EthosPowerAIApproach
The open-source stack
I build with the platforms below. Most are open source, and they can run on your own servers. That matters for governance: your compliance team can read the code, trace where the data goes and approve it without relying on a vendor's word. You can also leave at any time and keep the code. Technology is the third step of my method, which is on the approach page. This page lists the parts I use for that step.
On this page
AnythingLLM
Private AI Document Intelligence
Chat with your own documents, run retrieval and agents, and keep every file on your own hardware. I run it on my own servers.
Explore AnythingLLM
Click a capability to see what AnythingLLM does with your documents.
All capabilities work together — embed documents, chat with local or cloud LLMs, and let AI agents automate your workflows.
Ollama
Run LLMs Locally
Runs open-weight language models on your own hardware with one command. Nothing leaves the machine. It is the model server under most of my builds.
Click on any feature node to explore Ollama's capabilities
n8n
Fair-Code Workflow Automation
Workflow automation with more than 400 integrations, native AI nodes and a self-hosted option. It connects the CRM, the mailer and the models behind this site.
Click on any feature node to explore n8n's capabilities
Nextcloud
Self-Hosted Collaboration Platform
File sharing and collaboration on your own server, with end-to-end encryption and an AI assistant that runs inside it.
Click on any app node to explore Nextcloud's capabilities
ERPNext
Open Source ERP System
Accounting, inventory, HR, CRM and manufacturing in one open-source system with no per-user licence. I run this practice on it.
Explore ERPNext Modules
Click on any module in the diagram to see detailed information about its features and business value.
All modules share one database, so a change in one automatically update others in real-time.
OpenProject
Open Source Project Management
Open-source project management with Gantt charts, agile boards and time tracking, on your own server.
Click on any feature node to explore OpenProject's capabilities
Qdrant
Vector Database Written in Rust
Vector database written in Rust for semantic search, retrieval and agent memory. I use it for memory and document search.
Click on any feature node to explore Qdrant's capabilities
Neo4j
Native Graph Database
Graph database for knowledge graphs, agent memory and any data where the relationships matter most.
Click on any feature node to explore Neo4j's capabilities
ChromaDB
Embedding Database
Open-source vector database for embeddings, with a small API, built for local development.
Click on any feature node to explore ChromaDB's capabilities
Playwright
Browser Automation and Testing
Microsoft's browser automation framework, with cross-browser testing, test generation and an MCP server. It runs the copy audit and the visual tests on this site.
Click on any feature node to explore Playwright's capabilities
Firecrawl
LLM-Ready Web Scraping
Turns a website into clean markdown for a language model, with JavaScript rendering and chunking. It is the fallback scraper in my supply-chain showcase.
Click on any feature node to explore Firecrawl's capabilities
Postiz
Social Media Management
Self-hosted social media scheduling with AI content generation. An open-source alternative to Buffer and Hootsuite.
Click on any feature node to explore Postiz's capabilities
Plane
Issue and Sprint Tracking
Open-source project management with GitHub-style issues, sprints and cycles. A self-hosted alternative to Jira.
Click on any feature node to explore Plane's capabilities
Task Master AI
PRD-to-Tasks Decomposition
Turns a requirements document into a structured task hierarchy, reachable over MCP.
Click on any feature node to explore Task Master's capabilities
Msty
Local AI Interface
Desktop app for running local models, with GPU acceleration, Shadow Personas and MCP integration.
Click on any feature node to explore Msty's capabilities
SmythOS
No-Code AI Agents
Visual builder for AI agents, with drag-and-drop editing and connectors to company systems.
Click on any feature node to explore SmythOS's capabilities
LibreChat
Open Source ChatGPT Alternative
Self-hosted AI chat for many models, with agents, memory and MCP integration.
Click on any feature node to explore LibreChat's capabilities
Open WebUI
Self-Hosted LLM Interface
Web interface for local and cloud models that keeps conversations on your own server, with Ollama built in.
Click on any feature node to explore Open WebUI's capabilities
Matrix/Element
Federated Secure Communication
Open protocol for decentralised, end-to-end encrypted chat and video conferencing.
Click on any feature node to explore Matrix/Element's capabilities
Weaviate
Hybrid Vector Database
Open-source vector database with hybrid search, built-in vectorisation and AI agents.
Click on any feature node to explore Weaviate's capabilities
Windmill
Code-First Workflow Engine
Open-source serverless runtime for scripts, flows and apps in TypeScript, Python, Go, Bash and SQL. I use it when a workflow outgrows n8n's no-code editor.
Click on any feature node to explore Windmill's capabilities
Second Brain Pattern
Voice → Graph, Vectors, CRM → Human Gates
A pattern I run for my own two brands and adapt for clients. I record a voice note in Telegram. Whisper transcribes it on a local GPU, and a local model extracts the entities. The result goes to a Neo4j graph, a Qdrant vector store and Frappe CRM, all cross-referenced. Two crawlers look for leads but cannot write to the CRM. A person approves every prospect, every outbound email and every published episode. Windmill runs the schedule and the approvals, and Listmonk sends what a person has approved.
Click on any feature node to explore Second Brain's capabilities
DBOS
Durable Execution on Postgres
A lightweight library that turns ordinary Python or TypeScript functions into crash-safe, replayable workflows backed by Postgres. It checkpoints each step's output. After a restart it resumes from the last completed step and does not repeat side effects. The Dark Software Factory is built on it.
Click on any feature node to explore DBOS's capabilities
Milvus
Large-Scale Vector Database
Open-source vector database for similarity search at very large scale, with hybrid queries.
Click on any feature node to explore Milvus's capabilities
Frappe Learning
Open-Source Learning Management System
A self-hostable LMS built on the Frappe Framework. It has structured courses with chapters and lessons, auto-graded quizzes, manual-review assignments, embedded programming exercises, learner batches with Zoom live classes, completion certificates and a built-in job board. I provide setup, configuration and integration. You keep control of your learners' data.
Click on any feature node to explore Frappe Learning's capabilities
DeerFlow
Long-Horizon Research and Coding Agent Harness
DeerFlow is ByteDance's MIT-licensed agent harness for research and coding tasks that run for minutes or hours. It is built on LangGraph with sub-agents, skills, long-term memory and pluggable sandboxes. I self-host it with Docker Compose as a gateway, a frontend and Redis, and connect it to my models and tools. I provide setup, configuration and integration.
Click on any feature node to explore DeerFlow's capabilities
Honcho
Memory Infrastructure for Stateful Agents
Honcho, from Plastic Labs, is AGPL-3.0 memory infrastructure. An API server stores workspaces, peers, sessions and messages. A background deriver reasons over them to build representations that an agent can query. I self-host it on PostgreSQL with pgvector and connect it through the Python or TypeScript SDK. I provide setup, configuration and integration.
Click on any feature node to explore Honcho's capabilities
Hermes Agent
Self-Improving Agent with Skills and Persistent Memory
Hermes Agent, from Nous Research, is an MIT-licensed agent that runs as a terminal UI, a CLI or a messaging gateway. It writes its own procedural skills after complex tasks and keeps memory across sessions. I run it on my own hardware with the model provider I choose, behind a LiteLLM proxy where that helps. I provide setup, configuration and integration.
Click on any feature node to explore Hermes Agent's capabilities
Graphiti
Temporal Knowledge Graphs for Agent Memory
Graphiti, from Zep, is an Apache-2.0 Python framework that turns a stream of episodes into a bi-temporal knowledge graph. It records when each fact became true and when it was superseded. Retrieval mixes embeddings, BM25 and graph traversal. I self-host it against my own Neo4j (FalkorDB and Neptune are also supported) and expose it to agents through its optional MCP server. I provide setup, configuration and integration.
Click on any feature node to explore Graphiti's capabilities
LiteLLM
OpenAI-Compatible Gateway to 100+ Model Providers
LiteLLM is a proxy and Python SDK that gives one OpenAI-compatible endpoint across more than a hundred providers, with routing, fallbacks, virtual keys, spend tracking and logging callbacks. The core is MIT-licensed, and the repository's enterprise directory carries a separate licence. I self-host the proxy with Docker and PostgreSQL as the routing layer in front of local and hosted models. I provide setup, configuration and integration.
Click on any feature node to explore LiteLLM's capabilities
Listmonk
Self-Hosted Newsletter and Mailing List Manager
Listmonk is an AGPL-3.0 newsletter and mailing-list manager, shipped as a single Go binary over PostgreSQL. It has lists, campaigns, templates, bounce processing and a transactional email API. I run it on my own VPS for my lead-capture forms and this site's blog subscriptions. I provide setup, configuration and integration.
Click on any feature node to explore Listmonk's capabilities
OpenCode
Open-Source AI Coding Agent for the Terminal
OpenCode is an MIT-licensed coding agent with a terminal UI. A plan agent only reads and a build agent edits. It supports language servers and can run headless as an HTTP server. I self-host it as a service on my own machines, with the model provider I choose and optional MCP servers for extra tools. I provide setup, configuration and integration.
Click on any feature node to explore OpenCode's capabilities
Macro
Email, Chat, Docs, Tasks, Calls and CRM in One Workspace
Macro is an AGPL-3.0 workspace that puts email, channels, documents, tasks, calls and a CRM in one app. Any item can @-link any other, and a team memory is refreshed each night from the day's work. It runs hosted today, with an EU region. Self-hosting is free under the licence and Macro arranges self-hosted workspaces itself. There is no public guide yet, and its calls and sign-in run on services that need their own licences. I provide setup, configuration and integration, including agents that work through its MCP server.
Click on any feature node to explore Macro's capabilities
GSD Core
Spec-Driven Development Loop for AI Coding Agents
GSD Core is an MIT-licensed set of commands and subagents that takes an AI coding agent through one loop per phase: discuss the decisions, plan in fresh contexts, execute in parallel waves with a commit per task, verify with the person who asked, then ship the pull request. It continues Get Shit Done, which TÂCHES published in December 2025 and archived in May 2026. It runs on Claude Code, Codex, OpenCode, Cursor and several other agents, and writes its specs and state to a .planning folder in the repository. I use its loop in my own Claude Code sessions. I provide setup, configuration and integration.
Click on any feature node to explore GSD Core's capabilities
DeepTutor
Self-Hosted AI Tutor Built on Your Own Documents
DeepTutor is an Apache-2.0 tutoring agent from the Data Intelligence Lab at the University of Hong Kong. It builds a knowledge base from your own manuals and procedures, answers with citations to the page, writes quizzes, imitates an existing assessment paper and takes each learner through a mastery path with spaced review. It runs in Docker on your own server and can use local models through Ollama. Documents then stay on your server, provided its web search and other outbound tools are switched off. It was written for students, not staff compliance: accounts are local, with no single sign-on, no SCORM and no completion record. I provide setup, configuration and integration.
Click on any feature node to explore DeepTutor's capabilities
Want one of these on your own servers?
I help you choose, install and configure the tools that fit how your organisation works.