Reported News

TokenEconomizer: a Microsoft Research plugin that learns from past sessions to cut coding-agent costs

TokenEconomizer is an Apache-2.0 plugin from Microsoft Research, Microsoft Security and UC Berkeley for Claude Code, Codex, the GitHub Copilot CLI and Pi that learns from earlier sessions in a repository to cut waste; its README reports "18.4–30.9% lower mean cost across five agent–model combinations" on SWE-Bench tasks. It had about 3.2k GitHub stars when we checked on Nov 1, 2026.

Share on X
Changes reported
6
listed below, each from a source we read
We tested
0
we did not run it ourselves
Gaps we flag
3
things we could not verify
Sources
2
2 primary, linked below

What changed

Changes the sources report, with the evidence status of each
#What the source saysStatus
01TokenEconomizer is a plugin for Claude Code, Codex, the GitHub Copilot CLI and Pi, with experimental opencode support. It leaves the agent's model and code alone and instead learns across the sessions an agent has already run in a repository.Reported
02Micro-skills: it records repository facts (a build step that must run first, the command spelling that works, a needed environment variable) and states a fact at session start once three sessions have confirmed it, dropping it when later sessions contradict it.Reported
03Tool pruning: a loaded tool that goes unused over 10 prompts across at least 2 sessions, or over 5 sessions, is left out of later sessions through the agent's own settings, and one command brings it back.Reported
04Loop guard: when a turn keeps hitting the same failures, a judge model checks whether it is stuck and, at 0.7 confidence or more, adds one redirect note; the README says it never stops a turn itself.Reported
05Its own model calls run through your existing agent login and are counted in its cost reports; it needs Python 3.10 or newer, no separate API key, and keeps what it learns in ~/.tokeneconomizer on your machine.Reported
06The README lists tested version ranges (Claude Code 2.1.300 to 2.1.310, Codex 0.153.1 to 0.164.0, Copilot CLI 1.0.83 to 1.0.102, Pi 1.2.0 to 1.3.0) and says newer versions have not been tested yet.Reported

reported by the sourcemeasured or run by usnot verified

Primary source: TokenEconomizer repository and README (GitHub). We are the fast, independent index; the vendor’s own page is the source of record for this release.

In the source’s words

“18.4–30.9% lower mean cost across five agent–model combinations”

“Coding agents are built to finish tasks, not to keep them cheap.”

Early verdict

Worth trying if you run Claude Code, Codex or Copilot CLI on the same repositories every day: it targets repeated rediscovery, unused tool definitions and stuck loops, and every action is reversible per the README.

Basis: Based only on the project's README as read on Nov 1, 2026; we have not run it.

What we have not verified

  • We did not install or run the plugin; all cost and Pass@1 figures (SWE-Bench, Web-Bench and the security-auditing benchmarks) are the authors' own results from the README and paper.
  • We did not check whether it works on agent versions newer than the tested ranges, which the README says are untested.
  • The repository was created on Oct 13, 2026; we first saw it on Nov 1, 2026 (UTC) when it surfaced in a GitHub new-repos sweep, so this is not a same-day report.

This brief is Reported: written from the sources below, not from running the release ourselves.

Sources

Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/microsoft-tokeneconomizer-coding-agent-cost-plugin.md

Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.

The Weekly Verdict · free

5 new-model use cases. Every Monday.

A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.

Free. Double opt-in. One click to unsubscribe.