Featured

JacHacks came to San Francisco for the first time on July 26, taking over Founders, Inc. for a single day of building. More than 200 builders shipped 77 finished AI projects in Jac in under 9 hours, competing across three tracks: Agentic AI, Fintech, and Social Impact.

The 77 projects covered a wide range of use cases, across industries from healthcare to fintech to disaster response, and every one of them was built in Jac in that same 9 hours, a genuinely tall order for a single sitting.

Jac is what made the timeline realistic: one language covers the backend, the database, and the frontend, and by llm() adds AI to any function with no prompt engineering required. Builders also had the opportunity to host and deploy their apps through Jac Hammer.

Main track winners

Special awards

Spotlight

Sigil: don't prompt the skill, compile it

Sigil is a skill compiler, built in Jac. It compiles a plain Markdown SKILL.md into a typed agent harness: every mandatory step becomes a node the run has to visit, every prohibition a constraint the structure enforces.

The chart is the whole argument. Across 30 skills, prose runs land anywhere between 0 and 100; compiled runs pile up at 100. That lifts required-step compliance from 56% to 86% on GPT-4o, on 0.58× the tokens, and a better model doesn't close the gap: compiled wins on GPT-5 too.

Jayanaka's write-up covers the full evaluation, and sigilagent.com is the place to start if you'd rather just try it.

Ecosystem

Tutorials

Jac Tutorials: learn Jac from scratch

The official tutorial series for Jac and Jaseci, ordered from lesson one so you can follow along step by step. It covers the core concepts and how Jac differs from Python, Object-Spatial Programming and Meaning-Typed Programming, building backend APIs, frontend UIs, and AI features, and deploying an app locally or to the cloud. No prior Jac experience needed, just some comfort with Python and basic web concepts. New lessons land regularly.

Articles

Hijac: how the JacHacks Winter Social Impact winners put autonomous Jac agents on a phone

Nitin Shankar Madhu, Chuka Ezeoke, and Meron took 1st in the Social Good track at JacHacks Winter with Hijac, a pipeline that compiles Jac apps straight into installable native mobile apps. Their demo: an agent that watches a phone's sensors, judges whether a spike looks like a fall, and calls an emergency contact on its own if the owner doesn't respond within 30 seconds. It runs entirely on Object-Spatial Programming, with sensor readings modeled as nodes and by llm() handling the judgment call with no JSON-parsing boilerplate.

Community

Join the Jaseci Discord

Most JacHacks teams are still shipping in there. Debugging help, release questions, RFC chatter, and a steady stream of show-and-tells.

Subscribe to Jaseci on YouTube

Talks, demos, and the Jac tutorial series land here first.

Releases

jaclang 0.36: Postgres-native persistence, dependency-free serving

What matters if you're upgrading:

• Breaking: persistence is Postgres-native. The Mongo/Redis/Firestore stack is gone. Each request runs in one SERIALIZABLE transaction, zero-config runs use an embedded Postgres, and JAC_DB_URL points at any managed one. 0.35.x session data is not migrated.
• Breaking: serving is Jac-native. FastAPI, uvicorn, pydantic, and friends are gone from the runtime. Password hashing moved from bcrypt to scrypt, so pre-0.36 credentials need a reset.
• Breaking: ReadPerm and friends are now AccessLevel. One ambient enum (.READ, .CONNECT, .WRITE, .NO_ACCESS) everywhere a level is taken.
• Graph queries compile to data. Edge references lower to a query IR resolved inside the storage engine's own indexes, and every filter spelling is now equivalent.
• flow for goes parallel. A lent collection fans out over threads, with unsound captures caught at compile time: about 6× on 8 threads.
• One Kubernetes pipeline. Monoliths and microservices share the same deploy target: zero-downtime rolling updates, --dry-run previews, and a programmatic deploy SDK.