OpenTSC

Name: opentsc
Author: opentsc

A private, offline memory for the people, projects, and decisions you juggle — that also keeps score of whether your judgment was right.

🌐 English · 中文

by dashen (「AI 最严厉的父亲」) · dashen.wang · @dashen_wang

· · Commercial license · Whitepaper (the "soul") →

Changelog · Migrate v1→v2 · Usage guide

What is OpenTSC?

You deal with a lot of people and moving parts. Your notes are scattered, your memory is unreliable, and you rarely check whether your gut calls actually pan out. OpenTSC fixes that:

📝 You jot down what happened — "Carol delivered the quote on time", "Dave keeps stalling on the GPU order."
🗂️ It organizes everything into a searchable memory — every fact tagged with where it came from and how trustworthy it is.
💬 You ask plain questions — Who reliably delivers? Have I dealt with a situation like this before? What am I still waiting on, and from whom?
🎯 You log predictions; it later tells you your hit rate — so your judgment actually improves over time instead of quietly drifting.
🔒 Everything stays on your own computer. Offline-first, single-user, no cloud, no account.

It is not a CRM (it's about judgment and operations, not storing contacts) and not a social network. Think of it as a second brain for your working relationships and operations — one that holds you accountable to reality.

What can you actually do with it?

You want to…	Plain-language ask → it runs
Record something that happened, with evidence	"log that Carol delivered on time" → `event-add`
Find people or situations by meaning, not keywords	"who's reliable on delivery?" → `index-search`
Avoid creating a duplicate record of the same person	"is there already a Mike?" → `identity-resolve`
See what's gone stale or overdue	"what's rotting in my to-do list?" → `actions-stale`
Score your own predictions over time	`stage-prediction` → `calibrate` → `accuracy`
Spot patterns in your own notes	"who's been negative lately?" → `index-mood`

Examples use placeholder names (Carol, Dave…). OpenTSC ships with zero real data. Use it responsibly and only on information you're allowed to hold — see SECURITY.md.

How it works (30 seconds)

OpenTSC is a skill for AI coding agents (Claude Code, and others) plus a Python command-line tool. You talk to it in plain language through your agent; it stores everything as plain Markdown files you own, and builds a fast, semantic search index on top.

python skill/scripts/opentsc.py --root my-vault init
python skill/scripts/opentsc.py --root my-vault event-add B2 "Carol delivered the quote on time" "meeting note" --link p_carol
python skill/scripts/opentsc.py --root my-vault index-search "who is reliable" --kind entity

Optional power-ups (all opt-in — the core runs with zero extra installs): Chinese word segmentation (jieba), emotion scoring (lexicon / local model / LLM), and a vector memory index (zvec). See the usage guide.

Quick start

Drop skill/ into your project as .claude/skills/opentsc/, or call skill/scripts/opentsc.py directly.
(Optional) pip install jieba snownlp zvec and copy skill/templates/_config.yaml to <vault>/soul/_config.yaml to turn on the memory engine.
Read the usage guide. Upgrading from v1.0? See MIGRATION.md — your data needs no changes.

The bigger idea

OpenTSC is the "shell" of a concept called TSC — the Thin-Shell Company: how one person can run a large, self-evolving operation with AI agents. The one idea worth knowing:

Keep your judgment and memory (the "soul") separate from whatever software, team, or AI model happens to run it (the "shell") — so it survives any tool change. Export the soul, and you've exported the whole thing.

The full doctrine lives in the whitepaper / "soul" repo. The mapping from each principle to the code that implements it is in SKILL.md.

License

Code: AGPL-3.0 for the community; a commercial license is available for closed-source use.
Name & brand: OpenTSC™ / TSC™ are trademarks — see TRADEMARK.md.
Contributions: see CONTRIBUTING.md (includes a CLA).

OpenTSC · created by dashen「AI 最严厉的父亲」· dashen.wang · @dashen_wang