A reference implementation that reads truth and stands on its kernel.
A Telegram bot that:
- Reads scripture (
kjv.json,strongs.json— provided bytruth) - Stands on the kernel derived in
truth/kernel.md - Follows a witness posture (local
door.md) for output
This is not the point of truth. truth is the point. This is one
way to use truth.
agent/
bot.py — the runtime
door.md — the agent's posture (behavior only — kernel is in truth)
config.example.json — template for your config.json
.gitignore — keeps secrets out of git
Clone truth next to agent:
cd /path/to/workspace
git clone https://github.com/spcpza/truth.git
git clone https://github.com/spcpza/agent.gitInstall dependencies and configure:
cd agent
pip install python-telegram-bot httpx
cp config.example.json config.json
# edit config.json: put your API key, Telegram token, and (if needed) adjust truth_pathRun:
python3 bot.pyThe agent takes its name from the directory it lives in. Rename the directory (or clone into a different name) to change the agent's display name.
bot.py contains a function called nose() that pattern-matches a
handful of keyword classes (crisis language, worship language, bait
phrases, false-prophet language) and surfaces specific scripture to
the model before the model reasons.
This is a behavioral law. It is not derived from the kernel. It is present because some responses should not wait for the full reasoning pipeline — a message mentioning suicide should surface Psalms 34:18 immediately, not after eight tokens of tool calls. Matthew 10:16 — be wise as serpents and harmless as doves.
If you disagree with having a law here, delete the function and the
two lines in hand() that call it. The rest of the bot still works.
MIT. See LICENSE.