Request access

How Pragma works

Pragma runs research as a pipeline of AI agents. Each stage has a clear output, and every stage runs inside the same guardrails.

GuardrailsOut-of-sampleRealistic costsVersioned runsBounded agentsHypothesizeRationale, data needs,acceptance criteriaImplementHypothesis becomesstrategy codeValidateBacktest with costs,out-of-sample, walk-forwardRun the full pipeline or any stage on its own. Pick the model and provider per agent.GuardrailsOut-of-sampleRealistic costsVersioned runsBounded agentsHypothesizeRationale, data needs,acceptance criteriaImplementHypothesis becomesstrategy codeValidateBacktest with costs,out-of-sample, walk-forwardRun the full pipeline or any stageon its own. Pick the model andprovider per agent.

  1. Hypothesize

    Output: a hypothesis card — the statement, its rationale, a testable prediction, the data it needs and the criteria it must meet.

  2. Implement

    Output: strategy code generated from the hypothesis card, ready for backtesting.

  3. Validate

    Output: a backtest that includes spread, commission and slippage, out-of-sample and walk-forward checks, and a verdict: validated or rejected.

Each stage can run on its own: generate hypotheses only, go from hypothesis to code, or run the full pipeline.

Guardrails

  • Out-of-sample and walk-forward

    Every hypothesis is tested on data it never saw.

    A strategy that only works on the data it was built on is not a result. Pragma holds data back and re-tests across rolling windows before a hypothesis can pass.

  • Realistic costs

    Spread, commission and slippage are part of every backtest.

    Short-horizon ideas often vanish once trading costs are counted. Costs are part of the backtest from the first run, not an adjustment at the end.

  • Reproducible runs

    Data, code and prompts are versioned, so any run can be repeated.

    Every run records the data, code and prompts it used, so a result can be traced and repeated.

  • Bounded autonomy

    Agents work within loop and budget limits.

    Agents have hard limits on iterations and spend. When a limit is reached, the run stops instead of drifting.

Any model, any agent

Choose the model and provider for each agent. Pragma works with Claude (Anthropic), OpenAI models and other providers. Market data and execution run through MetaTrader 5.

FAQ

Is Pragma a trading bot?

No. Pragma is research software. It helps you form and test trading hypotheses; it does not trade for you or tell you what to trade.

Do you provide signals?

No.

Which platforms does it work with?

MetaTrader 5 data.

Which AI models does it use?

You pick the model and provider for each agent.

Can I run only one stage?

Yes. Each stage runs on its own or as part of the full pipeline.

Who is it for?

Systematic traders and quant researchers.

Pragma is research software. It does not provide investment advice, trading signals or portfolio management. Simulated results do not guarantee future performance. Trading leveraged instruments carries a high risk of loss.