A real personal assistant. In your Telegram.
It reads your mail, guards your calendar, remembers what you asked for and quietly runs the routine — all from one private chat. Inside it runs Hermes Agent, the open-source AI agent — launched, updated and watched over for you. No VPS, no terminal, no DevOps.
- Ready in about three minutes
- Your own AI provider
- Keys encrypted, never shown again
This is the entire product, hour by hour.
No dashboard to learn, no commands to memorize. You type like you would to a person; the platform does the unglamorous parts around it.
You didn't ask this morning — you asked once. Recurring work runs on the platform, between your messages.
Mail passes through only when you ask. Message bodies are never stored or logged on our side.
Sending always waits for your confirmation in the chat. A plain “yes” isn't enough — you press the choice.
Read-only or read & write, per account — and you pick exactly which calendars it may touch.
Thinking runs on the provider you connected — your ChatGPT subscription or your own key. Your model, your bill.
Automations live on the platform, not in a script you babysit. You can pause or delete any of them from the dashboard.
Updates, monitoring and recovery are our job. If something wobbles, we notice it — not you.
Connect Telegram, pick your AI provider, launch. Free during the limited beta — no card.
Three steps, about three minutes
A guided wizard walks you through each connection, remembers where you stopped, and never asks for a card.
Open our official bot with one secure link. That private chat becomes the place you talk to your agent.
Sign in with your ChatGPT subscription, or paste a key from OpenAI, Anthropic, OpenRouter or Nous Portal.
We prepare your agent and keep it ready. Calendar and mail are optional — connect them now or any time later.
No VPS, no terminal, no config files — ever.
Saved encrypted once, shown masked forever after.
Updates, monitoring and recovery are on us.
Free during beta · nothing to install · you can change everything later
It gets access to real accounts. Here's the deal.
Four promises we can point at in the product, not four adjectives.
Your keys are write-only
Provider keys and app passwords are validated once, stored encrypted, then shown masked forever — the dashboard and the API never return the value again, only its last characters.
The agent never holds your credentials
It reaches your accounts through our broker with short-lived, scoped access. Refresh tokens and passwords never enter the agent, its memory, its files or your chat.
You choose how far it can go
Read-only or read & write, per account — and for calendars, exactly which ones. Sending an email always needs your confirmation in the chat first.
Your content goes where you send it
Your agent runs as a private instance. Mail bodies pass through the broker only when you ask — not stored, not logged on our side. What you send does reach the AI provider you connected, and Telegram keeps its own copy; the agent keeps only what you asked it to remember.
Automated content checks are narrow and best-effort — they can miss things and can flag safe things, and they don't certify what you send as safe or lawful. Read the details in Privacy, Prohibited data and Security.
Powered by Hermes Agent — open source, in the open
Hermes Agent is an open-source personal AI agent by Nous Research, released under the MIT License. It is the engine inside every Talaria agent — we don't fork it into a black box, we run it for you.
A fine weekend project if you enjoy ops — the repository has everything you need, and self-hosting is a legitimate choice.
- Rent and secure a VPS
- Install the runtime and dependencies
- Keep API keys in env files
- Apply every update yourself
- Notice when it breaks — and fix it
The same open-source engine — with the wings already attached.
- Press Launch — ready in minutes
- Nothing to install, ever
- Secrets encrypted, masked forever
- Updates arrive automatically
- Monitoring and recovery included
Talaria is an independent managed service and is not affiliated with, sponsored by or endorsed by Nous Research unless separately announced in writing.
Two free ways to start during beta
Payment isn't enabled while legal and Paddle launch requirements are unresolved. Choose your own AI credentials now, or request managed access.
- Active immediately during beta
- Your provider bills model usage directly
- Telegram workspace, apps and automations
- No card and no Talaria payment
No administrator approval required.
- No model-provider key required
- Small one-time starter allowance while pending
- Full monthly allowance after admin approval
- Telegram workspace, apps and automations
A beta request is sent to an administrator for a capacity decision.
Common questions
Meet your new assistant today.
Mail, calendar, reminders and research — handled from one Telegram chat. Free during beta, live in about three minutes.