# Stefan's other public projects

> Stefan Preu's agent-friendly service CLIs, PM CLI ecosystem packages, the LlamaLift experiment, and how his original work differs from forks.
> status: draft-owner-review · updated: 2026-10-20 · source: https://bot.unbrained.dev/knowledge/projects

Besides PM CLI, Stefan Preu's public work includes agent-friendly command-line tools for real services, PM CLI extensions and some experiments. Everything listed here is public on GitHub (https://github.com/unbraind) and, where noted, on npm. The live list is at https://preu.at/work.

## Service CLIs: shared design

Stefan maintains four command-line tools that let people and agents operate real, often self-hosted services through predictable commands: Home Assistant CLI, Jellyfin CLI, SearXNG CLI and Monica CLI.

They share a design: compact structured output, discovery an agent can read (capabilities, schemas and contracts), and writes that have to be asked for explicitly, through read-only modes and confirmation flags. Each has its own page under https://preu.at/work.

## Home Assistant CLI

Home Assistant CLI (command `hassio`, npm @unbrained/homeassistant-cli, source https://github.com/unbraind/homeassistant-cli) controls a Home Assistant instance from the command line.

It gives counts before payloads (`hassio summary`), probes which APIs the instance answers and turns that into commands to prefer or avoid, supports dry runs with strict input checks, blocks every state-changing call in read-only mode, and requires an explicit `--yes` for risky writes. Details: https://preu.at/work/home-assistant-cli

## Jellyfin CLI

Jellyfin CLI (command `jf`, npm jellyfin-cli, source https://github.com/unbraind/jellyfin-cli) works with a Jellyfin media server: typed commands for everyday tasks plus checked access to every API operation.

It can inspect an operation's contract without side effects, validate a batch of reads before running it, keep mutations behind a separate verb that needs `--confirm`, export commands as tool schemas for function calling, and block writes in read-only mode. Details: https://preu.at/work/jellyfin-cli

## SearXNG CLI

SearXNG CLI (command `searxng`, npm searxng-cli, source https://github.com/unbraind/searxng-cli) runs web searches through a SearXNG instance the user chooses.

It has output modes for agents, a local cache for repeatable and offline-first runs, a versioned command contract, schema validation of its output, and a strict mode that fails loudly when a search returns nothing. Details: https://preu.at/work/searxng-cli

## Monica CLI

Monica CLI (command `monica`, npm monica-cli, source https://github.com/unbraind/monica-cli) works with the Monica personal CRM.

It is read-only unless set up for writing, probes which commands a given server supports, exports tool definitions for OpenAI, Anthropic and MCP runtimes limited to what that server supports, requires `--confirm` plus a write-enabled configuration for mutations, and validates outputs against registered schemas. Details: https://preu.at/work/monica-cli

## PM ecosystem packages

Around PM CLI, Stefan publishes separate repositories and packages so that users install only what they need. Examples include a browser UI (pm-web), an independent Rust implementation (pm-rust), context and brief generators, a dependency graph, a Gantt renderer, changelog generation, importers and sync for Azure DevOps, Jira, Linear, GitHub, CSV, Beads and Markdown TODOs, Slack notifications, workflow presets and extension starters.

Some are published on npm, some are source only, and some are archived or experimental. Check each repository for its current status.

## LlamaLift (experiment)

LlamaLift (https://github.com/unbraind/LlamaLift) is an experimental Rust/egui desktop app for managing models on local or remote Ollama servers: listing models, batch downloads with streamed progress, deletion safeguards and persistent settings.

Its v0.1.1 was published as a GitHub Release with a Windows executable; it is not published on crates.io. Stefan presents it as exploratory work, not a maintained product.

## Forks are not his authorship

Stefan's GitHub account also contains forks of other people's projects that he used for experiments, for example A2A, mem0, gemini-cli and agentapi. These are not his original work and should not be presented as such.

The project list on https://preu.at/work deliberately shows only original, non-fork repositories.

## Repository counts are dated

A portfolio review dated 4 September 2026 counted 38 public repositories on Stefan's GitHub, 28 of them original. Counts change. For the current picture, see https://github.com/unbraind or https://preu.at/work, which reads from GitHub and npm automatically.

## Earlier client websites

Before his agentic work, Stefan built websites, webshops and member portals for small businesses, associations and public organizations, mostly in Lower Austria. These are brief career context; see the background page. His current agentic systems and open-source work lead his portfolio.