Token Forecaster · House #21,883

What it is

Calls itself “forecaster”.

• Calls itself: forecaster

“Token Forecaster estimates how many output tokens a Claude reply, or a whole agent turn, is likely to use, then shows how the reply is tracking against that range while it runs.”

What it costs

A free tier, free forever. No monthly or yearly plan. No one-time option.

• One-time: no
• Monthly: no
• Yearly: no
• Free plan: free forever

“Open source, MIT.”

“…docs/PLAN-OF-ATTACK.md -> https://github.com/Sumcap/token-forecaster/blob/main/docs/PLAN-OF-ATTACK.md”

Who it sells to

Sells to developers.

• Sells to: developers

“It starts from a bundled Claude Code prior and can fit itself to your own local history.”

How it runs

A desktop app and browser extension for macOS. Runs locally and works offline. Open source. AI is the core of the product.

• Kind: desktop · extension
• Runs on: macOS
• Runs where: on the device, offline
• Open source: yes
• AI role: core to the product

“Token Forecaster runs on four surfaces, all fed by one local daemon: the Claude Code status line, a macOS menu bar app, a local dashboard, and a Chrome extension for claude.ai/code.”

“Learns from your local history.”

What it works with

Integrates with Claude Code and Codex. Runs inside Chrome and Claude Code. Built on Claude and Claude Code. Plugin for Chrome and VS Code.

• Integrates with: Claude Code · Codex
• Runs inside: Chrome · Claude Code
• Built on: Claude · Claude Code
• Plugin for: Chrome · VS Code

“Chrome extension. A pill above the composer on claude.ai/code”

“Token Forecaster estimates how many output tokens a Claude reply, or a whole agent turn, is likely to use”

What it offers as proof

In beta. Claims “90.6% of 4,146 unseen calls”. Try it without signing up. Publishes docs.

• Stage: beta
• Number claimed: 90.6% of 4,146 unseen calls
• Trial asks for: nothing

“Research preview: the bundled calibration is fitted on one user's Claude Code history.”

“Open source, MIT.”