Agent Cost Lens

AI coding cost is spread across multiple tools, providers, repos, models, retries, and failed runs. By the time finance asks, it's hard to explain.

Agent Cost Lens collects local cost records (or estimates them when exact usage is missing) and rolls them up by repo, model, provider, day, and developer. Catch expensive runs and retry waste before they become a quarterly problem.

• How much did agent runs cost today?
• Which repo is the most expensive?
• Which model or provider is driving spend?
• How much cost came from retries?
• Which runs failed after spending tokens?
• Which tasks should use a cheaper model?

One command. No account. No telemetry.

1# install (binary is `agent-cost`)
2npm install -g @agentopssec/agent-cost-lens
3 
4# today's spend
5agent-cost today
6 
7# break it down
8agent-cost by-model
9agent-cost by-repo
10agent-cost expensive-runs
11 
12# set a daily budget
13agent-cost budget set 50

A real summary from the last 24 hours of agent activity.

One record per run. Imported from Agent Flight Recorder when available.

1{ "runId": "run_001", "agent": "codex",
2  "repo": "dashboard-app",
3  "provider": "openai", "model": "gpt-5.2",
4  "inputTokens": 18400, "outputTokens": 2400,
5  "toolCallCount": 12, "latencyMs": 252000,
6  "retryCount": 2, "estimatedCost": 0.42,
7  "taskLabel": "fix failing tests" }