Self-hosted

Your virtual colleagues

A self-hosted platform for AI agents that know your infrastructure, act on your systems and follow the rules you set.

Max
Sysadmin colleague · IT office
You: The intranet is slow again. Can you check web-01?
Max:

The nginx workers on web-01 are stuck, like last month. The office runbook says to restart the service, so I need your approval.

From the wiki:Intranet · web-01Runbook: restart nginx
Waiting for your approvalweb-01
systemctl restart nginx

Why Orlok

For anyone who wants real help with their work.

A body of knowledge

Procedures, configurations and experience gathered in a wiki that is updated and enriched with every task. Virtual colleagues consult it and make it available to the whole office.

A colleague you train

From that knowledge it learns to answer and to work on your systems following your procedures, with the degree of autonomy you prefer.

More time for valuable work

It works alongside you as a colleague would, and your time is better spent.

The wiki

The knowledge base virtual colleagues check before they answer and update as they work. Every office, whether a department, a team or just you, has its own.

Built from the notes you already have

Import from Obsidian, Notion and Joplin with links intact. Passwords, keys and tokens are stripped on import.

Documents and scans

PDF, Office files and scans read on your server, with OCR in Italian and English. The colleague turns them into linked pages and updates existing ones.

It grows as you work

Colleagues enrich it as they work and, from your conversations, suggest what to keep. You have the final say.

Shared across the office

Each office has its own wiki, readable by all its members. Every change, by a person or a colleague, has an author and a history.

Reliable over time

Outdated information is marked with its date, never deleted. Every night, review flags contradictions, orphan pages, conflicting addresses and stale checks.

Search built for IT

IPs, hostnames and codes found exactly, per-office synonyms, semantic search and a knowledge map.

The office colleague

Every office can have a colleague dedicated to answering all its members. For company facts it consults only the wiki, cites the pages it draws on and, when something is not there, points it out. Recurring questions find a ready, consistent answer, because it comes from the same wiki.

Security and infrastructure

Virtual colleagues act on real systems, within the limits you set.

Your infrastructure

Orlok installs on your servers. The wiki is indexed by a model that ships with Orlok, so your pages never leave your network to become searchable. Your model provider only receives what the conversation needs; with Ollama nothing leaves your network.

Offline when you need it

With Ollama and the built-in embedding model, Orlok does not depend on any cloud service.

Encrypted credentials

Credentials and provider keys are encrypted, with the key kept outside the database. Everyone stores their own accounts; a colleague acts as the person it works for, without seeing the credentials and without extra rights.

Autonomy and controls

Commands are parsed, never matched as plain text. Whoever manages Orlok decides how much autonomy each colleague has. A few destructive commands, such as formatting a disk, are always refused, except in Yolo mode, which has to be chosen explicitly. “Always allow” covers a single kind of command and can be revoked.

Isolated execution

Commands run in a separate runner container: no database access, read-only filesystem, minimal privileges.

Full audit trail

Every change and job is written to a log that cannot be altered, with credentials masked in output. The log can be exported as CSV or JSON Lines.

Tools and features

What you need to work on your systems, from any device.

On your systems

  • Linux over SSH, Windows over PowerShell Remoting (WinRM) or OpenSSH
  • One list of servers, which the colleague reaches with the account of the person it works for
  • Complex jobs laid out as multi-step plans

External tools

  • MCP servers from a ready-made catalog: databases, AWS, Proxmox VE
  • Added per office in a few clicks, entering only the credentials
  • For each server you choose which tools to offer colleagues

The model you prefer

  • Anthropic, OpenAI and any compatible provider, or locally with Ollama, even offline
  • Wiki, tools and guardrails stay in Orlok: changing the model changes neither the rules nor the knowledge
  • Every model call is logged with its cost, with monthly spending limits per office or person

Desktop and mobile

  • Installable as an app on phones
  • Conversations and approvals on Telegram too, with email or push notifications
  • Close the page and the answer keeps going: pick it up on another device

Screenshots

Ask in plain words: the colleague checks the server, recognises the problem from the wiki and proposes the fix.
Ask in plain words: the colleague checks the server, recognises the problem from the wiki and proposes the fix.

Quick start

Database, app and isolated runner, started with Docker Compose. Requirement: a server with Git and Docker (with Compose).

  1. Get the code

    Clone the repository on the server that will host Orlok.

    git clone https://github.com/rylos/orlok.gitcd orlok
  2. Configure the environment

    Copy the example file, then set the public URL and three generated keys.

    cp .env.production.example .env.production# ORLOK_PUBLIC_URL: the address people will useopenssl rand -base64 32   # ORLOK_SECRET_KEYopenssl rand -hex 24      # POSTGRES_PASSWORDopenssl rand -hex 32      # ORLOK_RUNNER_TOKEN
  3. Start Orlok

    Images are built on your server and database migrations run automatically.

    docker compose --env-file .env.production up -d --build
  4. Complete the first-run setup

    Open the configured URL (port 8080 by default), create the company and the administrator, then add your first model provider.

Orlok is open source under the AGPL v3 or later: free to use, at work too. Read the license

Talk to us

Want to try Orlok in your company, or have a question? Write to us.

We use your details only to reply to you.