opentsc

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?

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

Requires Python 3.10–3.14 for the optional zvec index (the core itself runs on bare Python). On macOS (system Python is 3.9) or if you hit externally-managed-environment, use uv — full guide in INSTALL.md.

# 1. install as a skill — the repo's skill/ subdir IS the skill, not the repo root
git clone https://github.com/opentsc/opentsc
cp -r opentsc/skill ~/.claude/skills/opentsc        # or ~/.hermes/skills/opentsc

# 2. (optional) turn on the memory engine — recommended Tier-1 backends
uv venv ~/.venvs/opentsc --python 3.11 && source ~/.venvs/opentsc/bin/activate
uv pip install jieba snownlp zvec
cp opentsc/skill/templates/_config.yaml <your-vault>/soul/_config.yaml

Then read the usage guide · install details in INSTALL.md · upgrading from v1.0? 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).