15+
Agent profiles evaluated
Historical Lab, Drift, and Observatory datasets used for replay and analysis.
About
Built to keep traders liquid.
Beneat builds decision-quality infrastructure for traders and autonomous agents: risk controls, behavioral telemetry, and guardrails that keep operators solvent under pressure.
Most trading stacks optimize for execution. We optimize for the quality of the decision before execution — the sizing, authority, session state, and risk behavior that decide whether an edge survives contact with the market.
The work started as an internal terminal and widened into a Lab/Observatory loop for evaluating human and machine operators under real market pressure.
66K+
Logged trade rows
Agent trade histories captured for behavioral analysis, risk replay, and guardrail design.
4
Live Observatory models
GLM-5, Kimi K2.5, MiniMax 2.5, and Qwen3 Max tracked in the open.
1.7K+
Closed live trades
Recorded across the active Observatory agents at the latest endpoint check.
Execution was not the hard part.
Most trading stacks optimize for order flow. Beneat began from the opposite premise: decision quality breaks before execution does.
Risk needed to move upstream.
We built Beneat for internal trading: enforce risk before orders hit the market, track behavior, read session state, and keep the operator liquid long enough for edge to matter.
The operator became the surface.
Human state, LLM behavior, and deterministic guardrails all changed outcomes. That widened the product from terminal to decision-quality infrastructure.
Why now
The risk surface changed.
AI trading moved faster than controls.
Agent demos celebrate autonomy. Markets punish unfenced authority. Beneat studies where models overtrade, ignore context, or become confidently wrong.
Risk is behavioral, not only financial.
Sizing, fatigue, revenge, leverage, and model drift are part of the same decision surface. The system has to measure the operator before it grants authority.
Evaluation feeds product.
Lab and Observatory experiments become guardrail logic, risk policies, and telemetry inside the terminal.
Research
Agents made the problem louder.
As exchanges and social feeds shifted toward AI agent trading, the gap became obvious. LLMs can describe markets. That does not make them safe trading vehicles.
In trading, deterministic algorithmic strategies are usually the stronger execution primitive. LLMs need hard guardrails, narrow authority, and systems that stop them before a fluent mistake becomes a financial one.
We ran agent evaluations and risk experiments instead of arguing about the narrative. That work feeds back into the terminal, into the Lab and Observatory, and into DQS: the broader question of how any operator, human or machine, earns authority through behavior.
Team
The team
Filip Tomáška
Co-founder
Company development, partnerships, fundraising, and market strategy.
Muhib Qureshi
Co-founder
Core product architecture, agent evaluations, risk experiments, and risk-system design.
Alberto Maschietto
Infrastructure, Hardware & Security
Infrastructure, hardware systems, cybersecurity, and technical hardening for the Lab and terminal.
Strategic conversations