Web Workspace
Onboarding, dashboards, AWS account setup, tenant configuration, approval queues, and audit history.
Arclight
AWS FinOps AI Agent for investigation and controlled action.
Arclight gives finance, platform, and operations teams one interactive runtime for AWS cost questions, anomaly analysis, recommendations, approval workflows, and governed remediation.
Interaction Model
Arclight works through Web, Slack, and MCP.
Slack Operator
Alerts, summaries, quick RCA, cleanup proposals, approval buttons, and execution receipts in the team workflow.
MCP Runtime
Connect Claude, Codex, Cursor, and custom agents to a governed Arclight runtime for deeper multi-turn investigation.
Agent Loop
From question to evidence to action.
A user asks a cost question, Arclight clarifies intent, plans the investigation, invokes AWS billing and operational tools, gathers evidence, explains the finding, proposes action, waits for approval when needed, executes, and reports the result.
Capability Landscape
Built for real FinOps operating work.
Investigate
Cost anomaly RCA, service-level drivers, account spikes, resource inventory, and evidence trails.
Detect
Idle resources, unused reservations, underutilized compute, storage waste, and unexpected usage changes.
Recommend
Rightsizing, cleanup, lifecycle policies, reserved capacity options, and architecture improvement paths.
Forecast
Monthly run-rate, budget risk, expected savings, seasonal patterns, and optimization impact estimates.
Act
Approval-gated cleanup, tagging changes, snapshot review, non-production shutdown, and controlled remediation.
Govern
Audit history, tenant isolation, policy guardrails, permission boundaries, and change traceability.
Slack
Approval-based automation in the channel.
Arclight can post a cost anomaly alert, answer follow-up questions, investigate root cause, propose cleanup or rightsizing, wait for approval, execute a controlled action, and post the audit result.
MCP
Open runtime for power-user workflows.
MCP lets external AI clients connect to Arclight, use governed tools, and run deeper interactive investigations without turning Arclight into a closed chat bot.
Enterprise Runtime Isolation
Designed around customer separation and delegated AWS access.
The target architecture uses a shared control plane with per-tenant runtime accounts, delegated customer access, role assumption boundaries, audit logging, and approval gates for actions that can change infrastructure.
Shared control plane
Per-tenant runtime account
Delegated customer access
Human approval gates
Pilot