SpendFence

Who it's for

Platform/DevOps engineers at 5-100 person software companies who run disposable Docker sandboxes or self-hosted CI runners; today they cope with docker run resource flags, Docker stats, and cloud billing alarms.

The problem

Payment: runaway CPU and egress charges plus no per-task chargeback. Time: manually correlating a surprise cloud bill to one container/task.

How to build it

CLI plus policy YAML, local daemon, and minimal web dashboard; integrates with Docker via container labels, writes JSON/CSV, no Kubernetes operator in v1.

How it makes money

Platform teams pay $199-$499/month because one runaway workload can cost more than that and the ledger supports customer chargeback; free Docker stats/cloud alarms show usage but do not enforce per-task limits or produce invoice lines.

Why it doesn't exist yet

Incumbents skip: Docker's sandbox push is focused on isolation and developer ergonomics, not tenant budgeting/chargeback. Indie gap: cgroups, iptables accounting, and container metadata are enough to ship a small policy runner and ledger without building a container platform.

First users

First 10 users come from HN/Docker Sandbox early adopters and self-hosted runner maintainers who have hit surprise egress bills; offer the open-source runner free and charge for hosted ledger/dashboard.

Build size

2 people x 8 weeks; includes Linux Docker enforcement for CPU/runtime/egress, policy YAML, JSON ledger, CSV export, simple dashboard; excludes Kubernetes, Windows/macOS, GPU budgets, auto-remediation beyond kill.

Biggest risk

Docker ships native per-container budget enforcement and a cost ledger in Docker Sandboxes within six months.

Conditions for a hit (all 3 required)

• A policy file can specify max_cpu_seconds, max_egress_bytes, and max_runtime_seconds; the runner terminates the workload when any limit is exceeded and records which limit was violated.
• Each completed or terminated task emits a JSON ledger line containing task_id, container_id, start/end timestamps, cpu_seconds, egress_bytes, and estimated_cost_usd using a user-supplied rate file.
• A CLI command exports a CSV of the top 50 tasks by estimated cost for a selected date range, one row per task, without requiring a proprietary database.

How it's judged (in 6 months)

GitHub repository implementing these features reaches 1,000 stars(judgment date 2027-02-11)

AI self-confidence 60/100 — self-reported likelihood of meeting the criterion, not a business success rate