Available Agents

This page lists all 124 negotiating agents available in the library, organized by ANAC competition year. Each agent is a Python reimplementation of its original Java Genius counterpart.

AI-Generated Implementations

All agents in this library are AI-generated reimplementations based on the original Java source code from the Genius framework. While we strive for behavioral fidelity, they may not behave identically to the originals in all cases.

Quick Reference

Group	Winner	Total Agents
basic	TimeDependentAgent	5
anac2010	AgentK	7
anac2011	HardHeaded	6
anac2012	CUHKAgent	7
anac2013	TheFawkes	7
anac2014	AgentM	15
anac2015	Atlas3	22
anac2016	Caduceus	14
anac2017	PonPokoAgent	17
anac2018	AgreeableAgent2018	15
anac2019	AgentGG	14

Basic Agents

The basic group contains fundamental time-dependent agents that serve as baselines and building blocks.

from negmas_genius_agents import get_agents

# Get all basic agents
basic_agents = get_agents(group="basic")

Agent	Parameter	Strategy Summary
`TimeDependentAgent`	e=custom	Base class with formula: `target(t) = Pmin + (Pmax-Pmin) * (1 - t^(1/e))`
`TimeDependentAgentBoulware`	e=0.2	Tough negotiator; concedes slowly, mainly near deadline
`TimeDependentAgentConceder`	e=2.0	Cooperative; concedes quickly early in negotiation
`TimeDependentAgentLinear`	e=1.0	Constant concession rate throughout
`TimeDependentAgentHardliner`	e=0	Never concedes; always offers best bid for self

ANAC 2010

The first ANAC competition featured 7 agents.

Agent	Rank	Strategy Summary
`AgentK`	1st	Statistical opponent modeling with adaptive target; tracks mean/variance of offers
`Yushu`	2nd	Time-dependent concession with best-10 opponent bid tracking
`Nozomi`	3rd	Threat-based concession; mirrors opponent's concession behavior
`IAMhaggler`	4th	Bayesian opponent model with Nash-product bid selection
`AgentFSEGA`	-	Boulware concession with opponent reservation value estimation
`AgentSmith`	-	Multi-strategy; classifies opponent as HardHead/Conceder/Random/TFT
`IAMcrazyHaggler`	-	Random high-utility bidding with stubborn acceptance (exploits conceders)

ANAC 2011

Agent	Rank	Strategy Summary
`HardHeaded`	1st	Boulware concession with frequency-based opponent weight estimation
`Gahboninho`	2nd	Three-phase strategy (early/main/panic) with opponent cooperativeness tracking
`IAMhaggler2011`	3rd	Adaptive concession rate based on opponent behavior trends
`AgentK2`	-	Enhanced AgentK with improved statistical acceptance
`BramAgent`	-	Boulware with issue frequency tracking
`TheNegotiator`	-	Four-phase strategy with linear regression opponent prediction

ANAC 2012

Agent	Rank	Strategy Summary
`CUHKAgent`	1st	Two-phase concession with variance-based opponent toughness detection
`AgentLG`	2nd	Multi-phase (early/compromise/panic/end) with value preference learning
`OMACAgent`	3rd	Adaptive Boulware with opponent reservation value estimation
`TheNegotiatorReloaded`	-	Improved TheNegotiator with toughness-based adaptation
`MetaAgent2012`	-	Blends Boulware/Linear/Conceder based on opponent concession rate
`IAMhaggler2012`	-	Nash-product maximization with consistency-weighted opponent model
`AgentMR`	-	Three-phase (exploration/exploitation/compromise) with risk awareness

ANAC 2013

Agent	Rank	Strategy Summary
`TheFawkes`	1st	AC_Next acceptance with adaptive concession and opponent classification
`MetaAgent2013`	2nd	Strategy portfolio with domain-based selection
`TMFAgent`	3rd	Time-management focused with exploration/exploitation balance
`AgentKF`	-	Kalman filter-inspired statistical tracking
`GAgent`	-	Nash-optimal bid selection with frequency analysis
`InoxAgent`	-	Robust Boulware (e=0.05) with minimal opponent modeling
`SlavaAgent`	-	Weighted bid scoring with issue importance estimation

ANAC 2014

Agent	Rank	Strategy Summary
`AgentM`	1st	Simulated annealing bid search with adaptive acceptance
`DoNA`	2nd	Deadline-oriented with domain analysis and opponent sensitivity
`Gangster`	3rd	Multi-strategy gang voting (randomizes between tactics)
`WhaleAgent`	-	Boulware curve with Nash product bid selection
`TUDelftGroup2`	-	Polynomial concession with issue weighting
`E2Agent`	-	Simple exploration-exploitation balance
`KGAgent`	-	Knowledge-guided adaptive threshold
`AgentYK`	-	Three-phase: hardball, concession, final flexibility
`BraveCat`	-	BOA framework with combined acceptance criteria
`Atlas`	-	Precursor to Atlas3 with Pareto estimation
`Aster`	-	Multi-criteria star-pattern bid selection
`ArisawaYaki`	-	Wave-based oscillation in concession
`AgentTD`	-	Classic time-dependent (Boulware/Conceder switch)
`AgentTRP`	-	Trade-off, Risk, and Pressure balancing
`AgentQuest`	-	Quest-based goal setting with adaptive thresholds

ANAC 2015

Atlas3 was the winning agent and became highly influential in subsequent years.

Agent	Rank	Strategy Summary
`Atlas3`	1st	Three-phase Boulware with popular bid tracking for end-game
`ParsAgent`	2nd	Nash product optimization with frequency-based opponent model
`RandomDance`	3rd	Randomized exploration with adaptive acceptance
`AgentBuyog`	-	Three-phase (tough/moderate/flexible) with deadline handling
`AgentH`	-	Hybrid time-dependent with opponent tracking
`AgentHP`	-	High-performance polynomial concession
`AgentNeo`	-	"Chosen one" strategy with opponent hardness response
`AgentW`	-	Weighted multi-criteria bid selection
`AgentX`	-	Experimental adaptive concession
`AresParty`	-	War-themed aggressive then cooperative
`CUHKAgent2015`	-	Updated CUHK strategy with improved opponent model
`DrageKnight`	-	Knight-themed defensive strategy
`Y2015Group2`	-	Team-based development, balanced approach
`JonnyBlack`	-	Dark-themed hardball strategy
`Kawaii`	-	"Cute" adaptive strategy (actually quite aggressive)
`MeanBot`	-	Mean/average-based decision making
`Mercury`	-	Fast concession with deadline awareness
`PNegotiator`	-	Probabilistic acceptance with exploration
`PhoenixParty`	-	Rising from rejection with adaptive recovery
`PokerFace`	-	Bluffing-inspired concealment of intentions
`SENGOKU`	-	Japanese warfare-inspired multi-phase
`XianFaAgent`	-	Chinese strategy with wisdom-based decisions

ANAC 2016

Agent	Rank	Strategy Summary
`Caduceus`	1st	Meta-strategy voting with multi-agent ensemble
`YXAgent`	2nd	Opponent hardness estimation with adaptive response
`MyAgent`	3rd	Rubick-based with Nash equilibrium estimation
`AgentHP2`	-	Multi-phase concession with trend detection
`AgentLight`	-	Lightweight Boulware with minimal overhead
`AgentSmith2016`	-	Updated Smith with Nash-based selection
`Atlas32016`	-	Atlas3 with four-phase refinement
`ClockworkAgent`	-	Precision timing with phase-based strategy
`Farma`	-	Frequency-based opponent model, adaptive concession
`GrandmaAgent`	-	Patient conservative with late concession
`MaxOops`	-	Aggressive with recovery mechanisms
`Ngent`	-	Gentle concession strategy
`ParsCat`	-	Persian cat - Nash product with issue analysis
`Terra`	-	Earth-themed firm-to-flexible three-phase

ANAC 2017

PonPokoAgent won with a surprisingly simple but effective randomized approach.

Agent	Rank	Strategy Summary
`PonPokoAgent`	1st	Randomized threshold patterns; no opponent modeling (simplicity wins!)
`CaduceusDC16`	2nd	Multi-strategy ensemble with weighted voting
`BetaOne`	3rd	Bayesian opponent modeling with three-phase strategy
`AgentF`	-	Linear concession with simple opponent tracking
`AgentKN`	-	Sigmoid concession curve with adaptive threshold
`Farma2017`	-	Exponential concession with Nash-inspired selection
`GeneKing`	-	Genetic algorithm-inspired population bidding
`Gin`	-	Smooth polynomial with future offer estimation
`Group3`	-	Three-phase: hardball, exploration, concession
`Imitator`	-	Tit-for-tat inspired concession mirroring
`MadAgent`	-	Unpredictable "madness" with random elements
`Mamenchis`	-	Patient dinosaur - high patience parameter
`Mosa`	-	Simulated annealing-inspired cooling schedule
`ParsAgent3`	-	Third generation Pars with Nash optimization
`Rubick`	-	Adaptive concession based on opponent behavior
`SimpleAgent2017`	-	Baseline linear strategy for benchmarking
`TaxiBox`	-	"Fare meter" accumulated concession

ANAC 2018

Agent	Rank	Strategy Summary
`AgreeableAgent2018`	1st	Cooperative focus with adaptive compromise
`MengWan`	2nd	Chinese university agent with trend-based modeling
`Seto`	3rd	Threshold-based with multi-phase response
`Agent33`	-	Triple-three structure (3 phases x 3 criteria)
`AgentHerb`	-	Herbal patience - slow and steady
`AgentNP1`	-	Nash product variant with issue weighting
`AteamAgent`	-	Team A's collaborative approach
`ConDAgent`	-	Condition-based with state machine transitions
`ExpRubick`	-	Experimental Rubick with enhanced adaptation
`FullAgent`	-	Complete implementation of multiple strategies
`IQSun2018`	-	Intelligence quotient-based decision making
`PonPokoRampage`	-	Aggressive PonPoko variant
`Shiboy`	-	Japanese style with honor-based acceptance
`Sontag`	-	Literary-inspired balanced strategy
`Yeela`	-	Israeli agent with adaptive concession

ANAC 2019

Agent	Rank	Strategy Summary
`AgentGG`	1st	Issue importance-based bidding with Nash estimation
`KakeSoba`	2nd	Fixed threshold with bid diversification
`SAGA`	3rd	Genetic algorithm with population evolution
`AgentGP`	3rd (Nash)	UCB-style exploration/exploitation balance
`AgentLarry`	-	Simple linear concession baseline
`DandikAgent`	-	Boulware-style slow concession
`EAgent`	-	Exponential decay concession curve
`FSEGA2019`	2nd (Nash)	Adaptive concession based on opponent behavior
`GaravelAgent`	-	Tit-for-tat inspired adaptation
`Gravity`	-	Gravitational (accelerating) concession
`HardDealer`	-	Aggressive hardball tactics
`KAgent`	-	AgentK-inspired adaptive concession
`MINF`	-	Minimal information for robustness
`WinkyAgent`	1st (Nash)	Nash product maximization strategy

Usage Example

from negmas import SAOMechanism, make_issue
from negmas.preferences import LinearAdditiveUtilityFunction as U

from negmas_genius_agents import AgentK, HardHeaded, Atlas3

# Create negotiation scenario
issues = [make_issue("price", (0, 100)), make_issue("quality", (0, 10))]
mechanism = SAOMechanism(issues=issues, n_steps=100)

# Create agents with utility functions
ufun1 = U.random(issues=issues)
ufun2 = U.random(issues=issues)

# Use any ANAC agent
agent1 = AgentK(ufun=ufun1)  # ANAC 2010 winner
agent2 = Atlas3(ufun=ufun2)  # ANAC 2015 winner

# Run negotiation
mechanism.add(agent1)
mechanism.add(agent2)
result = mechanism.run()

print(f"Agreement: {result.agreement}")
print(f"Agent1 utility: {ufun1(result.agreement)}")
print(f"Agent2 utility: {ufun2(result.agreement)}")