Skip to content

Models Reference

agents.Agent

LLM-based agent configured to play IPD.

Field Type Default Description
name CharField(100) — Unique agent name
slug SlugField auto URL-safe identifier
role CharField(200) — CrewAI role
goal TextField — CrewAI goal
backstory TextField — CrewAI backstory
llm_model CharField(100) gpt-4.1-mini LLM model identifier
temperature FloatField 0.0 Sampling temperature
system_prompt_override TextField "" Optional full prompt override
persona_type CharField(50) — cooperative, tit_for_tat, selfish, forgiving, random, custom
avatar_emoji CharField(10) 🤖 Display emoji
color CharField(7) #3B82F6 Hex color for charts
is_active BooleanField True Whether agent appears in lists
test_chat_history JSONField [] Saved chat preview messages
created_at DateTimeField auto —
updated_at DateTimeField auto —

Methods:

  • to_yaml() → str — Export as CrewAI-style YAML
  • from_yaml(yaml_str, name) → Agent — Import from YAML

agents.PolicyAgent

Deterministic strategy agent (no LLM).

Field Type Default Description
name CharField(100) — Unique name
slug SlugField auto URL-safe identifier
policy CharField(20) — ALLC, ALLD, TFT, GRIM, GTFT, WSLS, RANDOM
alpha FloatField 0.5 Cooperation probability (for RANDOM policy only)
avatar_emoji CharField(10) ⚙️ Display emoji
color CharField(7) #6B7280 Hex color

games.Experiment

A full experiment with conditions and replicates.

Field Type Default Description
name CharField(200) — Experiment name
description TextField "" Optional description
n_rounds IntegerField 100 Rounds per game
horizon_type CharField(20) fixed fixed or geometric
stop_prob FloatField 0.02 Geometric stopping probability
replicates IntegerField 10 Games per condition
payoff_matrix JSONField {CC:[3,3],...} Payoff values
status CharField(20) draft draft, running, completed, failed
started_at DateTimeField null When run began
completed_at DateTimeField null When run finished

games.ExperimentCondition

A specific agent matchup within an experiment.

Field Type Description
experiment FK → Experiment Parent experiment
name CharField(200) Condition name (e.g., "Cooperative vs ALLD")
agent_a_llm FK → Agent (nullable) LLM agent for side A
agent_a_policy FK → PolicyAgent (nullable) Policy agent for side A
agent_b_llm FK → Agent (nullable) LLM agent for side B
agent_b_policy FK → PolicyAgent (nullable) Policy agent for side B
order IntegerField Display order

Each side must have exactly one agent assigned (either LLM or Policy, not both).

games.Game

A single game instance (one replicate of one condition).

Field Type Description
condition FK → ExperimentCondition Parent condition
replicate IntegerField Replicate index (0-based)
seed IntegerField RNG seed for reproducibility
total_rounds IntegerField Actual rounds played
agent_a_total_payoff IntegerField Cumulative payoff A
agent_b_total_payoff IntegerField Cumulative payoff B
cooperation_rate_a FloatField (nullable) Computed metric
cooperation_rate_b FloatField (nullable) Computed metric
mutual_cooperation_rate FloatField (nullable) Computed metric
mutual_defection_rate FloatField (nullable) Computed metric
status CharField(20) pending, running, completed, failed

games.GameRound

A single round within a game.

Field Type Description
game FK → Game Parent game
round_index IntegerField 0-based round number
agent_a_action CharField(1) C or D
agent_b_action CharField(1) C or D
agent_a_payoff IntegerField Payoff for A this round
agent_b_payoff IntegerField Payoff for B this round
agent_a_cum_payoff IntegerField Running total for A
agent_b_cum_payoff IntegerField Running total for B
agent_a_reasoning TextField LLM's explanation (optional)
agent_b_reasoning TextField LLM's explanation (optional)

Constraints: unique_together = ['game', 'round_index']