Splunk Open Sources Token Meter Tool for Application Developers

Splunk, a Cisco company, has released an open-source tool called Token Meter for application developers. The tool, which runs on Linux, macOS, and Linux machines, reads local Claude Code and Codex logs to provide real-time cost and activity information for artificial intelligence (AI) agent sessions.

Token Meter analyzes the trace files written by AI agents to disk and compares them against public model rates. This allows developers to understand the cost of long-running sessions or tasks that could potentially be performed using a less expensive AI model instead. The tool presents key metrics such as output speed in tokens per second, time spent waiting on the model, the balance between fresh input and generated output, and the amount returned by the AI agent after a tool call.

Developers can also identify current, completed, or historical sessions using filters based on runtime, model, state, and time. They can further analyze the data by tracing executions, events, tool activity, context growth, retries, failures, coverage, and warnings about session statistics. Comparisons can be made across agents, models, days, sessions, and tools for tokens, estimated cost, timing, context, executions, and tool activity.

The dashboard displays a timeline of raw events, a tools tab for usage statistics, an insights tab for derived signals, and an alerts tab for budget state. Token Meter sends notifications when a run exceeds a set threshold or when costs spike.

Hao Yang, vice president and head of AI at Splunk, emphasized that as AI technology advances, DevOps teams should strive to balance the use of AI models with the use of context engines to minimize token consumption. Eventually, AI agents' data access and processing will become more decentralized, with AI model service providers working to help organizations control costs.

While the cost of individual tokens is declining, the number of tokens consumed is increasing exponentially. Organizations are now spending significantly more on AI than initially anticipated. A survey by the Futurum Group found that nearly half of organizations are exceeding their AI budgets. Therefore, DevOps teams must focus on reducing AI costs as application developers increasingly rely on AI coding tools.

The challenge lies not only in reducing costs but also in ensuring that the allocated funds for AI are spent effectively to deliver the most business value to the organization.