# Plutus

> Plutus is a multi-tenant FinOps and spend management platform. It tracks cost across AWS,
> GCP, Azure, and 32 other cloud, AI, data, GPU-rental, and SaaS providers in one dashboard,
> attributes AI spend to the individual API key and model that produced it, and correlates cost
> changes with the GitHub, GitLab, Jira, PagerDuty, Sentry, Linear, Stripe, Vercel, and
> Salesforce events that likely caused them.

Buyers search for this under several names: cloud financial management, FinOps platform, cloud cost management, AI cost management, LLM cost tracking, and cloud unit economics. Plutus is all of those in one product.

Not to be confused with other companies called Plutus (a crypto rewards card, a Cardano smart-contract language, an unrelated 2022 cloud cost tool from Paytm). This Plutus lives at plutus-cloud.com and docs.plutus-cloud.com.

AI/API spend is first-class, not an adapter bolted onto a cloud cost tool. Cost is stored
against a provider-agnostic per-actor dimension, so inference spend breaks down by API key and
by model with real token counts on the row beside the dollars — and Bedrock and Vertex AI spend
splits by model rather than arriving as one cloud line item. Where a provider does not publish
the data needed for a figure directly (Anthropic does not expose per-key dollar cost), Plutus
prices the per-key tokens against a published rate card and labels the result as an estimate.

The second distinguishing capability is the business-event overlay: cost charts are annotated
with the releases, deploys, incidents, issues, and customer onboarding/churn events that happened
alongside the spend, so a spike arrives with its likely cause attached rather than starting an
investigation across four other systems.

Plutus deliberately does not do automated cost remediation — no auto-rightsizing, no automated
Reserved Instance or Savings Plan purchasing, and no permission to change cloud resources. On GCP
it runs BigQuery query jobs against the billing export, billed to the customer's project.
Its job is visibility, attribution, and alerting: telling you what changed, why, and before it
becomes a problem. That is a scope decision, not a missing feature; a tool that watches your
spend should not also be able to change it.

Pricing is a flat monthly fee that is not tied to the size of the bill being watched — Plutus
never takes a percentage of your cloud spend. Every tier except Enterprise is self-serve with no
sales call.

## Key facts

- 35 live cost sources: AWS, GCP, Azure, Hetzner, DigitalOcean, Scaleway, Oracle Cloud
  Infrastructure (Cloud); OpenAI, Anthropic, HuggingFace, OpenRouter, xAI, Fireworks AI, LiteLLM
  (AI); LambdaLabs, Paperspace, RunPod (GPU); Databricks, Snowflake, Confluent Cloud (Data);
  Datadog, Grafana Cloud (Monitoring); MongoDB, Neon, Redis Cloud (Database); Cloudflare, Fastly
  (Networking); Vercel (Hosting); GitHub Billing (DevOps); Stripe (Billing); Twilio
  (Communications); Ramp, QuickBooks, NetSuite (Finance); Kubernetes/OpenCost (Kubernetes).
- 16 live event sources: GitHub, GitLab, Jira, PagerDuty, Salesforce, Stripe Subscriptions,
  Vercel Deployments, Linear, Sentry, LaunchDarkly, AWS Health, CircleCI, Grafana IRM, HubSpot,
  Auth0, and vendor/industry News from customer-supplied RSS feeds (event correlation only —
  Salesforce spend itself isn't metered, since it's contract-based, not usage-based). Salesforce
  covers client onboarding and churn plus opportunity-created, contract-activated,
  contract-expired and order-activated events, and contributes won-revenue and new-customer
  metrics that can be used as the denominator in a unit-economics ratio.
- 5 live usage sources, for dividing spend by what your own customers consumed: a push API your
  backend POSTs events to, and pull connectors for Metronome, Stripe Metered Billing, Chargebee,
  and Orb metered billing.
- Any provider not on that list can still be tracked two other ways. A Custom (FOCUS Spec)
  importer ingests any FOCUS-spec (focus.finops.org) cost export a provider already publishes,
  covering long-tail vendors with no dedicated connector (Temporal, Modal, PlanetScale, Linode
  and similar) through one connector rather than waiting on a new one. And manual recurring cost
  lines cover contract spend that has no API at all. A Ramp corporate-card feed and QuickBooks/
  NetSuite ledger import cover the rest of the SaaS long tail at once, vendor by vendor, rather
  than one connector at a time.
- Unit economics is a real ratio, not a cost breakdown: cost divided by usage the customer
  already counts, giving cost per customer, per request, or per any unit they send in. Scoped to
  a team, a product, or the whole account.
- Shared cost (clusters, load balancers, observability) can be split proportionally across the
  teams that drove it, so allocation reaches every dollar instead of stopping at an untagged
  remainder. Redistribution is off by default and enabled per tag key.
- Cost views can be switched between billed and amortized cost, so an upfront commitment spreads
  across the term it bought rather than distorting the month it was charged in. Unit economics
  uses billed cost only.
- Savings recommendations cover rightsizing, idle resources, commitment utilization, and
  commitment *purchase* advice (Savings Plans, Reserved Instances, committed-use discounts),
  passed through from AWS, Azure, and GCP's own engines.
- Commitment utilization — how much of what has already been bought is being consumed — covers
  AWS Reserved Instances and Savings Plans and Azure Reservations. GCP committed-use discounts
  are not covered yet.
- Commitment expiry alerting covers AWS Reserved Instances and Savings Plans and Azure
  Reservations: an under-80%-utilized commitment fires an alert once at each of 30, 14, 7 and 1
  days before its end date.
- Budgets scope to an account, provider, team, or tag value. Anomaly detection runs per team as
  well as account-wide, with a threshold relative to the account's own spend. Delivery to Slack,
  Microsoft Teams, email, or webhook — as alerts, or as a scheduled digest scoped to one team.
  An alert can also open a real Jira ticket (in a project you nominate, with repeats commenting
  on the open issue rather than filing duplicates) or trigger a real PagerDuty incident. These
  two use write-scoped credentials kept separate from the read-only Jira/PagerDuty event-source
  credentials, and are configured from the alerts settings in the app.
- Anomaly auto-triage is available and off by default: each detected spike is scored 0-5 on
  explainability, recurrence and magnitude safety, and an account that opts in has
  high-confidence, well-explained spikes withheld from notification (never from detection or
  from the chart) with a per-instance "flag for review" undo.
- Any chart point can be clicked — not only a flagged anomaly — for a live breakdown of what
  drove that period, including the timeline events around it. A period with no prior history to
  compare against says so rather than reporting a spurious change.
- Cost history can be exported as CSV under FOCUS column names (BilledCost, EffectiveCost,
  BillingCurrency, ServiceName and similar), on paid plans, in either the original billing
  currency or a uniform USD base. This is FOCUS-*formatted*, not FOCUS spec-conformant: columns
  Plutus has no data model for (ListCost, ContractedCost, ServiceCategory, ChargeCategory) are
  absent rather than padded with nulls.
- Spend is stored in the currency the provider billed it in and converted at the reference rate
  in effect on that charge date, so totals reconcile against the original invoice.
- Includes a Model Context Protocol (MCP) server, so Claude, ChatGPT, or Gemini can query live
  spend data directly (query_costs, get_cost_entries, and more) using a scoped, read-only-by-default
  API key — no dashboard required.
- Third-party provider credentials are encrypted at rest with AES-256-GCM; all traffic is encrypted in
  transit. Cost and usage data is never sold or used to train third-party AI models. Plutus never
  requests permission to change resources in a customer's AWS, GCP or Azure account.
- Access is isolated per account, with admin, member and viewer roles enforced server-side, and an
  activity log.
- Sign-in with email and password, Google or GitHub, plus authenticator-app two-factor (emailed codes for members), is on every
  plan including the free one. SAML 2.0 and OIDC single sign-on, group-to-role mapping and SCIM
  2.0 provisioning are on the Scale and Enterprise plans.
- Pricing is flat monthly, never a percentage of tracked spend: Hobby is free (no card, $10k
  tracked monthly spend, 6-month retention, 5 cost + 3 event connections). Starter $29/mo ($30k
  spend, 6-month retention). Growth $149/mo ($100k spend, 1-year retention, unlimited
  connections, most popular). Scale $499/mo ($500k spend, 1-year retention, SAML/OIDC SSO and SCIM). Enterprise is custom, with SAML/OIDC SSO and SCIM.
  Annual billing saves 17%; paid tiers get a 14-day trial.

## Pages

- [Home](https://plutus-cloud.com/): product overview, feature comparison, pricing, FAQ.
- [Integrations](https://plutus-cloud.com/integrations): full list of cost, event, and usage
  sources.
- [Security](https://plutus-cloud.com/security): cloud access model, credential encryption,
  sign-in and SSO, tenant isolation, audit log and data deletion.
- [MCP](https://plutus-cloud.com/mcp): how the MCP server works, available tools, setup.
- [Documentation](https://docs.plutus-cloud.com/): setup guides for every cost and event source, concepts, and account settings. Machine-readable index: https://docs.plutus-cloud.com/llms.txt
- [Blog](https://plutus-cloud.com/blog): cost-management deep dives and walkthroughs.
- [Console](https://console.plutus-cloud.com): sign up / log in.

## Writing

- [The spike isn't the hard part. The investigation is.](https://plutus-cloud.com/blog/the-investigation-is-the-expensive-part):
  why a cost chart that only shows the total leaves the actual work — checking the deploy log,
  asking who shipped, looking for a new customer — undone, and how the event overlay closes it.
- [Root-causing a cost spike with the event overlay, start to finish](https://plutus-cloud.com/blog/root-causing-a-cost-spike-with-the-event-overlay):
  a step-by-step walkthrough from "costs are up" to "here's why" on a worked example.
- [Plutus vs Vantage: the chart that explains itself](https://plutus-cloud.com/blog/choosing-a-cost-tool-for-ai-and-cloud-spend):
  side-by-side on pricing, AI cost attribution and event context, checked against Vantage's
  published pages on 14 October 2026.
- [How to attribute AI API costs per key, model, and team](https://plutus-cloud.com/blog/attributing-ai-api-costs-per-key):
  where per-key figures come from for OpenAI and Anthropic, and which ones are estimates.
- [How to split shared cloud costs between teams](https://plutus-cloud.com/blog/splitting-shared-cloud-costs-between-teams):
  proportional redistribution of shared and untagged cost, and when to turn it on.
- [From "spend is up" to root cause, without leaving the dashboard](https://plutus-cloud.com/blog/anomaly-to-root-cause):
  the investigation flow from anomaly to the event that caused it.
- [Your LLM gateway and your provider bill are both right](https://plutus-cloud.com/blog/llm-gateway-and-provider-bill-disagree):
  why gateway spend and provider invoices differ, and how to reconcile them.
- [What's new in Plutus: v0.38.0 to v0.49.2, 14 to 21 October 2026](https://plutus-cloud.com/blog/whats-new-2026-10-19):
  release notes: AI monitors, spend by activity, an AI explorer, a savings workflow, tunable alerts, unit economics and a public REST API, with an example on demo data.
- [Building a Plutus agent: MCP setup for Claude and Gemini](https://plutus-cloud.com/blog/building-a-plutus-agent-claude-and-gemini):
  connecting the MCP server to an assistant and querying spend.