Tabellio

Name: Tabellio
Author: yffo

by yffoversion 0.1.12i★ –iRuns on your computer

Registry nameio.github.yffo/tabellio

Install

Overview

Description from the maker (original English)

Local-first memory for AI agents: markdown source of truth, budget-aware recall, write gate.

PopNx MCP’s three pillars

These three are never merged into one score: opinions, measurements and popularity are different things.

User voiceiNo reviews yetReviews open together with accounts in a later step
Our basic checksi1 of 1 passedChecked from public information; code not scanned yet
UsageiNot measured yet★ –

What you can do with it

We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.

Connection: Runs on your computer
Transport: stdio
Category: Search & AI memory

Install

Pick the AI app you use, then copy the settings.

Before you install

Basic checks: 1 of 1 passed
Runs on your computer with your user’s permissions (files, network, apps).
You fill in yourself: TABELLIO_STORE

You need

npmNode.js installed (for the npx command)

Which app do you use?

ClaudeClaude CodeCursorVS CodeChatGPT

Pick your app above. We remember it, so next time the button says “Install in …”.

Open Claude Desktop → Settings → Developer → Edit Config, and paste this into claude_desktop_config.json

{
  "mcpServers": {
    "tabellio": {
      "command": "npx",
      "args": [
        "-y",
        "tabellio"
      ],
      "env": {
        "TABELLIO_STORE": "your-value"
      }
    }
  }
}

Run this in your terminal

claude mcp add tabellio --env TABELLIO_STORE=your-value -- npx -y tabellio

Fill in the empty values before running it.

Open in Cursor

Or copy the settings yourself

Paste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json

{
  "mcpServers": {
    "tabellio": {
      "command": "npx",
      "args": [
        "-y",
        "tabellio"
      ],
      "env": {
        "TABELLIO_STORE": "your-value"
      }
    }
  }
}

Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.

Open in VS Code

Or copy the settings yourself

Paste into .vscode/mcp.json in your project

{
  "servers": {
    "tabellio": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "tabellio"
      ],
      "env": {
        "TABELLIO_STORE": "your-value"
      }
    }
  }
}

Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.

This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.

Your choice stays on this device in a cookie; no account needed.

Values you fill in

TABELLIO_STORE env
Optional path to the memory store directory (default: ~/memory/tabellio).

These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.

What it can do

Runs on your computer; the tool list appears in your AI app after you install it.

Basic checksi

We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 7, 2026

✓ Pass
Maker matches the repo
The registry name was verified with the GitHub account yffo, which owns the repo
… Not checked yet
Still maintained
We have not read this repo from GitHub yet
… Not checked yet
Has a licence
We have not read this repo from GitHub yet
– Not applicable
Endpoint responds
This one runs on your computer; there is no internet endpoint
– Not applicable
Sign-in with OAuth
Runs on your computer and uses local settings instead
… Not checked yet
No hidden instructions in tool descriptions
Runs on your computer; we can only read tool lists from internet endpoints

Compared with others in Search & AI memoryi

More stars than 0% of the other 7251 servers in this category
Published to the registry more recently than 69% of the other 7251 servers in this category
Basic checks equal to or better than 100% of the other 7251 servers in this category

Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.

Common questions

Is Tabellio safe?

It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.

What keys or codes does it need?

The registry does not list a required key, though some servers ask you to sign in on first use.

Does it run on my computer or in the cloud?

It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).

Is it still maintained?

The GitHub repo last changed –, and the latest version (0.1.12) reached the registry 17 days ago.

Which apps can use it?

Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Node.js on your computer.

What can I use instead?

In the same category: Context7, Codebase Memory, Exa

Alternatives

Servers in the same category from other makers, sorted by stars.

Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi

Context7

by upstash

Feeds AI the latest docs for libraries and frameworks before it answers, so code it writes does not use outdated, removed functions.

Computer or cloud★ 64,224MITUpdated today

Install

Codebase Memory

by DeusData

Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.

Runs on your computer★ 48,148MITUpdated 20 days ago

Install

Exa

by exa-labs

Connect AI agents to Exa for web search, content fetching, and multi-step research.

Runs in the cloud★ 5,213MITUpdated last month

Install

README from the maker

Open the README

Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.

tabellio

罗马的契约代书人：公开立契，且必须留存副本以备日后核验。

一个通用记忆服务，让 agent 记住该记的事 —— 并且答得出「这条记忆谁写的、依据什么」。

tabellio 给 agent 用的长期记忆：本地优先、纯文本落盘、成本可控、可审计、可带走。一次落盘，任何 CLI / IDE / 自研 agent 通过 MCP 共用同一份记忆。

npm i -g tabellio

tabellio setup --write --verify # 写 MCP 配置，再用宿主自己的命令验证（默认 dry-run） tabellio hook install --host claude # 可选：装 Claude Code 会话结束钩子（默认 dry-run） tabellio hook install --host claude --write tabellio hook install --host claude --session-start --write # 再加「待批准候选」提示 tabellio schedule install --write # 可选：macOS launchd 每 6 小时自动发现近似重复（只产提案） tabellio schedule status # 看定时任务在不在、launchd 里有没有加载 tabellio status # 建库 + 看一眼 tabellio doctor # 只读体检：库 / 宿主配置 / 钩子 / 定时任务 tabellio recall --query "编辑器偏好" # 按预算取回记忆 printf '%s\n' '# 发布流程' '' '发版前跑 npm test 和 node bench/run-eval.mjs。' \ | tabellio capture --file - # 先收成候选，不静默入库 tabellio inbox --accept <候选 id> # 人看过、批准后才进主库 tabellio consolidate --auto --write # 自动发现近似重复，只写待批准提案（不改笔记） tabellio inbox --accept <提案 id> # 批准后才合并；可 --undo 撤销 tabellio-mcp # 给宿主用的 MCP stdio 服务器

零运行时依赖，Node ≥ 20.11，不需要 npm install。

tabellio-mcp 默认只读，工具面是 recall / search / read / list / status；recall 一次返回预算内的记忆正文，search 只给指针。加 --allow-write 才挂两个写工具：remember 直接写正式记忆，capture 只写候选、由人在 CLI 批准。

setup 只认 Claude Code / Cursor / Codex 三家，写前备份、默认只预览，已有的不同配置要显式 --force； --verify 会调用宿主自己的 CLI：Claude Code 报 Connected 才算连上；Codex 只接受配置、不做连接检查； Cursor 没有无头检查，需要 GUI 确认。找不到宿主 CLI 不会误报成配置失败。 DSH 的 YAML loader patch 还不敢让程序改。capture / inbox 是「自动记忆」的诚实版本：默认不会从对话里自动抽取；只有你显式装 Claude Code 的 SessionEnd 钩子（tabellio hook install --host claude --write），会话结束时才会从用户发言里抓显式线索（「以后都用…」「记住…」「我偏好…」），生成候选，不写主库。候选仍要在 inbox 里由人批准，批准时才走与 remember 同一条写入闸门。加 --session-start 后，下次会话开始会提示待批准候选数与自动巩固提案数（只读，不写库）。钩子只支持 Claude Code、默认不装、不联网、最多 3 条/会话；它可能漏、也可能误，也不常驻后台、不会自动改笔记。macOS 上可以再装一个可选的 launchd 定时任务（tabellio schedule install --write），每 6 小时跑一次自动发现；它同样只产提案，执行仍要人批准。 tabellio doctor 是只读体检：把库的解析失败/索引陈旧/待批准候选、宿主 MCP 配置是否指向当前库、配置里的 Node/MCP 入口是否还存在、Claude 钩子和定时任务的状态一次列出来，并给下一步命令；它不会替你修任何东西。

它到底是什么

一条事实一个 .md 文件，frontmatter 记着 kind / trust / tags / 来源与有效期； index/ 全是生成物，删掉能重建。当前（M1 + M2：读取回执 / 巩固 / 撤销 / 自动发现提案 / 可选定时发现 / 只读体检 doctor）没有数据库、没有常驻后台进程、没有网络调用；可选的 macOS launchd 定时任务只是到点跑一次扫描、只产提案；常驻核心与自动执行仍在路线图里（docs/06 §2/§8）。

默认 pull，push 例外：agent 要什么就问什么；always-on 的只有极小核心档。
文件是唯一真相源：任何时刻可 cat、可 git、可带走；索引可以丢，事实不能丢。
写入过闸门：去重、长度、信任分级、破坏性改写，全部 fail-closed；拒绝是正常结果。
预算感知：调用方给 token 预算，服务端裁剪并说明裁掉了什么。
可解释（M1 + M2：回执 / 巩固 / 撤销）：recall 返回分数拆解、来源与 budget.dropped；加 --receipt / --save-receipt 会把这次判断落成可核验回执（不含正文，只记 noteSha256），explain 逐条解释纳入 / 未纳入 / 预算丢弃，verify 核对回执是否被改、库是否漂移。consolidate 默认只出计划，--apply 才写元数据并留巩固回执，--undo 可整单撤销；当前支持 invalidate（按 supersedes 失效）与显式 archive（只改 status，不写有效期），两者都只动元数据。--merge / --summarize / --promote 默认只出内容类提案；加 --apply 会写目标，merge/summarize 还会把源写成 valid_to + superseded_by（正文不动），promote 只写候选、不关源；内容类 run 有独立回执与 --undo。自动发现（consolidate --auto、Claude 钩子或 macOS launchd 定时任务）只产待批准提案，不直接改笔记；写入回执、自动执行与常驻后台巩固仍未做。

用法、命令清单（含 doctor 的体检项）、--json 的冻结形状、库布局：docs/07-usage.md。

为什么要新起一个项目

前作 dsh-engramory-recall（Engramory 格式 DSH 插件）工程上完成度不低（839 测试、已发布、有面板与 MCP 只读视图），但它回答的是一个错的问题：把「每轮往上下文里塞几条指针」当成记忆的主要形态。它自己的账本给出了证据：

指标	实测
注入	131 轮、655 条指针
基础打开率（代理指标）	2.4%
携带 ≥8 次、一次没被翻开	23 条
能不能证明「让回答变好」	不能（指标只到「后来被打开」）

结论：push 式注入的信息密度太低，而且没有评测就没有方向。tabellio 当前已落地的不同：默认 pull、评测先行、索引分层、可核验回执；巩固已开工（显式 consolidate：invalidate + archive + 回执 + 撤销；merge / summarize / promote 默认提案、--apply 执行且有内容类回执与撤销；--auto 自动发现近似重复、只产待批准提案；可选 macOS launchd 定时任务只是定期触发这条发现），自动执行/自动巩固仍是 M2 计划。

但这不是替代关系（2026-10-16 的项目决定：旧插件不许冻结 —— 「一个不变的项目没有任何意义」）：旧插件继续迭代，新能力优先在它身上验证；通用化与跨宿主的部分进本项目。两条线共享一个 memory-core 内核。

文档导航

文件	内容	状态
docs/00-brief.md	立项书：目标、非目标、判据、里程碑、风险	已定稿
docs/01-landscape-systems.md	竞品架构调研（mem0 / Letta / Zep / Cognee / …）	已完成
docs/02-evaluation.md	评测方法与自建评测集方案	已完成
docs/03-hosts-and-interfaces.md	宿主支持、MCP 现状、分发渠道、进程模型	已完成
docs/04-cost-model.md	模型选型与月成本模型（三档方案）	已完成
docs/05-agi-direction.md	记忆在 AGI 方向的前沿与可下注之处	已完成
docs/06-architecture.md	架构设计（分层、闸门、账本、检索）	草案（M1 + M2 读取回执/巩固已实现）
docs/07-usage.md	用法：十六个命令、MCP、候选闭环、自动巩固提案、会话钩子、定时任务、只读体检、回执、巩固、库布局、退出码	已定稿
docs/08-eval-runbook.md	评测跑法与判据	已完成
docs/09-naming.md	命名调研（为什么叫 tabellio）	已完成
docs/11-semantic-recall.md	语义召回进核心：索引、合并、生命周期与信任过滤	已实现（0.1.7–0.1.8）
docs/12-conformance-and-tuning.md	适配器一致性自检（22 项）与评测门控调参	已实现（0.1.9）

状态口径（2026-10-19）：M1（CLI + MCP + 检索闸门 + 导入 + 判据 1 评测）已完成；M2 已开工（读取回执 / explain / verify / 显式巩固 consolidate（invalidate / archive）/ undo 已实现；合并、摘要、提升的提案层与执行层（--apply + 内容类回执 + undo）已实现；setup + capture / inbox 上手闭环已实现；Claude Code SessionEnd 钩子已实现、默认不装、只生成候选；--session-start 可在会话开始提示待批准候选（只读）；自动发现（consolidate --auto / 钩子）只产合并提案、仍要 inbox 批准；macOS launchd 定时任务（schedule）只是定期触发发现、默认不装、只产提案；doctor 提供只读体检、不自动修复；反馈排序与遗忘衰减（recordFeedback / applyFeedback / decay，默认关闭、只影响排序）已进核心；语义召回的核心部分（indexSemantic / semanticRecall / mergeSemanticItems，embedding 由宿主注入）已进核心；适配器一致性自检（tools/conformance.mjs，22 项）与评测门控调参（tools/tune-recall.mjs，只给建议不自动落盘）已提供；自动执行、自动衰减开关、写入回执与常驻后台巩固未做）；M3 分发侧完成（npm 包 + MCP registry），Claude Code 已实测 connected、Codex 只到 configured、Cursor 仍 unverified，三宿主录屏/日志未齐。调研文档截至 2026-10；2026 H2 的模型原生记忆与托管服务尚未纳入。

纪律（一开始就定下来，免得重蹈覆辙）

判据先行：三条必须满足的判据没定稿、没跑出基线之前，不写产品代码。
评测是产品的一部分：每个里程碑都必须能跑一次端到端评测，并留下结果；没有数字的功能等于没做。
旧插件是同伴不是遗产：两条线并行、共享 memory-core；它的库（72 篇）与账本（.engramory-recall.jsonl）直接当迁移样本与评测输入。新系统要在评测上赢过它 —— 赢的是长板，不是把它判死。
数据可带走：纯文本（或可导出为纯文本）永远是真相源。
成本是约束不是事后优化：默认档必须便宜到「不心疼」，贵的档必须可关。

Tabellio

Overview

PopNx MCP’s three pillars

What you can do with it

Install

Before you install

Values you fill in

What it can do

Basic checksi

Maker matches the repo

Still maintained

Has a licence

Endpoint responds

Sign-in with OAuth

No hidden instructions in tool descriptions

Compared with others in Search & AI memoryi

Common questions

Alternatives

Context7

Codebase Memory

Exa

README from the maker

tabellio

它到底是什么

为什么要新起一个项目

文档导航

纪律（一开始就定下来，免得重蹈覆辙）

License