1. Telegram Walt AI agents can now trade perpetual futures
Walt, the Telegram-based app formerly known as Wallet, announced on Tuesday, October 20, that users can now let AI agents open and close perpetual futures trades on their behalf. Three agents are supported: Mira, built by Walt's developer The Open Platform, plus OpenAI's ChatGPT and Anthropic's Claude. Agents trade only from an account the user funds, cannot withdraw money, and begin with order execution switched off.
The launch moves AI assistance from conversation into action. Instead of explaining a market or suggesting an idea, a connected agent can now place real leveraged orders, provided the user has deliberately enabled that ability. The company framed the safeguards as the core of the design, and they are what separates the feature from a simple chatbot plug-in.
Perpetual futures, often shortened to perps, are derivatives contracts that track an asset's price without an expiry date. Traders use them to take long or short positions with borrowed exposure. They are among the most heavily traded products in crypto, and also among the riskiest for inexperienced users.
2. What can the agents do inside Walt?
According to Walt, a connected agent can pull live perpetual futures prices and market news, and review a user's open positions, leverage and liquidation risk. It can open or close long and short trades when instructed, and it can place take-profit and stop-loss orders that close a position at preset price levels.
Beyond single commands, agents can run rules that users define themselves, such as dollar-cost averaging, which spreads purchases over time instead of entering at one price. They can also monitor positions continuously, a task that suits software better than a person who has to sleep, work and check a phone.
A take-profit order closes a position once it reaches a target gain, while a stop-loss order closes it once losses reach a chosen limit. Both are standard risk tools in derivatives trading, and handing their placement to an agent means the protective step can be set at the same moment a trade is opened, instead of being left for the user to remember afterward.
The range of functions covers the full cycle of a trade: researching, entering, protecting and exiting. That breadth is what makes the feature notable, because earlier AI tools in consumer finance mostly stopped at analysis and left the final click to the user.
3. Why the founder says chat beats trading terminals
Andrew Rogozov, founder and chief executive of both Walt and The Open Platform, argues that trading applications have traditionally demanded a lot of the people who use them. In his description, they present terminals, charts and order tickets that people have to learn how to use, a barrier that keeps many potential traders away.
His proposition is that natural-language instructions can replace much of that learning curve. A user who can describe an intention in a sentence no longer needs to master an interface built for professionals. Rogozov also pointed out that Telegram users already hold funds with Walt, which means the agents plug into an existing balance rather than asking people to move money somewhere new.
That convenience is the commercial logic behind the feature. A trading product that lives inside a messaging app and responds to ordinary language lowers the effort required to participate, which can raise activity for the platform. It also raises the question of whether lower friction is an unambiguous benefit when the product involves leverage.
4. How are the agents limited?
Walt says its agents can trade only with money that the user has moved into a dedicated account. They cannot withdraw those funds or send them elsewhere, and they place no orders at all until the user turns execution on. Users set the agent's instructions and trading limits, can watch each action as it happens, and can disconnect the agent whenever they choose.
These constraints address the most obvious fear about autonomous software handling money, which is that a mistaken or manipulated agent could drain a wallet. By confining the agent to a separate trading account and removing any ability to move funds out, the design caps the worst case at the amount the user chose to allocate, rather than the user's entire holdings.
The limits do not remove market risk, however. An agent that follows instructions faithfully can still lose the whole allocated balance if a leveraged position moves against it. The safeguards protect against theft and unauthorized transfers, not against bad trades, and the safeguards described are the company's own account of how the system works.
5. What Walt has not yet said
The announcement leaves several practical questions open. Walt did not spell out the maximum leverage agents may use, what the service will cost, or in which countries it is available. It also did not explain how ChatGPT and Claude connect to a Walt account, or which trading venue executes the orders that the agents place.
Some background helps fill the gap on the venue. Walt's perpetual futures launched in April through the decentralized exchange Lighter, according to The Open Platform. That offering covers more than 50 markets and allows leverage of up to 50 times. Whether agent-driven trades follow the same limits was not stated.
Until those details are published, prospective users cannot fully compare the agent feature with trading manually in the same app, nor judge whether the fee structure and execution venue suit their needs.
Those unanswered points matter because leverage is the central risk variable. At 50 times, a price move of only a couple of percentage points against a position can erase the margin behind it. Users weighing the feature will want to know whether agents can use the full range or are held to something more conservative.
6. Why leverage and automation raise the stakes
Leveraged positions can be liquidated quickly when markets move sharply. On October 10, 2025, more than $19 billion of crypto positions were liquidated in a single day, a record for the crypto market. The episode is a reminder of how fast forced closures can cascade when many traders hold borrowed exposure at once.
Adding autonomous agents to that environment introduces a new variable. A human trader who sees a violent move may hesitate, reduce size or step away, while an agent follows its configured rules at machine speed. That can be an advantage when the rules include stop-losses, and a liability when the instructions are poorly written or the market behaves in ways the user did not anticipate.
Liquidation itself is a mechanical process. When a leveraged position loses enough value that its margin no longer covers the exposure, the venue closes it automatically and the trader loses the posted collateral. An agent that reports liquidation risk, as Walt says its agents can, gives users a way to see how close a position stands to that threshold, but the information is only useful if someone acts on it.
For that reason the user-set limits and the ability to monitor each action carry more weight than they might in a less volatile product. The design leaves responsibility for the instructions with the person, which is a sensible allocation, but it assumes that person understands the risks of the strategy they hand over.
7. Mira, Walt and the wider race to automate trading
Mira became available in February. Its developer said in May that it had passed 2 million users, including more than 500,000 monthly active users. Walt describes its own user base as tens of millions of people worldwide, which indicates how large the potential audience for agent-based trading inside Telegram could become.
Walt is not alone in pursuing the idea. Robinhood unveiled its own AI trading agents last month. Like Walt's design, they operate through a dedicated account, but manual approval of each trade is on by default, a more cautious starting point than letting the user switch on unattended execution. Robinhood describes the product as coming soon.
The contrast shows that firms are choosing different balances between convenience and control. Walt starts with execution off but lets users turn it on and then leaves the agent to act within the limits set. Robinhood begins by requiring a person to approve each trade. Both approaches keep a human in charge of the funding decision, yet they differ on how often the human must intervene.
8. What to watch after the announcement
At its rebrand on September 28, Walt said AI trading tools would arrive in a separate launch, without giving a date. Tuesday's release delivers on that promise, and the next developments to watch are the details the company has not yet published: leverage ceilings, pricing, availability by country and the mechanics of linking outside assistants.
Users will also reveal how comfortable people are with delegating leveraged trading to software. Adoption figures, any reports of agents behaving unexpectedly and the response of regulators to consumer-facing automated derivatives trading will help determine whether this design becomes a template or a cautionary example.
For now, the feature gives Telegram users a way to trade perpetual futures by instruction rather than interface, within boundaries the company says it has built in. The permissions are narrow, the default is off, and the funds are separate from the rest of a user's holdings. Whether those protections prove sufficient will only become clear once real money starts to flow through the agents.