Abstract
Long-horizon strategic planning in complex strategy games requires coordinating tightly coupled decision domains, including technology, economy, diplomacy, and military, across hundreds of turns under imperfect information. Existing LLM-based agents face three challenges in this setting:
- They recover little relational structure, such as distance or threat, from raw coordinate observations.
- They serialize the entire growing game state into every prompt and plan all domains in a single output, causing context overflow and urgent domains crowding out long-term ones.
- The only reward is a lagged final score, which provides little progress signal within or across games.
We present SAGA, an LLM multi-agent framework that addresses these challenges with three mechanisms:
- Map-Semantic Scene Graph: renders typed entity relations as concise per-entity text.
- Tool-Augmented Planner: retrieves state on demand and routes disjoint per-domain directives to specialist controllers.
- Dual-Horizon Feedback Loop: sets intermediate goals within a game and distills causal lessons across games.
On CivRealm, a benchmark built on the strategy game FreeCiv, SAGA attains the highest mean final score among six methods, the statistically strongest gains on infrastructure, and a 27% reduction in output tokens. With cross-game evolution enabled, it reaches the highest score over five successive games.
Blogger's Review: The introduction of SAGA not only effectively addresses long-term planning issues in complex strategy games but also showcases the potential and flexibility of LLM in dynamic environments through its multi-agent collaboration and feedback mechanisms. Its future development and applications are worth monitoring.