RFC on DotBot, an LLM based intent-driven AI assistant for Polkadot

Hello everyone, this is Vikk from the Hungarian Polkadot DAO, and I’m writing in connection with a new proposal on the idea of DotBot, which is currently in the RFC (Request for Comments) phase. We would like to gather as much feedback as possible from the global Polkadot and Kusama communities so we can better understand the community’s perspective on DotBot and adjust our plans accordingly. At a later stage - probably 2-3 weeks from now - we will submit the proposal to OpenGov to request funds to further develop the current POC. Thank you very much the community’s contribution! Please find DotBot’s proposal below.:backhand_index_pointing_down:

In 2026, AI is everywhere, it’s our personal assistant, our co-pilot, and our researcher. But in crypto, we’re still stuck in the “manual transmission” reality, clicking through five screens just to move tokens across chains.

We think it’s time for Polkadot to have a conversation. We’re building DotBot, a chatbot that lets you execute transactions using simple English. No more overcomplicated flows, just tell the bot what you want to do, and it handles the technical heavy lifting while keeping you secure and asking for your permission when needed.

What is the goal?

Our goal is to bridge the gap between user intent and on-chain execution. We’re building the “brain” that knows which request to call and how to handle transactions easily, so you don’t have to worry about complex situations.

We know there’s still much work ahead. We’re focusing on refining the system right now and would appreciate any feedback.

What it does

The project focuses on translating user intent into safe, runtime-aware transactions across Polkadot Asset Hub and selected parachains. While basic asset transfer functionality is already implemented, further work is required to ensure correctness guarantees, robust transaction validation, and predictable behavior across wallets and runtimes.

We’re excited to share our progress on stabilizing and preparing DotBot’s execution layer for release. This includes enhancements like transaction simulation, end-to-end scenario validation, and modular infrastructure that can be reused by wallets and applications.

How can you test it?

Development of DotBot is already underway, and a functional prototype exists.

The current implementation focuses on intent-driven transaction execution rather than traditional form-based workflows. In particular, DotBot currently supports:

  • Conversational input for asset transfers (only on AssetHub at the moment) using natural language
  • Context-aware interpretation of user intent across multiple messages
  • Clarification prompts when user intent is ambiguous (e.g. resolving “Who is Alice?”)
  • Explicit user confirmation before any transaction is constructed or submitted
  • Visualized execution flows that expose each step of the transaction lifecycle
  • Generation of runtime-compatible Polkadot extrinsics based on interpreted intent
  • Wallet-based signing using standard Polkadot browser extensions

Ready to see it in action? Let’s give it a try!

How it helps the ecosystem

Polkadot is powerful but sometimes hard to use. DotBot makes it accessible to everyone, not just the power users. It’s a new way to interact with the network, new users don’t need to learn about blockchain and what’s the exact process of sending crypto to a friend. They just ask and DotBot help to do this.

8 Likes

Hey Vikk!

Are you using MCP for the bot?

Hey @alice_und_bob
I think we could say that, yes. We are developing Agents, which are basically classes in the lib. For example AssetTransferAgent. It has certain capabilities. The LLM knows about the capabilities of that agent. It will fulfill your request, by using the capabilities of that agent. Thinking in longer terms, there will be multiple agents, if a user prompt is complicated, it will compose the flow from multiple agent calls, it will use the capability of the first agent, it will then use the capability of the second agent, probably, it will then go back to the first agent, for example:

  1. transfer
  2. swap
  3. transfer again

For this, you need 2 agents.

4 Likes

GM @alice_und_bob, thanks for your questions. I hope you find @optr‘s answer helpful. Do you have any other questions / recommendations regarding DotBot’s POC? We’d really like to hear your words on the topic! :slight_smile:

Let's be honest, is this something absolutely necessary to spend money on? In my opinion, it isn't. Sorry.
1 Like

Hi @ultracoconut!

Please don’t be sorry! You absolutely have the right to express your opinion. If you feel that an AI chatbot for Polkadot, capable of performing different kinds of on-chain actions on behalf of still-lost users, is unnecessary, that’s completely okay.

That said, we believe an AI chatbot for Polkadot is very much needed, and it’s unfortunate that there is still no viable product in this direction within the Polkadot ecosystem. Meanwhile, other blockchain ecosystems are far ahead in this area, investing significant amounts of R&D efforts and huge amounts of money into LLM-based, user-centric chatbots and other AI tools.

Please find the examples below:

  1. Solana just announced their Colosseum Hackathon, focusing on AI agents:
  1. Galichat & SendAI, both are AI chatbots to help users performing on-chain actions, like token sends and swaps, etc… on Solana.
  1. Polygon just announced that they integrates x402, to enable autonomous, programmable payments for AI agents and APIs on Polygon.
  1. Base chain just announced that they are open for a more agentic ecosystem with OpenClaw: https://open-claw-ecosystem.vercel.app

  2. Lastly, if you didn’t know that, even the Web3Foundation announced in one of their earlier Medium Post the following guideline:

    “Verticals that Excite Us:
    We’re particularly interested in projects in these areas:

    • Cross-border payments & stablecoins — especially solutions serving LATAM & Southeast Asia
    • AI applications — For example: content authentication, verification systems, and AI-powered tools
    • DePIN/IoT Solutions — projects with concrete, marketable products that bridge physical and digital worlds.; …..”

    → The second bullet point leads to the conclusion that the W3F would like to see “AI-powered tools” in the Polkadot ecosystem. In our view DotBot is such a tool.

    Thank you for reading all this and we wish you a wonderful day!

2 Likes

You don’t need a Dotbot; any agent today would be capable of performing a transaction on Polkadot. An agent was able to autonomously purchase a SIM card, add a voice protocol to it, and call their human on the phone in the morning.

Agents today can perform transactions on Polkadot while having coffee using a Unitree robot.:robot:

They wouldn’t be able to safely execute transactions, this is a much more critical task than you think, with several nuances.

Beside, you don’t want to give full control to it, you want it to help you prepare transactions, which you can then overview. This is not something, ChatGPT can do for you as of today.

What you’re describing is a problem with the agents. So the Polkadot treasury would spend money to improve them? They’ll be improved in a few months without us spending any money.

Let me answer to your question from a technological standpoint:

I think we’re mixing up two different things here.

We’re not building or “improving” AI agents. And we’re definitely not assuming agents are dumb or incapable — we know they’re getting better fast.

The problem isn’t whether an agent can create a Polkadot transaction. Many already can. The problem is whether you want to trust a generic agent to safely execute on-chain actions in a complex, multi-runtime ecosystem.

That’s where things get tricky.

DotBot isn’t about giving an agent the keys and letting it do whatever it wants. It’s the opposite. The agent only helps interpret intent. DotBot then constrains it, checks it against the actual chain state, simulates what will happen, and shows the user exactly what they’re about to sign.

This is not something that “just fixes itself in a few months” as agents improve, because agents don’t magically gain Polkadot-specific safety guarantees, runtime awareness, or edge-case coverage on their own.

That’s also why we’re building things like the ScenarioEngine. You can’t just assume all paths work. On-chain mistakes are expensive and irreversible, so you need to actively test weird and bad scenarios — not hope the agent gets it right.

So no, the Treasury wouldn’t be paying to make agents smarter.
It would be paying to make agent-driven UX safe enough to actually use on Polkadot.

Westend AssetHub RPCs don’t work, so we will start implementing Paseo, as a substitute.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔍 Westend Relay Chain
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[06:32:18] ✅ OK   3175 ms  wss://westend.api.onfinality.io/public-ws
[06:32:21] ✅ OK   2367 ms  wss://westend-rpc.polkadot.io
[06:32:23] ✅ OK   2331 ms  wss://westend.public.curie.radiumblock.co/ws

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔍 Westend Asset Hub
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[06:32:23] ❌ FAIL 267 ms  wss://westend-asset-hub-rpc.polkadot.io
[06:32:24] ❌ FAIL 81 ms  wss://westmint.api.onfinality.io/public-ws
[06:32:24] ❌ FAIL 243 ms  wss://sys.ibp.network/westmint

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📊 CUMULATIVE STATISTICS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Endpoint                                                 Total     OK     TO   Fail      OK%     Avg ms
wss://westend.api.onfinality.io/public-ws                 4059   4019     40      0      99%       3324
wss://westend-rpc.polkadot.io                             4059   4054      1      4      99%       2412
wss://westend.public.curie.radiumblock.co/ws              4059   4056      2      1      99%       2472
wss://westend-asset-hub-rpc.polkadot.io                   4059      0      4   4055       0%        —
wss://westmint.api.onfinality.io/public-ws                4059      0      7   4052       0%        —
wss://sys.ibp.network/westmint                            4059      0     22   4037       0%        —