Deterministic Agentic Engine
Algosmithy
Deterministic Go-based engine for autonomous LLM-generated trading strategies and reproducible backtesting.
Demonstration examples — not trading advice.
What Algosmithy Is
Algosmithy runs a fully autonomous agentic loop where an LLM generates a strategy, compiles it, executes it inside an isolated sandbox, evaluates the results, and iterates until performance targets are met.
- •Deterministic Go actor-engine with sync and async modes.
- •LLM tool interface for data, backtests, memory, and finalization.
- •Sandboxed Docker execution for safety and reproducibility.
- •LLM-generated research artifacts (not trading strategies) available in the public repo.
Key Features
Agentic workflow
- Fully autonomous strategy synthesis → backtest → refinement loop
- Structured tool interface for market data, backtesting, and memory
- Deterministic execution and reproducibility
Go-based execution engine
- Actor-model runtime (sync deterministic + async multi-threaded modes)
- Unified strategy API (
ICandleStrategy,IOrderBookStrategy) - Multi-currency accounting and precision via
decimal.Decimal
Backtesting
- Parameter grid search via YAML
- Sandbox execution inside Docker
- Time-ordered event playback
Data & extensibility
- Pluggable exchange connectors (currently Bybit)
- In-memory and file-based historical caching
- Multi-instrument, multi-timeframe pipelines
LLM-Generated Strategy Examples
These examples were generated entirely by the Algosmithy agentic loop without manual edits.
EMA Crossover
Multi-EMA/MACD-style crossover with grid-searched params.
Bollinger Bands Breakout
Breakout logic with LLM-generated indicator implementation.
Demonstration examples — not trading advice.
Architecture Documentation
Get in touch
Open to collaboration with quant teams, AI infra groups, and funds exploring autonomous research loops.