Open source voice AI learning for everyone
UnaMentis is an open source, voice-first AI learning platform. It delivers structured educational content through natural, fully interactive voice conversation, so people can learn while walking, driving, doing chores, or anywhere they can't sit in front of a screen.
The platform is MIT-licensed, provider-agnostic, and designed to run entirely on-device when needed, with zero dependency on any single cloud provider.
The core experience is full hands-free, bidirectional voice interaction with barge-in capability. This is not an audio book. The learner can interrupt at any point to ask questions, explore tangents, challenge ideas, or request examples, and return seamlessly to the material.
The name means "one mind" in Latin.
The intended audience is every human being on earth with a capable smartphone.
Every major voice AI learning experience today exists inside a proprietary silo. ChatGPT's voice mode is extraordinary for learning, but it locks users into OpenAI's models, pricing, and decisions about how the experience works. When OpenAI changed voice mode behavior in a product update, extended learning conversations broke overnight. Users had no recourse.
Education remains stuck in a text-first paradigm. Academic competitions like Knowledge Bowl, Quiz Bowl, and Science Bowl are conducted entirely by voice, yet every preparation tool is text-based.
There is currently no open infrastructure for voice-native AI education. No standard curriculum format for conversational AI learning. No platform where you can swap your LLM, TTS, and STT provider without changing a line of code. No option to run the entire stack on-device.
UnaMentis is working software.
Fully functional voice learning client with real-time voice interaction, curriculum browsing, and session management. Android beta intended to follow shortly.
9 speech-to-text providers, 8 text-to-speech providers, 5 LLM providers. Cloud, on-device, and self-hosted options. Any component swappable without code changes.
117M-parameter TTS model ported to iOS using Rust and Candle. Near-server-quality voice synthesis running entirely on-device. No API keys. No data leaving the phone.
JSON-based curriculum specification for conversational AI learning. Maps to IEEE LOM, LRMI, SCORM, xAPI, QTI, CASE, and Open Badges.
Python-based server with WebSocket support for remote logging, real-time metrics, curriculum management, and hosted inference. React-based operations console.
In development.
UnaMentis is MIT-licensed. That will not change.
Even the largest enterprise customer in the world would run the exact same open source server as a self-hosted individual user. There is no planned "community edition" that gets neglected while the real product goes behind a paywall.
The open source project is the mission. Everything else exists to sustain it.
When the open source project achieves sufficient traction, we intend to form a company called UnaMentis Learning, incorporated as a European entity.
Add-on capabilities that extend the open source server for organizations with specific needs. Everything the server does out of the box stays open source.
Implementation support, custom curriculum development, integration consulting, and training for organizations deploying at scale.
Managed UnaMentis infrastructure for organizations and individuals who want the platform without running their own servers.
We will always retain the ability to give away enterprise features to anyone we believe should have them. We intend to make real money with this company, but the mission comes first and the ideals behind this effort are not negotiable.
When we formulate the governing documents for the corporate entity, a permanent commitment to shepherding the open source project will be part of the foundation, including a default percentage of profits dedicated to the open source project. The company is meant to be genuinely profitable. Supporting the open source project and building a successful business reinforce each other.
30+ year technology career spanning software development (Python, C#, Java), cloud infrastructure and DevOps (Azure, Terraform, Kubernetes, CI/CD), data analysis, and architectural design. Currently Senior Cloud Engineer at CD Baby (Downtown Music Holdings / Virgin Music Group) since 2014.
In 2025, began exploring AI at a fundamental level: published a knowledge distillation framework, ported a 117M-parameter TTS model to iOS, built governed multi-agent coding systems (OpenClaw, Agent Vision Team), created a codebase architecture visualization tool (Solution Explorer), and became active in Portland's AI community, speaking 2-3 times monthly.
Before tech: US Army Presidential Honor Guard (3rd US Infantry, Old Guard), hang gliding instructor at Kitty Hawk, EMT, decades-long traditional Irish musician.
Contributing to architecture, development, and project direction alongside Richard.
The team is intentionally small. We use AI-assisted development practices to achieve output that would traditionally require a larger team, and we want to grow only when it serves the mission.
The server technology already exists. What's needed is a hosting environment to run it publicly, serving inference to connected clients. Growing from a handful to 100-200 active users.
AI-driven marketing and content creation. API costs, AI video tools for demos, and basic tooling.
Presenting at conferences, meetups, and industry events. Speaking often pays for itself, but upfront costs require capital.
Enabling one or both founders to dedicate themselves entirely to UnaMentis and formation of UnaMentis Learning.
Focused integration work, curriculum development, and infrastructure provisioning to begin generating revenue.
Someone who looks at this project, understands the vision, and wants to help make it happen.
We are not looking for decision-making authority to change hands. We are not looking for pressure to close the open source core, restrict free distribution of enterprise features, or prioritize profit extraction over educational impact.
The support can take many forms: a donation through Open Collective Europe, a personal angel investment, a convertible note or SAFE-equivalent with terms that protect the mission, or a combination. We are open to structuring this in whatever way makes sense, as long as the core principles are preserved.
Relevant reference points for the commercial open source model:
Moodle started as one person's project. Now serves 150 million learners across 230 countries. Never taken venture capital. Commercial model is a partner network on the open source core.
Ghost is MIT-licensed, structured as a non-profit foundation. Generates $7.5M annually from managed hosting, serving Apple, DuckDuckGo, and Mozilla. Started with two people and a Kickstarter.
Cal.com raised $32M for open source scheduling infrastructure. PostHog raised $3M on an MIT-licensed product analytics platform through angel investors.
The commercial open source model has worked at scale across many verticals. Voice-native AI education is not yet one of them.
The best way to understand what we're building is a conversation. Richard is reachable through any of the channels listed on the portfolio site.