AgentsView now supports custom pricing for unlisted models

AgentsView, the token usage analytics tool developed by Wes McKinney, has a specific and predictable problem: when a new model launches, its pricing database doesn't include it immediately. Simon Willison encountered this situation on June 9th and, rather than wait for the app to update, decided to find a workaround and add the price manually.

The result is a brief but useful technical note published on his blog: a step-by-step procedure for registering a model with custom pricing in AgentsView, allowing the interface to continue calculating and displaying costs even when the model isn't yet in the official index.

What AgentsView is and why keeping it calibrated matters

AgentsView is a local application that aggregates token consumption from different code agents running on the same machine. Its value proposition is straightforward: see in a single dashboard how much each project costs, broken down by model and by agent, typically represented as a treemap. For teams or developers working with multiple agents in parallel, Claude Code, LLM CLI, and custom tools, having that consolidated view saves time and prevents billing surprises.

The problem emerges whenever Anthropic or another provider launches a model that AgentsView doesn't yet recognize. Without a reference price, the app can't calculate the cost, and the treemap loses its primary utility. It's not a critical failure, but it is a real friction point in the workflow.

The solution: quick reverse engineering

What makes Willison's case interesting isn't just the trick itself, but the method. By his own account, he used the new model to reverse-engineer AgentsView and understand how it manages pricing internally. From there, he deduced the procedure for inserting a custom price.

The post doesn't detail the complete code in the public summary, but it points to his TIL entry (Today I Learned), where he documents the exact steps. This is the kind of contribution that carries more value than it initially appears to have: when thousands of developers use the same tool and the same new model, someone needs to first document how to bridge the gap. Willison is usually that someone.

Who this is useful for

The audience this knowledge targets is fairly specific:

• Developers using AgentsView to monitor costs of local agents and don't want to lose visibility each time they switch models.
• Teams adopting models quickly and can't afford to wait days for analysis tools to support them natively.
• Anyone using the `llm` CLI or other agents compatible with AgentsView who wants to keep their cost dashboards updated without relying on the app provider's update cycle.

This isn't a tutorial for beginners or a feature that Anthropic controls directly. It's an ecosystem workaround, which is exactly where the most practical tricks usually live.

Cost as a real work metric

The screenshot accompanying Willison's entry shows something worth noting in its own right: over $74 spent on a single project (`prod_datasette_agent`) in the course of just one day. That gives a sense of the scale at which some developers operate with autonomous code agents, and reinforces why tools like AgentsView exist and have demand.

When daily token spending can easily exceed $80 on a single machine, having a broken dashboard because a price is missing is a real problem, not a cosmetic detail.

---

We appreciate that the ecosystem has documentation like this: small, concrete, and published the same day the problem emerged. AgentsView would be more useful if it had a native mechanism for users to add temporary prices while waiting for official updates, but until that arrives, having someone like Willison document the workaround is the next best thing.