For anyone deeply embedded in crypto and Web3, Telegram has become the de facto communication hub. Every DAO discussion, every investment group, every project coordination happens there. Telegram is home base, but it’s also chaos. Blink twice and there are 150 unread messages across 20 chats. Important pings get buried. When a team member's account gets compromised, manually removing them from every group becomes a tedious nightmare. Onboarding new contributors means countless copy-pasting invites across multiple channels.
Dory was created to fix exactly that.
Built as a Telegram-native concierge, Dory helps you tame the chaos with AI-powered summaries, user and chat management, onboarding/offboarding flows, nudges, memory search, and more.
The obvious objection? "I don't trust a bot with all my messages."
And that objection is entirely fair. Traditional chatbots require sending your conversations to centralized servers where they're processed in plaintext. Your sensitive discussions, alpha calls, and private conversations become vulnerable to leaks, hacks, or simple corporate overreach. A privacy-guaranteed architecture for Dory has become more important than ever.
And at the heart of Dory’s privacy architecture lies Phala.
Dory takes a fundamentally different approach from every other Telegram bot, by leveraging Trusted Execution Environments (TEEs), specifically through Phala’s confidential computing infrastructure. All Dory core services run inside Phala Confidential VMs, which are TEE-backed secure enclaves. Your code and data stay encrypted even while in use. Neither the Dory team nor the Phala team can access your plaintext conversations.
More importantly, Dory does NOT store your Telegram messages. Messages are fetched on demand inside the enclave, processed, and discarded; only lightweight metadata is stored, and even that lives inside the encrypted TEE perimeter.
AI inference happens inside Phala Confidential AI as well, in separate enclaves that process prompts and outputs entirely within TEE boundaries. Data moves enclave-to-enclave encrypted, never exposed to the host machine or cloud provider.
By running computation inside secure enclaves, Phala enables a new category of applications that can be both powerful and genuinely private. Without Phala's Confidential VMs and Confidential AI capabilities, Dory would be forced to choose between functionality and privacy. TEEs represent what the Founder of Dory, Captain Nemo, call "the closest we got to OpSec nirvana.”
How Dory Works
Most Telegram bots operate through Telegram’s Bot API, which is intentionally limited and cannot perform core management tasks. Bots can’t manage members, can’t handle access control, and lack the privileges needed for real team workflows. Dory solves this by running a real Telegram account inside the enclave. For teams, the recommended approach is creating a dedicated Telegram account and adding it to all work chats you want managed through Dory.
The authentication flow is straightforward: enter your phone number, receive a code, and authorize. Once authenticated, Dory’s enclave-hosted MTProto client:
- Syncs chat metadata
- Syncs member lists
- Subscribes to message and event streams
- Acts on your behalf for chat/user management
Because everything happens inside the TEE, plaintext never leaves the secure perimeter.
Core Dory Features
Dory focuses on three foundational categories that directly address the broken Telegram workflow.
Chat & User Management: This is where Dory replaces dozens of manual tasks:
- Add or remove users across multiple chats with one click
- Bulk-add members to all relevant groups
- Apply labels to chats for structured access control
- Manage users by label (e.g., “BD,” “Core Dev,” “Investors”)
Insights (AI-Powered): Ask natural-language questions about any Telegram group you’ve connected:
- “What were the main discussion themes in the BD chats this week?”
- “Summarize partner discussions from the last 3 days.”
- “Did we miss any unanswered customer questions?”
All inference is done inside Phala Confidential AI, so your prompts and outputs maintain the same privacy guarantees as the core system.
Nudges: Nudges are one of Dory’s most powerful capabilities — they act like an intelligent assistant that watches selected chats and notifies you when certain conditions are met.
Dory comes with ready-made templates:
- Unanswered Questions — ping when someone’s question goes ignored
- Important Mentions — track when you or your team get tagged
- Long Silence — detect when important chats go cold
- Commitment Tracker — follow up when someone says “I’ll handle this,” then forgets
You can also build fully custom, LLM-powered nudges:
- “Alert me whenever someone shares a transaction.”
- “Notify me if anyone raises a concern about staking or new DeFi primitives.”
- “Tell me when there’s an escalation in any partner chat.”
From Experiment to Breakthrough
Dory didn’t start as a polished product. It started the way most good tools do: with a few builders falling deep into the TEE × AI rabbit hole, trying to answer questions that whitepapers never really solve. What actually works in practice? What quietly breaks the moment you leave the lab? What turns out to be insanely expensive? And, maybe most importantly, what’s truly useful to real people?
And they built. They spun up enclaves, wired up MTProto clients, ran models inside Phala’s Confidential VMs, and let the architecture teach them where the sharp edges were. And while they were experimenting, they kept running into the same, very personal frustration — Telegram chaos. Missed pings, buried alpha, unmanageable groups. It became obvious that the problem they lived with every day was the perfect proving ground for what they were trying to learn.
Some experiments end up as quirky prototypes that never leave the lab. Others reveal themselves as something bigger — a glimpse of the future before it’s obvious. Dory feels like the latter.
If you have questions or want to get in touch with Dory, feel free to message the Founder of Dory on Telegram: @ncerovac 😄
The Bigger Picture
In Dory’s case, every part of the workflow lives inside the protected perimeter. The Telegram client fetching your messages, the systems handling your chat management, the models running your summaries and insights — all of it happens inside Phala’s TEEs designed to keep plaintext out of reach.
Phala’s confidential computing stack changed the equation. Phala makes it possible for Dory to operate without ever needing access to plaintext. It transforms privacy from a promise into a property of the architecture. And when you interact with Dory — whether you’re summarizing a chat, tracking commitments, or managing access across multiple groups — every step of that workflow stays inside the secure perimeter that Phala provides.
What you get as a user is something rare: an AI assistant that doesn’t need your trust because it is structurally incapable of betraying it. That’s the real breakthrough — not just for Dory, but for how the next generation of AI tools will be built.
