Jaseci Digest · June 18, 2026

Welcome to Issue N.02 of Jaseci Digest, a biweekly roundup of what's happening across the Jaseci and Jac open-source ecosystem.

This issue's lead: Jac Hammer turns a prompt into a real, full-stack app: one Jac program where the UI, data, authentication, and AI are all the same language, deployed in one click.

Featured

Jac Hammer: an AI app studio that builds in Jac ►  Watch video Try Jac Hammer What if one prompt gave you a real, full-stack app with authentication and a database, that you deploy frontend and backend in one click and scale to millions? ✗  With Lovable You get a React frontend and a separate Postgres backend wired together: one app spread across two layers and three languages (React, TypeScript, and SQL) to understand, change, and own. ✓  With Jac Hammer You get one Jac program where the UI, data, authentication, and AI are all the same language and runtime. One thing to read, change, and own. Describe the app, and ship it. A real app, not a demo. Jac Hammer vs Lovable. The differences that actually matter. Jac Hammer Lovable Database ✓ One node. Frontend and backend read the same definition, nothing to keep in sync. A SQL table plus a matching TypeScript type: two places to drift when the model changes. Authentication ✓ Built in. Every user gets their own graph. Supabase Auth, plus Row-Level Security policies you own and debug. Per-user data privacy ✓ On by default. Nothing to author, nothing to forget. You write the RLS rules. Miss one and data leaks. AI in your app ✓ by llm(). Call a model like a typed function. Wire up an SDK, write the prompts, parse and retry the output. Languages to debug ✓ Jac. TypeScript (UI), TypeScript (edge functions), and SQL, across two systems. Deployment ✓ Frontend and backend, one click. Frontend deploys; the Supabase backend runs as a separate service. Builds for ✓ Web and desktop Web. Want the full write-up? Read it on the site.

Articles

How Tobu used Jaseci to build an AI-powered memory graph

Tobu is an AI memory app that captures the people, places, and stories behind your photos. It models each one as a connected memory on a Jac graph, not a file in a folder. The highlights:

• 2 weeks to a working demo, 6 weeks to production, zero prompt-engineering pipelines
• 92.84% retrieval accuracy, 2.2× fewer missed memories than traditional RAG
• $1.5M raised on the early traction

Community

JacHacks Spring winners are live

JacHacks Spring wrapped with more than 300 participants and 81 projects shipped in four days. Winners landed across Agentic AI, Consumer Healthcare, Fintech, and Social Impact, alongside specialty awards like Best Use of Jac and Best Demo. See them all on the winners page, or explore all 81 submitted projects on Devpost.

Join the Jaseci Discord

The fastest way to talk to the team and other builders. RFC chatter, debugging help, release questions, and a steady stream of show-and-tells. Worth joining whether you're shipping in production or just curious.

Releases

jaclang v0.16.5: eject a Jac project into a deployable app

jac eject now compiles a project into a runnable FastAPI backend, with walkers exposed as POST endpoints and authentication wired up automatically, plus a compiled JavaScript frontend. jac install is faster too, using uv as its backend when available.

jac-desktop v0.2.2: a Jac-native desktop target

The desktop target is now fully Jac-native: jac build --client desktop compiles a single self-contained binary that embeds the OS webview. No more Rust toolchain, PyTauri, or PyInstaller in the chain.

jac-scale v0.2.28: event-driven autoscaling, down to zero

A new KEDA autoscaler engine (autoscaler_engine = 'keda') scales on real demand and can idle workloads all the way to zero replicas (idle_replicas = 0). The admin console also adds deploy health, endpoints, and storage pages, so you can watch rollouts and pods without touching kubeconfig.

byllm v0.6.17: sturdier structured outputs

by llm() now retries malformed or empty structured outputs automatically, regenerating up to three times (configurable via max_output_retries) with corrective prompts, and streaming is now safe to consume across threads.

Issue N.02 · June 18, 2026 Jaseci Digest

© 2026 Jaseci Digest. Part of the Jaseci open-source ecosystem.