TOMC 0.2.0 preview: desktop companion and automatic request processing

TOMC 0.2.0 (research preview)

TOMC now has a desktop companion for Windows and macOS, and on Windows it can process Codex and Claude Code requests automatically.

What's new

• Desktop companion (Windows and macOS). A small cat shows the tokens TOMC saved, their value at each model's listed input price, savings per model and per day. The context library keeps prepared contexts under your original conversation names (Codex thread names, Claude Code titles), with a copyable continuation prompt, export and import for another computer. The analysis report shows savings by conversation and kind of work, how requests were handled, how much each compression removed and context lengths.
• Automatic request processing (Windows, optional). A local gateway on 127.0.0.1 compresses eligible old history in Codex and Claude Code requests before they leave your computer. Requests it cannot shorten safely pass through unchanged.
• Windows installer. py scripts\install_windows_companion.py shows the plan; --install installs the cat; --gateway adds automatic processing and routes Codex and Claude Code only after the gateway answers its health check, keeping exact copies of both settings files; --uninstall <receipt> restores them. On macOS, use scripts/install_macos_pet.py.
• Anonymous feedback. From the analysis report you can send a report as a pull request after reviewing it. It contains kinds of work (coding, writing, papers …) and numbers only: no conversation names, content, folders or IDs.
• MCP runtime 0.2.0 for Claude Desktop, Claude Code and Codex: ten tools, including get_status, list_contexts, get_context and make_handoff. Prepared contexts are saved locally by default; context recording can be turned off.
• API reader: keeps provider-reported cache-token usage; a multi-turn example shows per-request preparation.

Install

• Desktop companion: Windows and macOS guide.
• Claude Desktop: install tomc-memory-0.2.0.mcpb below. Codex: tomc-memory-codex-0.2.0.zip below. Claude Code: install the plugin from this repository's marketplace. Plugin guide

Good to know

• Savings appear only for requests TOMC actually shortens. Codex and Claude Code manage much of their context themselves, short histories already fit, and some content formats are not supported yet; the analysis report shows how many requests fell into each group.
• Automatic request processing is Windows-only for now, and the gateway supports Codex signed in with ChatGPT (not API-key setups).
• Everything stays on your computer; the gateway forwards request bodies and sign-in headers in memory only.
• There is no PyPI package. MIT License.

中文简介

• 桌面小猫（Windows / macOS）：显示节省的 token 和按各模型输入价折算的金额；上下文库按原对话名整理，可复制接续提示词、导出到另一台电脑；分析报告按对话和工作类型统计节省。
• 自动请求处理（Windows，可选）：本机网关在请求发出前压缩 Codex 和 Claude Code 中符合条件的旧历史。
• 一条命令安装：py scripts\install_windows_companion.py --install，加 --gateway 开启自动处理；可按回执卸载并恢复设置。
• 匿名反馈：只提交工作类型和数字，提交前预览完整内容。

TOMC assistant packages 0.1.1 (research preview)

TOMC assistant packages 0.1.1

This research preview distributes the existing MIT-licensed 0.1.1 assistant packages from source commit a6c75f2e4af4eed1528c0923176567766f52956c as versioned release assets. Package bytes are unchanged from the checked-in download files. The compiler core remains 0.1.0.

The packages include the host-managed conversation policy: task notes require explicit save and recall. TOMC does not automatically intercept native assistant conversations or provide cloud synchronization.

The release includes MCP Registry metadata in server.json. A registry receipt is attached only after the public registry returns the matching version and package hash. No new model benchmark is claimed.

Installation | Validation

TOMC assistant preview.6 - MIT source and plugins

TOMC is an open-source research project for preparing compact task context and saving project notes in Claude Code and Codex. Preparation runs on CPU through an MCP server or Python API.

This preview fixes the release snapshot at commit f595b0a86e6038ce918a96cb642c480f0fe0fa2a. The tagged source and the attached assistant packages use the MIT License, including commercial use under its terms. The earlier preview.5 source tag is preserved as a historical version under its original license; its updated MIT attachments should not be confused with that source tag.

The assistant packages are unchanged from the verified MIT packages attached to preview.5. This release does not claim a new assistant installation test or new benchmark result. It also corrects outdated private-staging text to reflect the author's already-authorized public research preview.

• Installation and usage
• Project page and English demo
• Validation and limitations

Use prepared context in place of the original history, or pass it to a new session. Savings depend on the task and budget. Saved notes are recalled explicitly; this plugin does not reset usage limits or intercept every assistant request.

Package checksums:

c4ea4aae5ec5995a72f3e0891e0708f880035ff7dc7e3d91f57393de3c8ad4ac  tomc-memory-0.1.0.mcpb
a7b3aa22451f4769cfccfff47d00c6cb09e5782af45e048160823c7fcf131c34  tomc-memory-codex-0.1.0.zip

TOMC assistant preview.5 — public release, MIT License

First release from the public TOMC repository.

• License: TOMC is released under the MIT License. It is free to use, including commercially; modification, redistribution and integration into other projects are allowed, and copies or substantial portions must retain the copyright and license notice. Both bundles include the MIT LICENSE.
• Default budget: when no budget is given, TOMC keeps about 80% of the history (at least 1,024 estimated tokens, so short histories stay intact). Explicit budgets work as before.
• Clients: Claude Desktop opens the .mcpb; Claude Code registers the extracted .mcpb with claude mcp add; Codex users extract the ZIP and run install.py. Replace an earlier installation to get this version.

This package's checks: both bundles match the files and SHA-256 build receipts in plugins/downloads, and their LF checksum files pass sha256sum -c. The Claude bundle passed the packaged MCP smoke test; repository CI passes on Ubuntu and macOS (Python 3.10 and 3.12). Model calls and GUI installation were not repeated for this build.

中文：这是公开仓库的首个版本。TOMC 以 MIT 许可证发布：可免费使用，包括商业用途，允许修改、再发布和集成到其他项目，复制或分发时须保留版权声明和许可证。不指定预算时默认保留历史的约 80%。请用新包替换旧安装。