Abstract
The agent-to-map (A2Map) Protocol streams high-frequency geospatial vector data from remote large language model (LLM) agents to interactive client viewports through a distributed architecture. A remote LLM agent generates spatial updates, which a coordinate compression engine converts into token-compact serialized representations using techniques such as variable-length ASCII polyline encoding before streaming. A client-side spatial runtime receives these incremental delta packets and applies them to an offline, double-buffered virtual map state tree. The runtime atomically commits snapshots of the state tree to a single, hardware-accelerated canvas overlay synchronized with the display’s refresh rate, achieving smooth rendering without engaging the document object model (DOM). An asymmetric camera state machine decouples user-initiated navigation from agent-commanded camera updates to prevent viewport snap-back. Furthermore, a bidirectional geospatial event bridge performs client-side hit-testing, serializes user gestures into typed actions, and transmits them back to the remote agent to enable a closed-loop interaction. This design supports high-throughput data streaming and rendering without freezing the user interface, while also reducing LLM token consumption.
Keywords: A2Map protocol, spatial vector data, generative artificial intelligence, coordinate compression engine, double-buffered rendering pipeline, asynchronous network stream
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Vijayvargiya, Arjun, "Agent-to-Map Protocol: High-Throughput Spatial Vector Streaming for AI Agents with Double-Buffered Synchronization", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12247