Interactions

Turn your own AI conversations into training data — capture, curate, and fine-tune your local models, all on your machine.

• A privacy-protected chat client for your local agents that captures every interaction in transit — the prompt, the model's output, and your signals (like, edit, regenerate) — into a per-field-encrypted, per-user store
• Curate what's worth keeping in a review queue — approve, reject, annotate and tag captured pairs — then export curated batches as SFT/DPO training data to AI Training
• Closes the consolidation loop — chat → capture → curate → export → fine-tune — so improvements are baked into your model's weights, not replayed into its context
• Local-first and privacy-bounded — an MCP server exposes only derived state (preferences, themes, queue metadata), never your raw conversations, and doubles as the cluster's capture hub for interactions pushed in by sibling apps