Building a Self-Hosted AI Agent Platform Users Fully Own
DBB Software designed and built a personal AI platform that runs entirely on the user's own machine, shipping three language runtimes as a single installable binary, with the user's memory, credentials, and files never leaving their hardware.
Industry
Retail & E-Commerce
Service
Agentic AI Development
Team
1 Senior Full-stack Developer
Project State
March 2026 - Ongoing
Country
NDA
About the Client
The client is a venture-backed, remote-first technology studio. Their platform is a personal AI operating system that runs on the user's own hardware or a private hosted pod, giving each person long-term memory across every conversation and file, multiple collaborating AI agents with distinct roles and personalities, and direct connections to the accounts they already use.
The Client's Initial Request
The client engaged DBB Software to build a personal AI platform from scratch because their product promise ruled out the architecture that every competing assistant relies on: a vendor cloud that holds the user's conversations, memory, and API keys.
Data Sovereignty by Architecture
Make user ownership a system property rather than a policy statement, with conversations, files, memory, and credentials residing only on hardware the user controls.
01
Single-Artifact Distribution
Package the entire backend so that a non-technical person can install one thing and have a working platform, with no services to orchestrate and no database to provision.
02
An Extensible Agent Platform
Allow each user to run multiple AI agents with their own memory, personality, and skills, and let third-party skills, tools, and connections extend them without forking the product.
03
Delivery Across Three Surfaces
Ship the same product as a self-hosted install, a managed hosted instance, and a desktop application, without maintaining three divergent codebases.
04
Solutions We Delivered
DBB Software built the platform end-to-end, from system architecture through deployment, with every technical decision traced back to a single constraint: the software runs on machines the team cannot reach and on data the team must never see.
The Pod Runtime
A single Go binary carries the whole backend: authentication with JWT and OIDC, multi-user workspaces, an encrypted vault for the user's own model keys and OAuth credentials, a file drive, browser automation sessions, audit logging for every sensitive action, and self-updating. State lives in an embedded database and the local filesystem, so there is no external database to run and nothing to phone home to. Internal services are bound to loopback behind one authenticated trust boundary.
The Supervised Agent Layer
A TypeScript service spawns and supervises one Python agent process per AI agent, managing its profile and configuration and restarting it when it fails. The agent runtime extends NousResearch's open-source Hermes agent framework, productized for multi-user, multi-agent operation. Agent events are translated into a standard streaming chat protocol so the browser receives token-by-token responses and live tool-call progress across three process hops and three languages.
Persistent Memory and Agent Identity
Each agent accumulates a durable memory graph from the user's conversations and connected sources, with a retrieval layer that lets it recall facts and decisions months later. Personality definitions, a portable skills format, and the Model Context Protocol tool servers let the same agent gain new abilities without a code change. Scheduled and background runs let agents work while the user is away, and multi-provider speech support adds voice to the same loop.
Cross-Surface Packaging and Delivery
A Tauri desktop application bundles Go, Node, and Python into a single installer, provisions a managed HTTPS tunnel so the local pod is securely reachable, and updates itself in place. A React single-page application delivers streaming chat, memory, and personality management, and the skill marketplace across every surface. Roughly 2,900 automated tests run across the Go packages, the TypeScript service, the interface, and end-to-end browser journeys, with continuous deployment on every merge.
Results Achieved
Data Stays Local
The user's conversations, files, credentials, and their agent's memory never reach a server the client or DBB Software operates. Users achieve full ownership of the solution.
Install the Deployment
A person with no infrastructure knowledge runs one installer and has a working multi-agent AI system, with three language runtimes, a database, and a supervised process fleet hidden behind it.
Model Choice Stays Open
Users bring their own provider keys and switch between them without losing their agent's accumulated memory. Nobody is locked into one model vendor's pricing or roadmap.
Re-Architecture Without a Freeze
Mid-project, roughly 70 API routes moved from an in-house agent engine onto the Hermes-first architecture, and about 21,750 lines of superseded code were retired, while the product stayed runnable throughout.
Releases Happen on Merge
Around 2,900 automated tests across four layers gate every change, and merges deploy continuously. A single developer maintains a three-runtime platform without a manual regression pass standing between a fix and its users.
Zero to Working Platform
Roughly 86,000 lines across Go, TypeScript, Python, and React went from an empty repository to a demonstrable end-to-end product in about four months, built by one senior engineer.
Ship AI That Runs on Your Customers' Infrastructure
DBB Software builds AI products that keep customer data on customer hardware, without giving up the experience of a hosted product.
Contact Us
"Most of our work starts with a 30-minute call where someone describes a product they're trying to ship and one part of the engineering picture they can't get around.
If that's where you are, let's set one up; I'll tell you straight whether we're the right fit.”
Mina Morkos
Business Development Manager
Want a similar outcome for your team?
Ask our AI assistant — it can pull related case studies, talk through the approach, and put you in touch with the team if you want a deeper conversation.
