.
Is Grafana safe for company data?
Self-hosted Grafana keeps data on your servers. Grafana Cloud lists SOC 2, PCI DSS and FedRAMP. Grafana's AI pages do not say which model providers Assistant uses or whether inputs are used for training, so ask before sending prompts that contain personal data.
What are the best Grafana alternatives?
The closest alternatives we have reviewed are Langfuse, Arize Phoenix, OpenLLMetry and OpenTelemetry Collector.

Answers checked Nov 18, 2026.

Make Grafana? Show this rating on your siteGrafana rated 4.4 out of 5 on ReviewRamble