Via AI
An agentic workspace built on the Claude Agent SDK — the same core capability as Anthropic's Cowork, engineered with the security and controls that regulated, document-heavy work demands. I co-founded Via and lead its product, design, and engineering — from the system architecture to the interface.
Overview
Via is an AI workspace where an agent works next to you — reading your files, using your tools, and running real multi-step work with a person in the loop. It's built on the Claude Agent SDK, the same foundation Anthropic's Cowork runs on.
As Cowork took off, we deliberately matched its UX. Teams get an interface they already recognize — chat and canvas, live connectors, an approval flow — wrapped in the security a sensitive back office actually requires. It's live in production, not a demo.
Built on the Claude Agent SDK
At the core is an agent runtime on the Claude Agent SDK — tools, memory, and connectors (Drive, Slack, Notion, a private RAG index, web search, browser control) so the agent gathers context and takes action, not just answers. A workflow engine wires it into operations: any record event fires an agent against a typed output schema.
It runs on React/TypeScript, GraphQL, and Postgres, with workspace isolation, RBAC, SSO, and audit built in — the baseline for sensitive data, and why it clears procurement where most AI tools stall.
The agent does the real work
Documents and data flow in; the agent drafts the comparisons, proposals, and forms, works the back-and-forth, and produces finished, review-ready artifacts.
Inside the product
One workspace ties it together — the AI copilot, enriched records, document generation, and automation, all over the same live data model and connectors.
The client portal
On the other side, policyholders get a portal to see their policies, claims, and certificates — with an AI agent that answers coverage and billing questions on the spot.
