Research Workflow
How to Build a Crypto Strategy with an AI Agent
How an AI agent can drive crypto strategy research in Traseq through the same authoring contract the no-code editor uses — build, backtest, compare, iterate.
Traseq··6 min read
How an AI agent can drive crypto strategy research in Traseq through the same authoring contract the no-code editor uses — build, backtest, compare, iterate.
Start with a no-code crypto spot strategy, lock the version, run the backtest, and keep the result traceable for comparison.
An AI agent can drive a full crypto strategy research loop in Traseq: author a strategy version, run a backtest, read the structured results, and decide what to test next. It does this through the Public Agent API — the same SignalGraph v2 authoring contract the no-code editor produces. Because the agent and the UI write to one contract, they cannot drift: a strategy an agent builds is the same kind of object a person builds by hand, and the results are reproducible either way.
Traseq is a research workspace, not a live trading or exchange-execution platform. An agent here does research only — it does not place orders, connect to exchange accounts, or guarantee performance. The output of an agent run is evidence to review, not a trade.
When people hear "AI trading bot," they usually picture something placing live orders. That is not what this is. In Traseq, an agent is an automation client that performs the same research workflow a person would, just programmatically:
The agent is doing the tedious, repeatable parts of research at machine speed. The judgment about whether a result is worth trusting still belongs to you. Every run produces a version-linked artifact you can open in the UI, inspect, and reproduce.
A useful agent workflow is a tight loop, not a one-shot call:
The discipline that matters in manual research matters here too: change one variable at a time, or you will not know what moved the result.
The point of routing both the agent and the editor through one contract is that nothing special happens for agents. A few practical consequences:
This is what separates agent-assisted research from a black box. The agent is fast, but the work it produces is the same inspectable, version-traceable artifact you would build yourself.
Be clear about what an agent in Traseq can and cannot do, so you aim it at the right job:
An agent that respects these limits is a research accelerator. One that pretends to do more is just a worse version of a trading bot you should not trust.
Before wiring up an agent, it helps to feel the research loop by hand. The interactive demo runs three system templates on real BTC/USDT 1h candles from a choppy, sideways-to-down window — and all three trend and breakout templates net-lost while mean-reversion barely broke even. That is not a flaw in the demo; that is exactly what honest backtesting looks like, and exactly the kind of result an agent should surface plainly rather than hide.
When you are ready to automate, the agent API workflow path walks through the loop end to end, and the docs cover the core concepts an agent and the UI both rely on.
To go deeper on the ideas behind this workflow, see research traceability and strategy versioning, how to build a crypto strategy without code, and how to compare backtest results. For the product scope, see crypto backtesting.
No. Traseq is a research workspace. An agent can author strategy versions, run backtests, and compare results, but it does not place orders, connect to exchange accounts, or execute live trades. The output is research evidence, not a trade.
It is the interface an AI agent or automation client uses to drive Traseq research — authoring a strategy version, finalizing it, running a backtest, and reading structured results. It writes to the same SignalGraph v2 authoring contract the no-code editor produces.
No. Both the agent and the UI write to one authoring contract, so an agent-built strategy is the same kind of object you build by hand. Results stay version-linked and reproducible, and you can open agent work directly in the editor.
No. An agent runs honest historical tests quickly; it does not find strategies that are guaranteed to work. A result that looks good in one window can fail in another, which is exactly why you backtest and compare before risking money.
Crypto spot research on major USDT pairs across large-cap, high-volume tokens, at 15m, 1h, 4h, and 1d timeframes. Conditions are evaluated on bar close, and signal-driven entries and exits fill at the next bar open.
| Property | Why it matters for agent research |
|---|
| No UI/agent drift | An agent-built strategy is the same object the editor builds — there is no second, divergent code path to maintain or distrust. |
| Reproducible results | Every backtest stays linked to a finalized version and its run configuration, so a result an agent produced can be reopened and re-run in the UI. |
| Human handoff | You can pick up an agent's work in the no-code editor, or hand a hand-built strategy to an agent, without translation. |
| Auditable research | Because versions and runs are traceable, you can answer "which exact rules produced this number?" months later. |
15m, 1h, 4h, and 1d timeframes.