timeweaver

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for timeweaver

Inspiration:

▎ The two hardest unsolved problems in AI agents are long-term memory and agentic planning. What if memory and plan are the same thing at different points in time? A plan is a not-yet-happened memory; a memory is a plan that came true. TimeWeaver was born from this insight.

What it does:

▎ Full-stack travel planning app. Past memories + future plans = one graph on a 3D board. LLM walks edges (Route, Within, Knows, HappeningAt) to recommend destinations. Chat-driven UI. "i will go sf" → plane auto-flies.

How we built it:

▎ Jac language — 1 Node type, 9 Edge types, 4 walkers. Zero database code. by llm with compiler-enforced types. .cl.jac React frontend + Three.js 3D board. 4 hours total.

Challenges:

▎ Mermaid version compatibility, React state batching after root spawn, CSS specificity on dual dark-background columns, seed data persistence across clean builds, Jac type checker limitations on list concatenation.

Accomplishments:

▎ 0 lines of DB infra. Plan→Memory = one field flip. LLM walks edges not queries. UI clicks = natural language. Schema only at render boundary.

What we learned:

▎ When the domain is a graph, graph-native languages eliminate an entire infrastructure layer. LLM calls need compiler enforcement. Attributes should be runtime, not compile-time.

What's next:

▎ Real by llm integration. Multi-user graph search. Social graph crawling. Memory→Vlog auto-generation. Real-time collaboration. Mobile WASM build.

Built With

• deepseek
• jac
• kimi