GPT-6 Astra Creates Custom Running Routes from Address

ChatGPT Work with GPT-6 Astra can generate custom running routes from your home address using OpenStreetMap data, though the process lacks transparency.

First Impressions

The tool accepts a straightforward prompt - a home address and distance requirements - and returns complete route visualizations with downloadable files. In one test, the system worked for 27 minutes to produce both 5K and 10K routes that loop from a specified address. The output included an embedded visualization, GPX files, and GeoJSON files, making the routes immediately usable with running apps and GPS devices.

The generated routes appear as interactive maps. One example showed an “El Granada harbor loop” measuring 5.1 km, rendered in an HTML file stored at /workspace/el-granada-5k-share. The visualization code uses D3.js (https://cdn.jsdelivr.net/npm/d3@7.9.0/dist/d3.min.js) and includes proper attribution to OpenStreetMap (https://www.openstreetmap.org/copyright).

Core Features

According to the system’s own explanation, the route generation process involves three steps. First, it uses Nominatim to geocode the provided address. Second, it queries the Overpass API to download local OpenStreetMap roads and trails data. Third, it calculates loop routes locally based on the requested distance.

The tool produces multiple output formats. Beyond the visual map, runners receive GPX files compatible with most GPS watches and fitness trackers, plus GeoJSON files for developers who want to manipulate the route data programmatically. The HTML visualization includes distance measurements and map data attribution.

Workflow Integration

The major limitation surfaces when trying to understand or modify the route generation logic. The actual Python code that runs behind the scenes remains hidden in the ChatGPT interface. When asked to provide the code after route generation, the system could not reproduce it, apparently due to thread compaction - a process where earlier conversation context gets compressed or removed.

This opacity creates problems for users who want to adjust the routing algorithm, understand why certain paths were chosen, or debug issues. The 27-minute processing time suggests substantial computation, but without code visibility, developers cannot optimize or customize the approach for specific needs like elevation preferences or traffic avoidance.

The lack of pre-compaction text preservation means the detailed work the model performed essentially disappears. For a tool aimed at technical users, this represents a significant gap between capability and transparency.

Verdict

GPT-6 Astra through ChatGPT Work demonstrates impressive autonomous capability for a specific task - generating geographically accurate running routes using open data sources. The multi-format output and proper data attribution show attention to practical usability.

However, the black-box nature of the execution undermines trust and limits iteration. Developers accustomed to inspecting and modifying code will find the hidden implementation frustrating. The inability to retrieve executed code after thread compaction particularly stands out as what the source describes as an “anti-feature.”

For casual runners who simply want route suggestions, the tool delivers. For users who need reproducibility, customization, or understanding of the underlying logic, the transparency gap remains a critical weakness that future iterations should address.