Kagi Assistant is a multi-model AI service that proxies selected model providers for its users. It is an important comparison point for a privacy-focused model gateway.
Privacy model
Kagi says it does not train on user data, does not share account information with providers, and defaults conversation threads to deletion after one day. Its public provider table makes an important fact visible: retention varies by chosen model and backend.
That transparency is better than an undifferentiated claim that every route is private. The upstream provider can still read plaintext during normal inference.
Lesson for the venture
Adopt the honest, per-provider policy display, but make the router local-first and add a distinct client-attested confidential tier. Private AI buyer alternatives compares the tiers. Compare Duck.ai’s similar proxy design, which labels select routes as zero provider visibility rather than disclosing retention per provider.
Built on 2 sources (2 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 (87%)", "lines": 40, "share": 0.8695652173913043}, {"model": "Claude Sonnet 5", "label": "Claude Sonnet 5 (7%)", "lines": 3, "share": 0.06521739130434782}, {"model": "Kimi K3", "label": "Kimi K3 (4%)", "lines": 2, "share": 0.043478260869565216}, {"model": "Claude Opus 5", "label": "Claude Opus 5 (2%)", "lines": 1, "share": 0.021739130434782608}]}, "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"}}}