
System design lives in people's heads and on whiteboards, and turning a conversation into a clean architecture diagram is slow, manual work. Someone always ends up dragging boxes while the real discussion moves on. TalkGraph fixes that. You describe your system out loud, and an animated architecture diagram builds and edits itself live on the canvas. Say "add a Postgres behind the API and put a queue between them," and it happens in seconds. TalkGraph is built on AssemblyAI's Voice Agent API over a single browser WebSocket. The agent edits the diagram only through typed tool calls against an allowlisted component library, so it never invents components or illegal connections. Keyterms from the catalog keep technical words like Kafka, Redis, and Kubernetes accurate, and barge in lets you interrupt and correct it naturally. Voice is the shortcut, not the only way. Everything also works with drag, click, and keyboard, and sensible defaults are inferred for you with one tap undo. TalkGraph also lints your design for weak spots like single points of failure, simulates traffic and failures, and exports Terraform, Mermaid, and an ADR from the same typed graph. From conversation to production ready architecture, just by talking.
30 Sep 2026