Zipeng Wu

Researcher in temporal representation

[Zipeng Wu @zipeng365]

How to save tokens in Claude Code and Codex with a free, open-source plugin

I've released TOMC, a plugin from my research on context compression. It aims to reduce input tokens while preserving answer quality.

Feedback and collaboration welcome.
https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@Hesamation Those long debugging runs sound like a useful test for TOMC, my free open-source research plugin. It aims to reduce input tokens by compressing supplied context for Claude Code/Codex; would you be open to comparing input tokens and answer quality? https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

@Youssofal_ A $200 Pro plan and the token meter still says “nice try” 😭 I made TOMC to trim the history you carry into the next task: https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@CodexResets1 10% left is when “one tiny fix” becomes a horror movie 😅 I made TOMC to trim the history you carry into the next task: https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@theo 200 “watch PR” scripts... the agents started a babysitting service 😅 I made TOMC for saved task notes and lighter context for the next task: https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

@cyrilXBT I’d like to test context compression alongside your model routing. TOMC is my free, open-source Claude Code/Codex research plugin, aimed at reducing input tokens; would you try it and compare answer quality on the same tasks?

https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@chetaslua Would you be open to testing TOMC with the history from a run like this? My free, open-source Claude Code/Codex research plugin compresses context to reduce input tokens, and I’d welcome a comparison of output quality.
https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

@cherry_mx_reds Specific handoffs help the next agent keep the constraints straight. I built TOMC to prepare compact task context from history you supply and reuse saved notes across Codex and Claude Code. CPU-only: https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@lydiahallie That 'one tiny follow-up' carrying the whole chat is the pain point. I built TOMC to prepare compact context from supplied history for the next task, plus saved notes for Codex/Claude Code. CPU-only, open source: https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

Same task. Less context baggage.

I built TOMC to compress history you provide for the next task and reuse saved notes across Codex and Claude Code. New computer? Bring the notebook.

CPU-only, open source: https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@mirku21 I’d like to test whether those auth and schema requirements survive context compression. TOMC is my free open-source Claude Code/Codex research plugin for reducing input tokens; I’d welcome a comparison with your workflow.

https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@swordBluesy @CodexResets1 Thanks for checking this; please compare the prepared text with the original and test the answer in a fresh session. If either detail gets lost, a small redacted example and the budget would help me reproduce it.

https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@Bober_smart Would you be up for testing TOMC with that decision-log workflow? My free open-source research plugin compresses context for Claude Code/Codex to reduce input tokens, and reuses saved notes through one notebook.

https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

Claude Code / Codex: “one more tiny task.”
My token budget: 💀

TOMC is my free open-source research plugin: compress context, save input tokens, reuse saved notes. Context can still fill; transfer between computers is manual.

https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

@Suryanshti777 Would you consider reviewing TOMC, my free open-source research plugin for Claude Code/Codex? It compresses supplied context to reduce input tokens, with task notes shared through one notebook or transferred manually between computers and servers.

https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

@ramxcodes I’d be curious whether the design requirements survive a handoff. My open-source Claude Code/Codex research plugin compresses supplied context to reduce input tokens while aiming to preserve answer quality; feedback from an animation task would help.

https://t.co/K5ggSvl1gi

[Zipeng Wu @zipeng365]

Here's the 30-second demo. Ask Claude Code or Codex to read the repo, help you install TOMC, and walk through an example. Supply the history you want to compress, then use the prepared context in a new chat.

https://t.co/K5ggSvktqK

[Zipeng Wu @zipeng365]

Please, just a few more tokens. I thought trying Claude Code once wouldn't get me hooked.

My free, open-source research plugin compresses context for Claude Code and Codex to save input tokens while aiming to keep answer quality.
https://t.co/K5ggSvl1gi
#ClaudeCode #Codex

[Zipeng Wu @zipeng365]

@VeraYugen @X hi，lets connect

[Zipeng Wu @zipeng365]

@fleyta88 Keeping routing decisions through a handoff is a useful test. I’ve open-sourced Claude Code/Codex research plugins to compress context, aiming to reduce input tokens while preserving quality; feedback on real tasks would help.

https://t.co/K5ggSvl1gi
https://t.co/C5nEIsCLn1

[Zipeng Wu @zipeng365]

@daniel_mac8 The cache-reset cost seems worth measuring over a full task. My open-source Claude Code/Codex research plugins compress context for a new session, aiming to reduce input tokens while preserving answer quality; I’d welcome comparisons.

https://t.co/K5ggSvl1gi