SMF·SHORTWAVE
Can a prompt replace Shortwave?
Email — email clients and hosted mail
Exhibit tracking slip
Verdict
AI-generated summaries of long email threads, and natural-language search translated into a real Gmail search query, Shortwave's actual AI-forward angle, is a real weekend build with your own model API key on top of the Gmail API. What doesn't survive: Shortwave's tight integration polish and its AI features tuned specifically for email content over months of iteration.
Exhibit A — The prompt
Received on31.07.2026Build a Gmail client with two specific AI features, thread summarization and natural-language search, rather than a general AI wrapper around email. Use Next.js/TypeScript with the Gmail API (OAuth) for the base email client (list, read, reply, archive), and a model key read from ANTHROPIC_API_KEY or OPENAI_API_KEY for the AI features specifically. For thread summarization: on a long thread, more than a handful of messages, offer a 'Summarize' button that strips quoted-reply boilerplate and signatures before sending the thread's actual content to the model, returning a short summary of the discussion and any open questions. For natural-language search: let the user type a plain-English query like 'the email about the contract from last month', send it to the model with instructions to translate it into Gmail's own search operator syntax (from:, subject:, after:, has:attachment), and run the translated query against the Gmail API — show the translated query to the user so they can see and correct what was actually searched. Do not build AI-drafted reply suggestions, native mobile or desktop apps, or team shared-inbox features — those are out of scope; this is a single-account web client with two specific, well-scoped AI features. Needs a Google OAuth app with Gmail API access, plus an Anthropic or OpenAI API key for the AI features — without a key, the client still works as a plain Gmail client with those two buttons disabled.
Opening prefills the prompt — press enter to run it.
Exhibit B — What you lose
- B.1 AI features tuned specifically for email content over months of iteration
- B.2 native macOS and mobile app polish
- B.3 team shared-inbox AI features
- B.4 AI-drafted reply suggestions
Prior art
Exhibit C — Why people still pay: mail infrastructure, integrations, and client polish
Calling an LLM to summarize a thread is a weekend; tuning AI features specifically for the quirks of real email, quoted replies, signatures, forwarded chains, so summaries stay accurate, is the ongoing refinement work.
Questions
What exactly does the AI do here?
Two specific things: summarizing long threads, and translating a plain-English search into Gmail's actual search syntax, not a general AI assistant bolted onto email.
Can I see what search query the AI actually ran?
Yes — the translated Gmail search syntax is shown alongside the results specifically so you can verify or correct what was actually searched, rather than trusting a black box.
Does it draft replies for me?
No — that's out of scope. This build's AI features are limited to summarization and search translation.
What does it cost to run?
Minimal hosting plus model API usage for the two AI features, billed per token — search translation and summarization are both short requests, so costs stay modest even with regular use.
Related tools
Receipt