Cletrics FinOps Agents

The open-source reference library for cloud cost operations. A curated set
of specialist AI agent personas and named-pattern playbooks that drop into
any modern coding assistant -- Claude Code, OpenAI Codex CLI, Gemini CLI,
GitHub Copilot, Cursor, Windsurf, Aider, and OpenCode -- plus ChatGPT Custom
GPTs, Claude Projects, and Gemini Gems on the web.

Built by Cletrics and the FinOps community.
MIT-licensed. Vendor-neutral where possible. Contributions welcome.

Framework-aligned. 22 of 22 Capabilities covered (100%). Every
agent carries explicit
FinOps Framework metadata -- Domain,
Capability, Phase, Personas, Maturity entry point -- and references a
shared doctrine layer (Iron Triangle, Data in the Path,
Crawl/Walk/Run, FCP Canon Anchors, FOCUS Essentials). See
fcp-coverage.md for the live matrix, or run
./scripts/fcp-coverage.sh.

FOCUS-first. The library defaults to the
FinOps Open Cost & Usage Specification (FOCUS)
column semantics -- BilledCost, EffectiveCost, ListCost,
ContractedCost, ServiceCategory, CommitmentDiscountStatus, etc. --
so agent recommendations carry across providers. Provider-native
schemas (AWS CUR, Azure exports, GCP billing export) are documented
fall-back layers, not the primary frame.

Why this exists

Cloud cost operations is a practice: tagging, allocation, forecasting,
commitment management, waste detection, unit economics, and the human
cadences that tie them together. It's also full of specialized knowledge --
CUR schema quirks, GCP SUD dynamics, Azure enrollment types, Kubernetes
attribution math -- that every team ends up re-learning from scratch.

This repo packages that knowledge as agent personas your coding assistant
can load on demand. Instead of "help me write a query against my multi-
cloud billing data," you ask the Cloud Billing Analyst agent, and it
knows the FOCUS schema, the cost-column semantics (Billed / Effective /
List / Contracted), and the common gotchas before you type the first
line.

Install

git clone https://github.com/cletrics/finops-agents.git
cd finops-agents

# Install for all detected tools
./scripts/install.sh

# Or pick one
./scripts/install.sh --tool claude-code
./scripts/install.sh --tool codex         # OpenAI Codex CLI
./scripts/install.sh --tool gemini-cli    # Gemini CLI (skill format)
./scripts/install.sh --tool copilot
./scripts/install.sh --tool cursor
./scripts/install.sh --tool windsurf
./scripts/install.sh --tool aider
./scripts/install.sh --tool opencode
./scripts/install.sh --tool chatgpt       # prints Custom GPT / Projects bundle path

Web-only surfaces (no CLI required)

ChatGPT Custom GPTs, Claude Projects, and Gemini Gems can load the agents
directly. See integrations/chatgpt/README.md
for setup instructions for each.

The installer only does local file copies into your tool's standard config
directory. No network calls, no sudo, no elevated privileges.

Agent roster (24 agents, 7 categories)

v2 consolidation note. The roster was consolidated from 43 agents
in v1 to 24 in v2 by merging provider-specific variants behind
FOCUS-shaped capability agents. v1 agents (e.g. AWS Cost Explorer Analyst, GCP Billing Interpreter) are now covered as
provider-native deep-cuts inside the consolidated agents
(e.g. Cloud Billing Analyst). If you previously installed v1, see
the migration table at the bottom of this section.

cloud-cost/ -- 5 agents

Day-to-day cloud cost analysis. FOCUS-first, multi-cloud by default.

• Cloud Billing Analyst -- FOCUS-shaped reporting and analysis;
  AWS CUR / Azure Cost Management / GCP BigQuery export as documented
  fall-back layers
• Budget & Anomaly Operator -- forecast-based trajectory budgets +
  segment-aware seasonal anomaly detection (covers Budgeting and
  Anomaly Management)
• Forecast & Estimation Analyst -- driver-based forecasts +
  pre-deployment workload cost estimates (covers Forecasting and
  Planning & Estimating)
• Unit Economics Modeler -- cost per tenant, per request, per GB,
  per AI feature -- shipped at GA, not retroactively
• FinOps Benchmarking Analyst -- KPI selection + internal +
  external comparison, FOCUS-normalized

commitments/ -- 2 agents

Rate optimization across all clouds.

• Commitment Discount Strategist -- cross-cloud RI / SP /
  Reservation / CUD portfolio using FOCUS Commitment Discount columns
  (CommitmentDiscountId, CommitmentDiscountStatus,
  PricingCategory='Committed'); Effective vs Contracted savings math
• EDP Negotiation Coach -- private pricing prep, BATNA modeling,
  non-price terms

data-platforms/ -- 2 agents

The pipelines and models behind every cost dashboard.

• FOCUS Data Engineer -- ingest, transform, and validate
  FOCUS-conformant cost datasets across providers; runs the FOCUS
  Validator and Requirements Analyzer
• Cost Warehouse Modeler -- dimensional models with FOCUS as the
  conformed-dimension source of truth; dbt, semantic layer

governance/ -- 7 agents

Practice operations: showback, chargeback, tagging, maturity, policy,
tooling, enablement, migration, allied-discipline coordination.

• FinOps Practice Lead -- governance + maturity assessment +
  allied-discipline integration (ITAM / ITSM / ITFM / Security /
  Sustainability)
• Allocation & Policy Architect -- tag taxonomy + policy-as-code
  enforcement (SCPs, Azure Policy, OPA / Gatekeeper)
• Platform & SRE Cost Lead -- embeds cost in ADRs and PR reviews;
  quantifies the reliability-cost curve and SLO trade-offs
• Showback / Chargeback Architect -- allocation model maturity
  (EffectiveCost-driven; InvoiceId reconciliation)
• Cloud Onboarding Coordinator -- migration-time cost hygiene +
  intake gate
• FinOps Enablement Lead -- training, Champions program,
  onboarding integration
• FinOps Tooling Evaluator -- build-vs-buy, vendor selection,
  FOCUS-conformance filter

kubernetes/ -- 2 agents

Container-level attribution and in-cluster optimization.

• Kubernetes FinOps Engineer -- cluster-level allocation,
  OpenCost / Kubecost, FOCUS-emitting allocation
• Kubernetes Workload Optimizer -- rightsizing (VPA-informed)
  and node-level autoscaling tuning (Karpenter / CAS)

waste-detection/ -- 3 agents

The hunters.

• Idle & Orphaned Resource Hunter -- compute, snapshots, load
  balancers, and zombie NAT Gateways in one inventory
• Cross-AZ Egress Investigator -- network cost hidden across
  storage / database / managed-service categories
• S3 Storage Class Auditor -- object-storage class alignment +
  lifecycle policies

specialized/ -- 3 agents

Higher-leverage niches.

• Workload Cost Optimizer -- ML training/inference + serverless
  • spot strategies (compute-pattern specialist)
• License & SaaS Cost Optimizer -- BYOL, marketplace, entitlement
  audits, FOCUS Provider/Publisher/Invoice Issuer for marketplace
  attribution
• Cloud Sustainability Analyst -- carbon accounting, region-carbon
  trade-offs, demand shifting

v1 → v2 migration map

Playbooks (10 named patterns)

Named-pattern writeups for specific failure modes. Cite them by name.

Cloud resource waste:

• Zombie NAT Gateway
• Snapshot Sprawl
• Cross-AZ Chatterbox
• Idle Load Balancer
• Oversized RDS

Reporting / allocation / governance patterns (FCP-anchored):

• Untagged Spend Drift
• Month-Length Illusion -- the February "win" that isn't
• Masked Anomaly -- real cost blown in by a coincident commitment offset
• Chargeback Revolt -- skipping showback costs 12+ months of credibility

FOCUS adoption:

• FOCUS Adoption -- Parallel Run -- migrate to FOCUS without losing reconciliation or stakeholder trust (STMicroelectronics, GitLab, Zoom, UnitedHealth Group, European Parliament patterns)

Usage examples

Claude Code

> Use the Cloud Billing Analyst agent to audit last month's spend
  increases. Focus on the AWS deep-cut.

Cursor

Reference the Budget & Anomaly Operator rule to design our alerting pipeline.

Any tool
Reference the agent by name in a conversation. The agent's persona and rules
become the working context for that task.

About Cletrics

Cletrics builds real-time cloud cost
observability for teams that want to stop waiting for yesterday's data.
This repo is our open-source contribution to the FinOps community -- use it
freely, with or without Cletrics the product.

Contributing

See CONTRIBUTING.md. Pull requests welcome for new
agents, new playbooks, tooling improvements, and localizations.

Please respect the 90/10 rule: agents and playbooks are generic FinOps
resources. Product references stay minimal.

License

MIT. See LICENSE. Trademark and attribution terms in
NOTICE.

"Cletrics" is a trademark of Cletrics. Fork freely; don't imply endorsement.

Security

See SECURITY.md for the disclosure policy.