Build local‑first,
private AI together.
Nokast is an open-source AI studio and community focused on small language models, SecureRAG, and personal assistants that run close to your data—not in someone else's cloud.
How Nokast works
4 principlesStart with a concrete problem
No vague "AI for everything." We zoom into one real workflow—staying informed with LocalizeAI, or running a cognition agent with Nodemind—and solve it fully before moving on.
Design local‑first by default
Data never leaves your machine unless you explicitly choose. Models are small, fast, and tuned for your hardware. Interfaces are built for people who actually use AI—not click‑through demos.
Open‑source the entire stack
Every project ships with clear architecture notes, configurable components for local SLMs and SecureRAG pipelines, and examples you can run on a laptop—not a cloud cluster.
Cycle feedback into the community
Build logs, mistakes, and trade‑offs go on Substack. Contributions flow back in via GitHub and Nokast Sessions—a tight loop between builders and users.
Everything runs on your machine. No cloud required.
Runs on your hardware
Small language models that fit on consumer GPUs and MacBooks. No cloud subscription, no API key.
SecureRAG
Private retrieval-augmented generation keeps your documents local and your context private by default.
Built in public
Every tool, pattern, and decision is shared openly. Fork it, improve it, ship it your way.
What we're shipping
Current projects
One mission across every tool: AI that runs where you are.
LocalizeAI
Newsletter intelligence, locally
Stop drowning in newsletters. LocalizeAI reads your inbox locally and generates a focused daily brief — top stories, key insights, original links — without sending your data to the cloud.
- ✓ Runs locally using SLMs
- ✓ No subscriptions, open-source
- ✓ Built for people who use AI daily
Nodemind
Voice-to-checklist, in your browser
Talk out your messy thoughts, get a clean checklist. Runs 100% in your browser — speech-to-text and cleanup both happen on-device, no install, no account, no server.
- ✓ Free and open source
- ✓ Runs fully on-device via WebGPU
- ✓ Installable as a PWA, works offline
JevControl
Offload agent decisions to a smaller model
Find out whether a smaller model can handle your agent's routine choices. Import a trace, pick the decisions to test, and compare speed, cost, and agreement against your main model.
- ✓ Reads real traces, no production changes
- ✓ You approve every decision tested
- ✓ Compares latency, tokens, and cost
Stay in the loop
Follow the Nokast build log
Want updates on Nodemind, LocalizeAI, and Nokast Sessions? Subscribe to the Nokast build log — deep dives on local-first AI, SLMs, and SecureRAG.
Subscribe on Substack