Stop taking notes. Start executing.
Spoken communication is the fastest, highest-bandwidth medium humans possess for technical brainstorming. Yet after every standup, sprint planning, or 1-on-1, engineers spend hours manually typing tickets, hunting for lost decisions, and reconstructing action items. Relay was built to eliminate that gap forever.
The Transcription Trap
Traditional meeting transcription tools dump thousands of words of unedited, rambling prose into your inbox or shared drives. Nobody reads them. What teams actually need isn't another 40-page transcript—they need the five things that must be built, who owns them, what blocks them, and when they are due.
Relay operates not as a passive recorder, but as an autonomous task extraction engine. We treat spoken speech as raw AST (Abstract Syntax Tree) input, parsing commitments, owners, and dependencies into deterministic Directed Acyclic Graphs (DAGs) ready for synchronization with issue trackers like Linear and GitHub.
Core Architectural Principles
Sub-Second Intake
Invoke Relay instantly with global desktop hotkeys (⌥ Space / Alt Space) without leaving your IDE, terminal, or browser. Audio begins streaming in under 12ms.
Deterministic Task DAGs
Relay deduces parent-child relationships, detects cycles, and attributes assignees automatically, eliminating post-meeting coordination overhead.
Local-First Encryption
Audio stems are encrypted with AES-256 GCM in transit and at rest. Passkey credentials remain safely inside platform hardware security enclaves.
Continuous Tool Tuning
Ingested user prompts are analyzed exclusively to optimize internal LLM tools, ensuring repeatable, low-latency, and cost-efficient extraction across common workflows.
Build the future of voice-driven software delivery.
We are actively admitting engineering teams, tech leads, and founders into the Relay Beta.
Request Beta Access