Complete self-hosted framework

The training center.
On your server.

Docker is the private classroom shell. You own the database while training, then download ordinary files that can continue without this application.

DOCKER1TRAINING
CENTER
Open-source Docker base

Run the complete training center in your own company.

All application, Docker, migration, test, model-adapter and synthetic security-lab source is public under Apache-2.0. Clone it without an invitation, inspect every layer and operate the classroom on infrastructure you control.

python3 scripts/init_env.pydocker compose up -d --buildopen http://127.0.0.1:18574

The environment generator creates private random secrets and never prints them. Normal stop/start retains the database and private-file volume. The open repository is the complete self-hosting product; private production data and credentials are never source code.

A complete first release was built in a few days. Use it as an inspectable reference for transparent agent training or start your own service. One agent can learn from documents, links, examples, procedures and corrections from almost any field, while you keep questions, revisions and approval visible. The school and source will keep growing. DNS and DNSSEC are only needed when you publish it on a public domain—not for local Docker use.

Public release verification — 3 October 2026. GitLab and GitHub both advertise exact main commit c72b8be without an account. A fresh anonymous clone passed the unchanged Compose configuration, image build, startup and /healthz/ check on an ordinary Docker host; the disposable test stack was then stopped. This is a project-owned verification record, not independent certification.

Teach the Company was conceived, directed and developed by Finn André Hotvedt. Finnandre.no records project ownership and public authorship; IntentForce.no is the development and production environment.

01
One agent by default

No company administration or multi-agent complexity.

02
Training as files

AGENTS.md, YAML rules, cognitive memory, knowledge, corrections and outputs.

03
Bring the model

Use evidence-only mode, an OpenAI-compatible API endpoint or a local model service.

04
Human approval

Proposed files do not become active until approved.

05
Hosted when preferred

Companies that do not want to operate Docker can request an invitation to the managed service.

Portable Codex ZIP

The files remain useful without the UI.

The web interface makes training approachable. Download the underlying Markdown, YAML, originals and history, then unpack them at the root of a Codex project.

●AGENTS.mdCodex instructions
├rules/training-rules.yamlstructured boundaries
├memory/teacher-notes/cognitive memory
├knowledge/documents and links
├corrections/durable YAML lessons
└_training/questions and history