Get started
Run it on your own machine
Python 3.11 to 3.13 and the standard library. No packages, no API key and no spending for the demo or the offline checks.
Windows (PowerShell)
From the repository folder, create the virtual environment once:
git clone https://github.com/billyrdavis1985-bot/ForgeVigil.git
cd ForgeVigil
py -3.11 -m venv .venv
.\.venv\Scripts\python.exe -B -m forge_vigil demo --review
Run the full offline qualification:
.\.venv\Scripts\python.exe -B tools\qualification_check.pyLinux and macOS
git clone https://github.com/billyrdavis1985-bot/ForgeVigil.git
cd ForgeVigil
python3 -m forge_vigil demo --review
Run the full offline qualification:
python3 tools/qualification_check.pyWhat the demo does
The offline demo uses disposable, fictional state. It shows ALLOW, DENY and REVIEW, a replay, a local release by the operator and a final report, then deletes its state. It makes no model calls.
Inspect a saved email
python3 -m forge_vigil inspect-email --file examples/emails/review-signals.eml
Shows real versus displayed link hosts and attachment-name signals. No links are opened, no attachment bytes
extracted and no model called. Outlook .msg files are not supported; save as .eml.
Before real traffic
The live Muse assessor is off unless you choose it, and requires your own Meta Model API key and a spend cap. Read these first:
- REGISTRY.md: the facts you own, such as documents, recipients and approvals
- Live Muse assessor: settings and spend limits
- Backups: back up the key together with the database
- Limits: what has not been established
Keep your API key out of chats and commits
Set MODEL_API_KEY in your own shell. On Windows the README shows a masked prompt so the key never appears
in your command history.
Documentation
- README: overview and command reference
- QUALIFICATION.md: reproducible offline checks
- DEMO.md: the scripted BLOCK and investigation rehearsal
- EMAIL_INSPECTION.md and SCANNING.md
- HANDOFF.md: current checkpoint and engineering plan