# llama-wolf

- **Event:** [Agentic Memory & Context Engineering Hackathon](https://cerebralvalley.ai/e/mongoDB-hackathon)
- **When:** Sat, Oct 11 at 9:00 AM – 10:00 PM (PDT)
- **Where:** Shack15, San Francisco, CA
- **Team:** [David Mayboroda](https://cerebralvalley.ai/u/davidmayboroda), [Baladhurgesh Balagurusamy Paramasivan](https://cerebralvalley.ai/u/baladhurgesh)
- **GitHub:** https://github.com/Baladhurgesh/llama-wolf
- **Demo video:** https://docs.google.com/presentation/d/1FHIYJSX8ulI9NxARyeWiCXKYnZ4d4DIe4asX7OmLSbk/edit?usp=sharing
- **Gallery:** https://cerebralvalley.ai/e/mongoDB-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongoDB-hackathon/hackathon/gallery/25

We build a social-deduction game arena (Werewolf) to test a practical memory architecture for language-model agents. Instead of retrieving long chat logs, our agent learns strategy memory: compact, reusable snippets that map situation → tactic → why. After each game, we distill tactics from logs (positive and negative), keyed by role (Villager/Seer/Werewolf), phase (accuse/defend/vote/night), and a small set of cue features (e.g., day bucket, counterclaim status, pressure, claims). At decision time, the agent filters by role/phase, recalls top-k tactics via a vector search over a canonicalized “situation string,” 

We run a teacher→student setup: GPT-5 and Llama-Maverick generate diverse training games; we extract and score tactics (win rate, vote-shift, acceptance/backfire, recency). At runtime, only Llama plays, with and without memory, against GPT-5 and Llama opponents under identical prompts and sampling to isolate memory effects.

---

Markdown version of https://cerebralvalley.ai/e/mongoDB-hackathon/hackathon/gallery/25. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt