Hours
Weeks
Commits
Role
Ola
AI-native multi-account Gmail client with offline drafts and real-time smart inbox
Ola is a modern Gmail client for people who live in their inbox, combining a Slack-like conversation view with an AI assistant that actually understands their mail.
ola-mail/ola
Ola is a modern Gmail client for people who live in their inbox, combining a Slack-like conversation view with an AI assistant that actually understands their mail. It connects multiple Gmail accounts under a single Supabase user, lets users switch between primary and secondary inboxes instantly, and keeps drafts and actions safe during flaky connectivity through an IndexedDB backed offline queue. AI categorization, summaries, and style aware smart replies help users clear their inbox faster while keeping all mail and labels stored directly in Gmail, not on a third party server. Under the hood, Ola uses Next.js and React for the application shell, then replaces Google SDKs with a custom TypeScript GmailClient that handles OAuth token refresh, multipart MIME construction with attachments, and error recovery. Gmail’s complex MIME and threading model is normalized into app specific thread and message objects, which power the UI, AI pipelines, and unsubscribe aware flows. Dexie provides a typed IndexedDB layer for drafts and queued actions, and Serwist enables offline capabilities and push notifications. AI features run through Azure OpenAI, exposed as streaming server sent events, and are gated behind a Pro subscription that is enforced on the server through Supabase and Polar billing data. Ola also implements end to end Gmail push delivery: Google Pub/Sub webhooks drive history API reads, which are filtered by per account importance rules and turned into browser web push alerts. This design keeps the client responsive, reduces unnecessary polling, and supports multi account setups without sacrificing performance.
Designed a Gmail parsing pipeline that normalizes raw threads into UI-friendly models, decoding complex MIME structures, threading headers, attachments, and unsubscribe metadata for consistent mailbox views and logic.
Crafted style-aware smart replies that analyze the user’s sent-mail corpus per account, mirroring language and tone while enforcing Pro subscription checks and per-user rate limiting.
Developed AI-powered email categorization that calls Azure OpenAI, maps model outputs into Ola-specific labels, and materializes those labels directly inside Gmail using the Gmail API.
Created streaming AI compose and summarize endpoints that translate Azure Responses API events into browser-friendly server-sent events, delivering low-latency text generation to the client.
Engineered a multi-account Gmail integration that layers Supabase user auth over separate gmail_accounts, enabling per-account token management, legacy compatibility, and seamless switching between primary and secondary inboxes.
Built a custom Gmail client that automatically refreshes access tokens and constructs multipart MIME messages with attachments, ensuring reliable sending without relying on heavyweight Google SDKs.
Implemented an IndexedDB-backed drafts and offline action queue using Dexie, persisting unsent changes and supporting recovery banners so users safely work through connectivity drops and refreshes.
Orchestrated end-to-end Gmail push notifications that consume Pub/Sub webhooks, query Gmail history, honor per-account importance filters, and dispatch browser web push alerts while pruning invalid subscriptions.
Ola provides a modern, AI-augmented Gmail client that supports multi-account workflows, offline-safe drafting, intelligent categorization, and real-time notifications for productivity-focused email users.