Why trust Nemo
Nemo is a personal intern: one for each person, reachable in chat, that does real work — drafts, sends, runs code, tracks things down — instead of just answering questions.
One intern, wherever you already talk.
Nemo lives in the Nemo web app, a full two-way Telegram bot, Slack (connect your own workspace), plus email. It remembers what you tell it — “remember I work at…”, preferences, recurring details — and carries that across every channel and every future conversation. It also runs background scans (like your Gmail inbox) to proactively surface things worth your attention — drafts, follow-ups, reminders — without you having to ask first.
Every credential is encrypted; nothing sensitive reaches the model.
Bot tokens and OAuth credentials (Slack, Telegram, X) are encrypted at rest with AES-256-GCM and decrypted only inside Nemo's own server — never inside a prompt sent to the LLM. For the broader catalog of connected apps (Gmail, Calendar, and others), Nemo doesn't even hold the token itself: a dedicated OAuth proxy holds it and injects it at call time, so the model only ever specifies what to call, never the key that authorizes it. Code the agent needs to run executes in an isolated, single-use cloud sandbox that's destroyed the moment the run finishes — no state persists between calls.
Nothing sends itself.
Emailing, messaging a group, or replying is a draft until you say go — the send action is code-blocked from firing in the same step that composes the message, so a second explicit confirmation is always required. For email replies and LinkedIn posts specifically, the only way anything goes out is a human clicking Send in an editable composer — there's no agent-only path to publish. Automations you schedule in advance lock in their recipients at creation time, so they can't later be redirected by anything the agent subsequently reads.
Hosted centrally, so you don't have to run anything.
Nemo runs as one service — a web app backed by a real-time, cloud-managed Convex database, deployed via Docker Compose sitting behind an nginx reverse proxy/load balancer — rather than something you deploy or self-host. Each user's data is scoped and isolated at the database level (no cross-tenant access), with infrastructure based in Singapore and data based in US East (N. Virginia).