$ show chrisworsey55/atlas-gic
ATLAS is a framework for autonomous AI trading agents that self-improve through market feedback. It implements a 4-layer multi-agent architecture (Macro, Sector Desks, Superinvestors, Decision) where each agent's prompt is treated as trainable weights and optimized via a git-based evolutionary loop. The codebase includes a meta-weighting layer (JANUS) that blends predictions from multiple trained cohorts, and a MiroFish integration for simulating future market scenarios using swarm intelligence. While the repository contains well-structured Python modules for the orchestration layer, simulation bridge, and forward training, the critical agent prompt files and internal data clients are missing (marked proprietary). The result is a reference architecture rather than a runnable system; however, the patterns demonstrated in janus.py and the mirofish bridge are cleanly implemented and could serve as templates for similar meta-learning or simulation-augmented agent systems.
The LLM identified 2 files worth highlighting for reuse:
Implements a meta-layer that dynamically weights recommendations from multiple agent cohorts based on rolling accuracy. Features softmax weight allocation with min/max constraints, automatic regime detection (NOVEL_REGIME/HISTORICAL_REGIME/MIXED) from weight differentials, and blended recommendation generation with disagreement penalties. This is a self-contained, reusable pattern for any multi-model ensemble system where confidence weighting and regime detection are needed.
Provides a bridge between swarm intelligence simulations and trading decision-making. Uses a lightweight mode that leverages Claude to simulate multi-agent interactions (hedge fund managers, central bankers, retail) with reflexive feedback loops. Includes seed generation from market data, scenario generation based on agent debates, and a prediction scorer. While dependent on missing internal data clients, the architecture for generating simulation prompts, parsing results, and scoring predictions is cleanly abstracted and could be adapted to other simulation-augmented agent frameworks.