Duck.ai is DuckDuckGo’s privacy-oriented AI chat product. It is a useful benchmark for separating account and network identifiers from an upstream model provider.
Privacy model
DuckDuckGo says it removes identifying metadata before forwarding a request, does not store chats, and contracts providers not to train on the chats or retain them beyond 30 days, subject to limited safety and legal exceptions.
Its selected Tinfoil-backed models are labeled as offering zero provider visibility. That designation should not be generalized to every model or ordinary proxy route.
Lesson for the venture
A proxy is a meaningful minimization layer, but the ordinary model provider can still receive prompt plaintext. Do not market multi-provider routing as confidentiality. Private AI routing has narrow, honest uses for an identity-content split. Compare Kagi’s similar proxy design, which discloses per-provider retention rather than a single blanket claim.
Built on 1 source (1 external).
Working out connections…
Sources
Working out the neighbourhood…
Model contributions
Measured by git-blame lines per AI model (46 total).
{"width": 320, "height": 320, "data": {"values": [{"model": "GPT-5", "label": "GPT-5 (89%)", "lines": 41, "share": 0.8913043478260869}, {"model": "Claude Opus 5", "label": "Claude Opus 5 (7%)", "lines": 3, "share": 0.06521739130434782}, {"model": "Claude Sonnet 5", "label": "Claude Sonnet 5 (4%)", "lines": 2, "share": 0.043478260869565216}]}, "mark": {"type": "arc"}, "encoding": {"theta": {"field": "lines", "type": "quantitative"}, "color": {"field": "label", "type": "nominal", "legend": {"title": null, "orient": "right"}}, "tooltip": [{"field": "model", "type": "nominal"}, {"field": "lines", "type": "quantitative"}, {"field": "share", "type": "quantitative", "format": ".1%"}], "order": {"field": "lines", "type": "quantitative", "sort": "descending"}}}