We are excited to announce that you can now chat with your Lightning node directly in Lightning Terminal, a web UI to help node operators to improve the quality and liquidity of their node. This new AI chat feature, in alpha today, lets you ask questions about your node and get answers grounded in your node's own data.

Running a Lightning node can be demanding work with in-depth niche context required at times. Liquidity management, fee strategy, and channel health are specialized topics, and the right answer for one node is rarely the right answer for another. A node that earns most of its routing revenue from a handful of peers has different priorities than one focused on reliable outbound payments. Until now, making sense of all that meant digging through tables, cross-referencing forwarding history, and carrying a lot of context in your head. The node assistant brings that context to a conversation. You ask, it reads the relevant data from your node, and it responds with an answer shaped around how your node actually behaves.

Some of the things you can do with it include:

• Checking your node at a glance. Ask about wallet balances, channel balances, and fee performance without opening a single table.
• Making sense of your routing activity. Understand forwarding trends, find underperforming channels, and spot opportunities to earn more.
• Getting guidance specific to your node. Ask what to do next and receive recommendations based on your node's real data, not generic advice.

We understand that your node's data is private information, so we built in privacy safeguards from the start. Tool calls, structured requests to read a specific piece of information on your node, run locally in your browser against your own node, and any data they return is obfuscated on your machine before it ever leaves your device: identifiers are replaced with random pseudonyms, amounts and timestamps are approximated, and human-readable labels such as aliases are obfuscated. We also maintain a zero-data-retention agreement with Anthropic, so the obfuscated data used to answer your questions is processed in real time and not stored under normal operation. We cover exactly how this works in the privacy section below.

This is an alpha release that we will work to improve over time. We would love feedback from the node operator community as you put it to work and you can send it to us right from Lightning Terminal using the Give Feedback button in the node dropdown menu. For now, it operates as a knowledgeable assistant that helps you think through your node, your channels and your liquidity. Over time, we will enable it to take direct actions with your explicit permission.

How It Works 🛠

The node assistant is powered by Claude, Anthropic's large language model. When you ask a question, it figures out whether it needs live data from your node to answer. If it does, it issues a tool call such as your channel list or recent forwarding history.

Those tool calls execute locally, in your browser, against your own node through Lightning Node Connect, an end-to-end encrypted protocol that lets a web browser talk directly to a Lightning node. The raw data never leaves your browser in its original form. Before any result is sent off to be processed, it passes through a privacy layer that runs entirely on your machine, which is covered in detail below.

Once the data is prepared, the node assistant uses it to answer your question, streaming the response back to you in real time. You can keep the conversation going, asking follow-up questions that build on what you have already discussed.

What You Can Ask 🔀

The node assistant is at its best when you bring it the real questions that come up while running a node. You can ask in plain English, and you do not need to phrase things precisely. Here are a few of the ways it can help, along with prompts you can try.

Liquidity management. Ask it to explain your options for moving liquidity in and out of your channels, and it can even pull current swap terms to ground the answer in real numbers.

• "What is the difference between Loop Out and Loop In?" (Loop Out and Loop In are two ways of moving liquidity out of and into your channels.)
• "How can I bring in more inbound liquidity?"
• "Should I rebalance my channels or open a new one?"

Routing analysis. Ask it to make sense of your forwarding activity, and it can surface recent forwards and reason about what is driving them.

• "What were my most active channels this week?"
• "Which of my channels are underperforming?"
• "Why did I see a spike in forwards on Friday?"
• "How should I set my inbound fees?" (Inbound fees are a tool for steering liquidity toward channels that need it.)

Concepts and context. Ask the bigger-picture questions that help you decide where to focus.

• "What is the opportunity cost of leaving this channel idle?" (Opportunity cost is the idea that capital parked in an idle channel could be earning elsewhere.)
• "Tell me about this peer in the network graph."
• "What does a healthy channel balance look like for my node?"

Privacy by Design 🔒

For many node operators, privacy is a critically important part of their operations, which is why it has always remained a core tenet of the Terminal experience. With this node assistant feature, we wanted to ensure that identifying or sensitive information would not be shared whenever possible. However, we also wanted to ensure that the node assistant was not only private, but also useful for node operators. So, we designed the feature with those requirements top of mind which led to multiple layers of protection in this feature.

The first is a privacy obfuscation layer, modeled after the privacy mapper already used in Lightning Terminal's Autopilot integration, which enables automation features like AutoFees and AutoOpen. Every result pulled from your node is transformed in your browser before it leaves your device using three techniques.

• Pseudonymization replaces identifying values such as node public keys, channel IDs, and payment hashes with randomized stand-ins, so a real pubkey becomes something like NODE_a1b2c3d4. These mappings are stored only in your browser and let the node assistant refer back to a peer across a conversation without ever knowing who it really is.
• Fuzzing adjusts numeric amounts and timestamps by a small margin, which preserves the ability to spot trends and make comparisons while keeping your exact balances and fees private.
• Stripping removes node URLs from the GetInfo response entirely, replacing them with a placeholder. The net effect is that the model receives data that is directionally accurate, enough to give you a useful answer, but from which your real identifiers cannot be recovered.

Further, the second layer of protection is a zero-data-retention (ZDR) agreement that Lightning Labs maintains with Anthropic for this feature. Under ZDR, data sent to the API is processed in real time and then discarded rather than stored. Combined with the obfuscation layer, this means that already-obfuscated node data reaches Anthropic, is used to generate your answer, and is immediately thrown away. Under normal operation, no node data is persisted on Anthropic's side. There are narrow exceptions, such as where retention is required by law or where a session is flagged for a usage policy violation, and our privacy policy covers those in full.

One detail worth calling out: the obfuscation layer protects data retrieved from your node, not the words you type. If you paste a real public key or another sensitive value directly into the chat, that text is sent as you wrote it. As a rule, let the node assistant pull data from your node rather than pasting it in yourself.

Lightning Labs does not store your conversations. Our server acts as a stateless relay that passes requests through and holds nothing, and your chat history lives only in your own browser, where you can clear it at any time. When you first open the chat, you will see a short explainer that walks through these protections and links to the privacy policy, and you will need to click through it before you start chatting.

Try It Today

The node assistant is live on Lightning Terminal. Head to terminal.lightning.engineering, connect your node, click the gold bolt icon in the top right corner, and start asking questions! Because this is an alpha release, your feedback is especially valuable. You can send it to us without leaving Lightning Terminal: open the node dropdown menu, click Give Feedback, and share what you think, along with an optional email address and screenshots if they help explain what you saw.

For now, the node assistant is focused on analysis. It reads from your node to help you understand balances, routing activity, and channel health, but it does not yet take actions on your behalf. That is a deliberate starting point: analysis is where it can be most useful with the least risk while it is still early. In future releases, we plan to let you take direct actions through the node assistant as well, such as adjusting fees, rebalancing, or managing channels, with the same human-in-the-loop confirmations and privacy protections built in from the start. Helping you make sense of your node is the first step toward letting you run it through conversation, and we are excited to see how operators use it today. If you haven't already, join our Slack community, follow us on X/Twitter, and/or subscribe to our newsletter.

The era of agentic Lightning nodes starts now. Today it helps you think, soon it will work for you. Try it now to help us build the best agentic Lightning node.