Tria, a self-custody neofinance platform that says it serves more than 600,000 users, has launched Agentic Trading: AI agents can now execute perpetual futures strategies automatically on Hyperliquid and Decibel. The core of the announcement is less the automation than the control architecture, since agents never receive access to user funds. This analysis reviews the design, its place in the market and the risks it highlights.
🔑 Key takeaways
- Tria launches Agentic Trading, which links AI agents such as Claude or ChatGPT to existing accounts to trade perpetual futures.
- Agents operate through restricted trading keys that can open and close positions but cannot withdraw funds.
- Execution runs on Hyperliquid and Decibel infrastructure, with no separate wallet controlled by the AI.
- The company itself warns that trade-only permissions do not prevent losses, which can reach the full trading collateral.
- No usage data, performance figures or independent security audit have been published so far.
A launch rooted in a private beta
Tria presents Agentic Trading as a direct response to requests from traders in its private beta. The logic is straightforward: users define the entry and exit conditions for their positions, then authorize an agent to execute supported transactions once those conditions are met. The announcement gives the example of an instruction telling the agent to monitor the price of bitcoin (BTC), open a position at a set entry point and close it when a predefined target is reached.
The product is powered by the Tria Trade SDK, whose integration instructions are published at api-docs.tria.so. The compatible tools cited are Claude, ChatGPT, Codex and Cursor, connected to existing Tria accounts. The stated goal is to remove the need for constant chart monitoring and manual order entry.
“Traders in our private beta have been asking for this capability, and we’re delivering it while continuing to build out the platform. AI is already changing how people analyze markets, but traders increasingly want to use it to execute their strategies as well.”
Vijit Katta, Co-founder and CEO, Tria
A permission system at the core of the design
Tria’s distinctive feature is the separation between execution and custody. Agents are not connected to a separate wallet funded by the user. They operate through trading keys created specifically for the chosen platform, Hyperliquid or Decibel. The table below summarizes the rights attached to these keys, as described in the announcement.
| Function | Trading key | User control |
|---|---|---|
| Open positions | Yes, on supported markets | Conditions defined in advance |
| Close positions | Yes | Targets and thresholds configured |
| Withdraw funds | No | Custody stays with the user |
| Validity period | Configurable | Set by the user |
| Activity tracking | Visible in the Tria app | Reviewable at any time |
| Trading access | Revocable at any time | User decision |
According to the announcement, activation involves selecting the platform in the app, creating a key for it, configuring its permissions and then following the integration instructions of the chosen AI tool. Once the connection is established, supported strategies can be automated and activity remains tracked directly within Tria.
What ‘trade-only’ really means
A trade-only key is a technical safeguard against withdrawals, not against losses. A misconfigured agent, or one reacting unexpectedly to a volatile market, can open unfavorable positions without being able to move the funds elsewhere. This distinction matters for judging the real scope of the protection, and it sits at the heart of the risk warnings in the release.
A trend that extends beyond Tria
Tria is not the first to offer agentic trading. According to Corporate Insight, Public launched in March 2026 the first brokerage offering to officially adopt the term ‘Agentic Brokerage’. Its investors describe a strategy in natural language, after which the agent monitors conditions in real time and executes trades. At the time of the analysis, the offering was limited to the desktop interface, with a mobile rollout planned.
- Robinhood: Agentic Accounts connects a third-party agent to a dedicated account funded by a separate budget. In the test reported by Corporate Insight, Claude could build a portfolio from the user’s preferences. At launch, the feature was limited to equities.
- eToro: Agent Portfolios connects a custom agent to a dedicated sub-portfolio through an API key with limited scope. The minimum investment is $200, with scheduled rebalancing or trading triggered by earnings releases or macroeconomic indicators.
- Gemini: Agentic Trading, launched in April 2026, connects third-party AI services to crypto account tasks, including risk management. Gemini also introduced modular ‘Trading Skills’ that provide real-time market data and historical candlestick data.
Retail interest that remains poorly measured
The announcement cites an Investing.com survey published in April 2026, in which 62% of US retail investors surveyed said they had used AI to inform their investment decisions. Among those who had experimented with AI, only 4.5% said they had used automated trading algorithms. The release does not disclose the sample size or the methodology, so these figures should be read with caution, even if they indicate a direction.
A documented precedent from MoonPay
An article by Team MoonPay published on April 8, 2026 places the development in a longer timeline. It states that from 2024, trading bots began moving beyond fixed rules to become autonomous agents powered by language models, able to interpret market narratives and execute multi-step strategies. According to the article, these agents can analyze social media and news sentiment, identify chart patterns, adjust risk management and modify their strategies based on results.
The article describes a 14-week beta program run from October 2025 to January 2026, during which more than 1,000 participants created over 9,500 agents, executing 187,000 autonomous crypto transactions. It also claims that several funds observed a measurable edge from AI-assisted strategies over human-managed portfolios, without naming those funds or disclosing their methods or results. These points remain claims made by the publisher and have not been independently verified.
Limits and risks: what the figures do not say
The available sources publish no quantified market reaction to the launch. No usage volume, number of users enabling the feature, revenue or price movement is disclosed. No performance data, transaction volume, commissions, position limits or geographic eligibility is published either, and no independent security audit is mentioned. The only indication of demand remains the request for the feature from traders in the private beta.
“AI-assisted trading does not guarantee profitable outcomes. Trade-only permissions prevent agents from withdrawing funds but do not prevent trading losses. Perpetual futures trading involves substantial financial risk, including the potential loss of trading collateral.”
Tria press release, distributed via Chainwire, October 30, 2026
Corporate Insight argues that automation can allow strategies to run continuously and reduce hesitation or panic. It also stresses that agents making errors, or acting as instructed in unexpected market conditions, can generate losses quickly. The firm warns that the ease of automation can create a false sense of security, particularly among investors who do not fully understand how agents work. These remarks are not an assessment of Agentic Trading’s performance.
Conclusion: a control innovation, not a yield promise
Agentic Trading marks a clear shift: AI moves from analysis to execution while custody of funds stays with the user. Technically, separating the right to trade from the right to withdraw is a sound architectural choice. Economically, however, nothing in the available material measures adoption or real-world performance.
Three scenarios deserve monitoring. The first would confine adoption to experienced traders. The second, more ambitious, would involve broader rollout if independent audits and usage data follow. The third, more adverse, would be a market incident involving misconfigured agents that weighs on confidence across the segment. Tria’s upcoming disclosures, particularly on volumes and security, will be decisive.
Sources
- The Defiant, Tria press release
- TradingView, Chainwire release
- Google redirect to the Tria release
- Washington City Paper, Tria press release
- Corporate Insight, analysis of agentic trading
- Chainwire, Tria release of October 30, 2026
Cet article est publié à titre informatif et éducatif. Il ne constitue en aucun cas un conseil en investissement. Faites vos propres recherches (DYOR) avant toute décision.
