Rewire

Inspiration

We walked around the hackathon floor and talked to a lot of teams. We noticed something — everyone is technically building the same thing. The only difference is which sponsor APIs they call, in what order, and what data they pass between them. That's it. Every project is just a different wiring of the same building blocks.

That made us think: what if we stop hardcoding one pipeline and instead build an agent whose brain is a visual canvas? You drag API blocks, connect them however you want, and hit run. Rewire the brain before every run - and suddenly one agent can do what dozens of hardcoded agents do.

What it does

Rewire is an autonomous agent with an open brain. The canvas is the circuit board. You drag sponsor API blocks from the toolbar, wire them together, configure each one, and execute the whole pipeline autonomously. Outputs flow from node to node using {{input}} templating.

Here are some flows you can build just by dragging and connecting:

Research & Extract — Tavily Search → Fastino GLiNER → Neo4j Graph → Output Search the web for a topic, extract people/orgs/locations from the results, and store the entities as a knowledge graph. Instant structured intelligence from raw web data.

Visual Content Moderation — Yutori Browser → Reka Vision → Modulate ToxMod → Output Have the browser agent scrape images from a website, run them through Reka to describe what's in each image, then flag anything toxic. Autonomous content moderation pipeline.

Enterprise Knowledge QA — Airbyte → Senso.ai → Fastino GLiNER → Output Pull customer data from Salesforce via Airbyte, query your knowledge base with Senso for relevant policy context, then extract key entities and relationships from the answer.

Competitive Intel Loop — Tavily Search → Fastino GLiNER → Tavily Search (loop) Wire two blocks into a loop and the agent becomes long-running and autonomous — search for a topic, extract entities, use those entities as new search queries, repeat indefinitely. A self-feeding research agent.

Web-to-Graph Pipeline — Yutori Browser → Fastino GLiNER → Neo4j Graph → Output Point the browser agent at a company's team page, extract all names and roles, and automatically build a graph of the org structure.

Change the wiring, change the behavior. No code changes needed.

How we built it

• React + TypeScript + Vite frontend
• React Flow (@xyflow/react) for the drag-and-drop canvas, nodes, and edges
• Custom topological sort execution engine that resolves the DAG and runs nodes in order
• Express.js backend as an API proxy to all sponsor services
• Hand-drawn UI aesthetic (Kalam + Patrick Hand fonts, wobbly borders, paper textures) to make it feel like you're literally sketching an agent's brain on a napkin

Challenges we faced

• Balancing between "build one deep integration" vs "build a meta-tool that connects them all" - we bet on the meta-tool and it paid off.
• Giving up on the idea we were working on at 2pm and pivoting to this with limited time left on the clock

What we learned

With AI, every system is just API calls stitched together and data being passed around. Orchestration is the real product. So why hardcode one pipeline when you can build an orchestrable agent - one whose wiring is the interface itself?

Built With

• airbyte
• fastino
• modulate
• neo4j
• react
• reactflow
• reka
• render
• senso.ai
• tavily
• typescript
• vite
• yutori