Everything a modern AI workspace needs — designed, built, shipped.
Each capability was scoped, architected and delivered as part of this portfolio build: schema, server functions, streaming, UI and polish.
Switch between 8 frontier models mid-conversation. Token streaming with stop, regenerate and message branching.
Drag-and-drop documents. Per-line confidence, dual-pass multilingual merge and an inline correction editor.
On-demand browsing tool with clickable citations, hover previews and 'used for this answer' highlighting.
Whisper STT in the composer and natural TTS playback on any assistant message.
Group threads into folders, scope context with project knowledge files, and pin work that matters.
Generated code and documents open in a versioned side panel — iterate without losing history.
Built-in image-generation tool the model can call when a visual is the right answer.
Custom instructions, per-thread system prompts, saved prompt library and cross-conversation memory.
Cron-style recurring tasks that run prompts on a schedule and post results back into a thread.
Send the same prompt to two models side-by-side and stream both responses simultaneously.
Per-day messages and tokens with one-click CSV and PDF export for billing or reporting.
Generate scoped keys to use DeshiGPT programmatically — keys shown once, hashed at rest.
Every table is RLS-locked. Users only ever read and write their own rows — enforced in the database.
Server functions stream from the edge for low-latency token delivery anywhere in the world.
GitHub-flavoured markdown, KaTeX math and Mermaid diagrams render inline in every response.