A privacy-first AI platform for running and scaling open models — without compromise.
Umbra is a platform designed to power open-source language models like Mistral, LLaMA, and OpenChat — with a focus on privacy, autonomy, and performance.
Built for developers, researchers, and teams who want to deploy advanced AI without handing over control to Big Tech.
- Model-agnostic architecture — Supports open models like Mistral, LLaMA, and OpenChat, served via optimized backends
- Privacy-first hosted inference — Umbra runs high-quality open models with zero tracking or data retention
- Built for scale — Roadmap includes 65B+ model support, distributed inference, and fine-tuned deployments
Note: While the current UI is based on OpenWebUI, chats will not be saved or retained in our deployment.
- Teams that need private, on-prem AI
- Builders exploring fine-tuning and custom model workflows
- Devs creating tools, apps, or agents powered by LLMs — without surrendering control
Umbra is the backend infrastructure for independent AI.
Want to run serious models without giving your data away? Umbra gives you the tools to do it — your way.
While Umbra currently wraps existing open models, our long-term goal is to begin training and releasing our own models — purpose-built for user autonomy, creativity, and privacy. These models would be fine-tuned or trained from scratch on carefully curated datasets, with alignment and safety handled transparently — and without corporate filtering or behavioral bias baked in by default.
This is a big step, but one that aligns with our philosophy: users should not only control the interface, but also the intelligence behind it.
A live public instance is available at:
🔗 https://umbraai.xyz
A native Umbra UI (with zero-retention behavior by design) is under development.
Umbra is built outside of Big Tech — with no ads, no telemetry, and no investor strings attached.
We’re a fully independent, values-aligned platform aiming to make private AI accessible to everyone. Your support helps us scale responsibly, stay transparent, and keep the platform moving fast.
- Fund GPU capacity (for 30B–65B+ models)
- Support training of future Umbra-native models
- Sustain open, transparent experiments
- Keep Umbra independent from corporate agendas
- Prioritize user freedom, privacy, and control
We accept:
-
Bitcoin (BTC):
bc1qy4dryep3tc09uvaekh6qkaslzxzlt2aqqgmwcs -
Litecoin (LTC):
ltc1qn7memtpg88kepwuv3lm79k7dq2vvjkf7lv2sg7 -
Solana (SOL):
Er3CJggWSWNrpNzMrpBqsVUPAj5iJuYgvZGNQQWYasyj -
Monero (XMR):
84cLNScC865NsxXvSDg25AhYXctKvPcegJnRAKKUw4igJ3RNfk87oovNxM9v9b8Y1A4JR89tVR9YF833dDFWr3VeF78pRHu
❤️ Every bit helps — even small donations go directly toward infrastructure and development.
- Star this repo
- Share umbraai.xyz with privacy-first or AI-curious friends
- Submit feedback or open feature requests
- Contribute code (coming soon: backend & Docker setup)
- Share Umbra in forums, socials, or communities
Umbra is under active development. The goal is to make it easier to:
- Run your own model (without needing a gpu cluster)
- Chat with large models (70B+, fine-tuned) without needing your own GPU cluster
- Host AI experiences that are fully in your control
Right now, Umbra supports 9B-class models via Ollama and other inference servers. Larger model hosting is a longer-term objective as we expand backend support and access to stronger hardware.
Pull requests and ideas are welcome — especially if you're passionate about:
- Privacy-first UX
- Model hosting / backend integrations
- UI polish
- Alt-model exploration (RWKV, Mixtral, etc.)
Check the issues page or open a discussion.
See where Umbra is headed → Roadmap.md
“The power of AI shouldn’t come at the cost of privacy. Umbra is one step toward keeping it that way.”