General Liquidity operates as an applied research lab, not just a product company. We're focused on problems that don't have off-the-shelf solutions.
Agent infrastructure — The unsexy plumbing that makes everything else possible.
State management, tool interfaces, and execution environments that let AI systems operate reliably when real money is on the line. A tool registry for standardized access to any exchange or data source. An event bus for real-time reaction without polling.
Multi-agent reasoning — Systems that argue with themselves before they act.
Single-agent systems hallucinate and lack adversarial pressure. We're building architectures that replicate institutional structure: analysts gather data, bull and bear researchers debate the thesis, risk teams pressure-test execution, portfolio managers synthesize and decide.
Financial memory — Systems that learn from their own history.
When a trade closes, the system reflects: What was the thesis? What did analysts see? What did risk assessment miss? Lessons get stored and retrieved. Over time, accumulated reflection becomes something like judgment.
Intent translation — Bridging human wants and safe machine execution.
"Hedge my ETH exposure if volatility spikes" contains remarkable implicit context. The system unpacks meaning, generates concrete plans, shows implications across scenarios, and waits for approval.
Simulation and time travel — Agents prove themselves before risking real capital.
Sandboxed environments where strategies run against historical data as if it were live. The agent wakes up believing it's January 2023, receives price data tick by tick, and watches decisions play out.
Regime awareness — Recognizing when the game has changed.
Markets shift. Correlations break. Liquidity evaporates. We're researching how agents perceive and adapt to changing regimes, understanding when historical assumptions no longer apply.
How We Build
Risk before speed.
Fast systems that blow up are worthless. We design for survival first. Every feature gets evaluated: what happens when this fails at the worst possible moment?
Memory over metrics.
Dashboards tell you what happened; memory tells you why. We're building systems that accumulate genuine judgment, not just data points.
Research and product together.
Papers without products are speculation. Products without research are toys. Deployment is part of the scientific method.
The best systems argue with themselves.
Confidence without doubt is dangerous. We design agents that debate and critique before acting, surfacing weaknesses internally before the market does.
Join Us
If you're interested in working on these problems, we'd like to hear from you. We're looking for people who want to build systems that matter, at the intersection of AI and markets.
Get in touch