TL;DR: Senpi and Trigr both run AI trading agents on Hyperliquid perps, but they start from different places. Senpi is conversational: you talk to a personal agent in Telegram, and its documented strategies are forward-tested live. Trigr is validation-first: strategies are backtested on point-in-time data with explicit costs, optionally checked by ML runs that report the Deflated Sharpe Ratio, then deployed as paper or non-custodial live agents, from Studio or from ChatGPT, Claude or Codex over MCP. Choose Senpi for chat-native trading and copy-style templates; choose Trigr if you want evidence before capital.
Senpi facts below come from its launch press release, its open-source repository and funding coverage, and are stated as of September 2026.
What is Senpi?
Senpi launched personal trading agents for Hyperliquid on February 24, 2026, according to its launch press release. The release describes:
- agents built on the OpenClaw agent framework, preconfigured with 31 trading tools;
- persistent memory that learns your risk profile over time;
- a Telegram co-pilot mode, where you chat with the agent;
- an autonomous mode, where the agent executes trades, manages leverage and applies dynamic stop losses on your behalf.
Senpi also publishes an open-source skills repository under the MIT license, with 80+ strategy templates (some of them copy-trading), and states that each strategy instance needs its own funded Hyperliquid wallet. The repository lists about 230 crypto assets plus roughly 95 equities, metals, indices and pre-IPO names on Hyperliquid. The company raised a $4M seed round led by Lemniscap with Coinbase Ventures in September 2025. Pricing was not disclosed in the sources we reviewed.
Senpi's homepage now describes the product as "your AI quant engineer," which puts it in direct competition with strategy builders such as Trigr.
What is Trigr?
Trigr is an AI strategy builder, point-in-time backtester, ML validation tool and trading-agent platform for Hyperliquid perpetual futures. You build a strategy as a node graph (trigger, filters, signal, risk, execute) in Studio, ask the AI copilot to draft it from plain English, or have your own assistant build it over MCP. The no-code strategy builder guide shows the structure.
Every strategy is backtested before it runs. You can then deploy it as a paper agent, a live Hyperliquid agent, or a Propr prop-firm challenge agent, with up to six strategy slots per agent on different assets.
How do Trigr and Senpi compare?
| Topic | Senpi | Trigr |
|---|---|---|
| Venue | Hyperliquid perps | Hyperliquid perps; Paper; Propr challenge accounts |
| TradFi markets | Equities, metals and indices on Hyperliquid, per its repo | HIP-3 TradFi perps for backtests and paper agents only in the current beta |
| Primary interface | Telegram chat with a personal agent | Studio node graph, AI copilot, or ChatGPT, Claude and Codex over MCP |
| Strategy evidence | Forward-tested live in its Agents Arena; no built-in backtesting documented | Point-in-time backtest on full history before deployment; paper forward test |
| Costs in testing | Not documented | Fees and builder fee always; slippage and funding opt-in, gross result labeled |
| Overfitting checks | Not documented | ML runs report leak verdict, OOS Sharpe, DSR and PBO |
| Custody | Press release: "no wallets to manage, no private keys to rotate"; mechanism not confirmed | Trade-only agent key that cannot withdraw; you keep the master key |
| Copy trading | Copy-trading templates in its library | Not live (paused) |
| Open source | Skills repo under MIT | Strategy recipes stay private; marketplace of versioned strategies |
| Pricing | Not disclosed in our sources | Free, Trader $29/mo, Pro $99/mo |
Backtested or forward-tested: why does it matter?
This is the core difference. A forward test is real: trades happened in live markets. But it covers only the period since the strategy started, and it cannot tell you how the same rules behaved through earlier regimes, such as a sharp drawdown or a long chop. A few months of live results can look very good or very bad for reasons that have little to do with the rules.
A backtest answers a different question: how would these exact rules have traded over years of history, with costs? It is only useful if it is honest. Trigr's engine follows fixed conventions described in the backtesting docs:
- Point-in-time data. Every feature uses only information available at the bar's close; slower inputs are carried forward, never read ahead.
- Next-bar-open fills after a signal.
- Conservative intrabar handling. If a take-profit and a stop-loss are both touched in one bar, the engine replays 5-minute bars; if still ambiguous, the stop wins.
- Explicit costs. Fees always apply. Slippage (flat bps per fill) and funding (historical or a flat per-8h rate) are opt-in, and the first result is labeled gross so you know which view you are reading.
When you test many variants, luck creeps in. Trigr's ML optimization runs report the Deflated Sharpe Ratio (Bailey and López de Prado) using the run's trial count, plus the Probability of Backtest Overfitting. A plain backtest shows those as not yet computed rather than implying them.
The two approaches are not opposites. Trigr also forward-tests: a paper agent fills at the current Hyperliquid price with the taker and builder fees, and the agent view marks where the backtest ends and the forward record begins. The point is the order: history first, then a forward test, then money. See paper trading agents for how to run that step.
Telegram copilot or your own assistant over MCP?
Senpi's interface is a personal agent you talk to in Telegram, with memory of your preferences. That is convenient: you can ask about a position from your phone and act in the same thread.
Trigr takes the opposite approach to AI access. Instead of shipping its own chat agent, it exposes a remote MCP server that ChatGPT, Claude, Claude Code and Codex can connect to. Your assistant can search the capability catalog, draft and edit strategies, run backtests with optional slippage and funding, read stored runs and full trade logs, iterate with labeled experiments, and create a paper agent in a paused state.
Hard limits are part of the design. Over MCP, an assistant cannot start execution, place a live trade, withdraw funds or edit exchange credentials. Consequential actions need a browser approval (manual mode) or run under an autonomous mode you choose per connection, and each authorization is single-use, lasts ten minutes and is bound to the exact tool arguments. Going live happens in the Trigr app. The article on AI trading agent safety and approvals covers the details.
Which is better depends on how you work. If you want a trading companion that acts, Senpi's model fits. If you want an assistant that does research and testing while live authority stays with you, Trigr's fits.
How does custody compare?
Senpi's press release says users have "no wallets to manage, no private keys to rotate," and its repository requires a funded Hyperliquid wallet per strategy instance. We could not confirm from a primary source whether that wallet is embedded, MPC-based or a delegated agent wallet, so we do not characterize it further.
Trigr uses Hyperliquid's API wallet model. You approve a trade-only agent key that can place and cancel orders but cannot withdraw or send funds to another address. You keep your master wallet key and custody of your funds. For autonomous agents, Trigr stores that trade-only key encrypted server-side; the key is valid for about 90 days. Hyperliquid's own documentation on API wallets and nonces explains the mechanism, and our explainer on Hyperliquid agent wallets covers what it means in practice.
When is Senpi the better choice?
Senpi is a strong product with real strengths:
- You want to trade by conversation. A Telegram agent with memory is a natural fit if chat is how you like to work.
- You want copy-style and trader-discovery strategies. Senpi's library includes copy-trading templates and tools such as trader risk scores. Trigr's copy trading is paused.
- You want to read or fork strategy code. Senpi's skills are open source under MIT.
- You want live TradFi exposure on Hyperliquid now. Senpi's repo lists equities, metals and indices. On Trigr, HIP-3 TradFi markets are backtest and paper only in the current beta.
When is Trigr the better choice?
- You want to see how a strategy behaved over years of history, with fees, slippage and funding stated plainly, before any capital is involved.
- You test many variants and want DSR and PBO rather than a raw Sharpe.
- You want to build with ChatGPT, Claude or Codex under exact-argument approvals, and keep live authority in your own hands.
- You want alternative data (funding, open interest, liquidations, macro series) as no-code building blocks.
- You want paper, live Hyperliquid and Propr challenge agents from the same strategy, and a marketplace of immutable, versioned strategies at Discover.
Backtests are not guarantees; perps are leveraged and can lose more than expected.
Next steps
If you are weighing Senpi against Trigr, pick one idea, backtest it on Trigr gross and then net, and run it as a paper agent for a few weeks. The deployment guide walks through the path from backtest to live agent.