264 lines
7.0 KiB
Plaintext
Raw Permalink Normal View History

feat: Phase 5.3 - Multi-Agent Learning Infrastructure Implement intelligent agent learning from Knowledge Graph execution history with per-task-type expertise tracking, recency bias, and learning curves. ## Phase 5.3 Implementation ### Learning Infrastructure (✅ Complete) - LearningProfileService with per-task-type expertise metrics - TaskTypeExpertise model tracking success_rate, confidence, learning curves - Recency bias weighting: recent 7 days weighted 3x higher (exponential decay) - Confidence scoring prevents overfitting: min(1.0, executions / 20) - Learning curves computed from daily execution windows ### Agent Scoring Service (✅ Complete) - Unified AgentScore combining SwarmCoordinator + learning profiles - Scoring formula: 0.3*base + 0.5*expertise + 0.2*confidence - Rank agents by combined score for intelligent assignment - Support for recency-biased scoring (recent_success_rate) - Methods: rank_agents, select_best, rank_agents_with_recency ### KG Integration (✅ Complete) - KGPersistence::get_executions_for_task_type() - query by agent + task type - KGPersistence::get_agent_executions() - all executions for agent - Coordinator::load_learning_profile_from_kg() - core KG→Learning integration - Coordinator::load_all_learning_profiles() - batch load for multiple agents - Convert PersistedExecution → ExecutionData for learning calculations ### Agent Assignment Integration (✅ Complete) - AgentCoordinator uses learning profiles for task assignment - extract_task_type() infers task type from title/description - assign_task() scores candidates using AgentScoringService - Fallback to load-based selection if no learning data available - Learning profiles stored in coordinator.learning_profiles RwLock ### Profile Adapter Enhancements (✅ Complete) - create_learning_profile() - initialize empty profiles - add_task_type_expertise() - set task-type expertise - update_profile_with_learning() - update swarm profiles from learning ## Files Modified ### vapora-knowledge-graph/src/persistence.rs (+30 lines) - get_executions_for_task_type(agent_id, task_type, limit) - get_agent_executions(agent_id, limit) ### vapora-agents/src/coordinator.rs (+100 lines) - load_learning_profile_from_kg() - core KG integration method - load_all_learning_profiles() - batch loading for agents - assign_task() already uses learning-based scoring via AgentScoringService ### Existing Complete Implementation - vapora-knowledge-graph/src/learning.rs - calculation functions - vapora-agents/src/learning_profile.rs - data structures and expertise - vapora-agents/src/scoring.rs - unified scoring service - vapora-agents/src/profile_adapter.rs - adapter methods ## Tests Passing - learning_profile: 7 tests ✅ - scoring: 5 tests ✅ - profile_adapter: 6 tests ✅ - coordinator: learning-specific tests ✅ ## Data Flow 1. Task arrives → AgentCoordinator::assign_task() 2. Extract task_type from description 3. Query KG for task-type executions (load_learning_profile_from_kg) 4. Calculate expertise with recency bias 5. Score candidates (SwarmCoordinator + learning) 6. Assign to top-scored agent 7. Execution result → KG → Update learning profiles ## Key Design Decisions ✅ Recency bias: 7-day half-life with 3x weight for recent performance ✅ Confidence scoring: min(1.0, total_executions / 20) prevents overfitting ✅ Hierarchical scoring: 30% base load, 50% expertise, 20% confidence ✅ KG query limit: 100 recent executions per task-type for performance ✅ Async loading: load_learning_profile_from_kg supports concurrent loads ## Next: Phase 5.4 - Cost Optimization Ready to implement budget enforcement and cost-aware provider selection.
2026-01-11 13:03:53 +00:00
"""
VAPORA Agent Pools Configuration
Defines scaling policies and configurations for each of the 12 agent roles
"""
import k.api.all as k
# ===== AGENT POOL DEFINITIONS =====
agent_pools = {
"architect": {
role = "Architect"
description = "System design and architecture decisions"
llm = "Claude Opus"
parallelizable = false # Initiator role, must run sequentially
min_replicas = 2
max_replicas = 5
target_cpu = 70
cpu_request = "2000m"
memory_request = "4Gi"
cpu_limit = "4000m"
memory_limit = "8Gi"
}
"developer": {
role = "Developer"
description = "Code implementation"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 5
max_replicas = 20
target_cpu = 60
cpu_request = "2000m"
memory_request = "3Gi"
cpu_limit = "4000m"
memory_limit = "6Gi"
}
"code_reviewer": {
role = "CodeReviewer"
description = "Code quality and review"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 3
max_replicas = 10
target_cpu = 65
cpu_request = "1500m"
memory_request = "2Gi"
cpu_limit = "3000m"
memory_limit = "4Gi"
}
"tester": {
role = "Tester"
description = "Test writing and validation"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 3
max_replicas = 10
target_cpu = 70
cpu_request = "2000m"
memory_request = "3Gi"
cpu_limit = "4000m"
memory_limit = "6Gi"
}
"documenter": {
role = "Documenter"
description = "Documentation and guides"
llm = "GPT-4"
parallelizable = true
min_replicas = 2
max_replicas = 8
target_cpu = 50
cpu_request = "1000m"
memory_request = "2Gi"
cpu_limit = "2000m"
memory_limit = "4Gi"
}
"marketer": {
role = "Marketer"
description = "Marketing content and campaigns"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 1
max_replicas = 5
target_cpu = 40
cpu_request = "1000m"
memory_request = "2Gi"
cpu_limit = "2000m"
memory_limit = "4Gi"
}
"presenter": {
role = "Presenter"
description = "Presentations and slides"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 1
max_replicas = 3
target_cpu = 50
cpu_request = "1000m"
memory_request = "2Gi"
cpu_limit = "2000m"
memory_limit = "4Gi"
}
"devops": {
role = "DevOps"
description = "CI/CD and deployment"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 2
max_replicas = 8
target_cpu = 60
cpu_request = "1500m"
memory_request = "2Gi"
cpu_limit = "3000m"
memory_limit = "4Gi"
}
"monitor": {
role = "Monitor"
description = "Health checking and alerting"
llm = "Gemini Flash"
parallelizable = true # Real-time monitoring
min_replicas = 2
max_replicas = 5
target_cpu = 30
cpu_request = "1000m"
memory_request = "1Gi"
cpu_limit = "2000m"
memory_limit = "2Gi"
}
"security": {
role = "Security"
description = "Security audit and verification"
llm = "Claude Opus"
parallelizable = false # Can block pipeline
min_replicas = 2
max_replicas = 5
target_cpu = 70
cpu_request = "2000m"
memory_request = "4Gi"
cpu_limit = "4000m"
memory_limit = "8Gi"
}
"project_manager": {
role = "ProjectManager"
description = "Project tracking and roadmap"
llm = "Claude Sonnet"
parallelizable = true
min_replicas = 1
max_replicas = 3
target_cpu = 40
cpu_request = "1000m"
memory_request = "2Gi"
cpu_limit = "2000m"
memory_limit = "4Gi"
}
"decision_maker": {
role = "DecisionMaker"
description = "Conflict resolution and decisions"
llm = "Claude Opus"
parallelizable = false # On-demand decision making
min_replicas = 1
max_replicas = 3
target_cpu = 70
cpu_request = "2000m"
memory_request = "4Gi"
cpu_limit = "4000m"
memory_limit = "8Gi"
}
}
# ===== HORIZONTAL POD AUTOSCALERS =====
hpas = [
{
name = "vapora-agents-developer-hpa"
target_deployment = "vapora-agents"
agent_role = "developer"
min_replicas = 5
max_replicas = 20
target_cpu_utilization = 60
metrics = [
{
type = "Resource"
resource = {
name = "cpu"
target = {type = "Utilization", averageUtilization = 60}
}
},
{
type = "Pods"
pods = {
metric = {name = "agent_queue_depth"}
target = {type = "AverageValue", averageValue = "50"}
}
}
]
}
{
name = "vapora-agents-reviewer-hpa"
target_deployment = "vapora-agents"
agent_role = "code_reviewer"
min_replicas = 3
max_replicas = 10
target_cpu_utilization = 65
}
{
name = "vapora-agents-monitor-hpa"
target_deployment = "vapora-agents"
agent_role = "monitor"
min_replicas = 2
max_replicas = 5
target_cpu_utilization = 30
}
]
# ===== POD DISRUPTION BUDGETS =====
pod_disruption_budgets = [
{
name = "vapora-agents-pdb"
selector = {matchLabels = {"app": "vapora-agents"}}
minAvailable = 2 # Always keep at least 2 agents running
}
{
name = "surrealdb-pdb"
selector = {matchLabels = {"app": "surrealdb"}}
minAvailable = 2 # Database must always have 2+ replicas
}
]
# ===== NETWORK POLICIES FOR AGENTS =====
network_policies = [
{
name = "allow-agent-to-nats"
ingress = [{
from = [{podSelector = {matchLabels = {"app": "vapora-agents"}}}]
ports = [{protocol = "TCP", port = 4222}]
}]
egress = [{
to = [{podSelector = {matchLabels = {"app": "nats"}}}]
ports = [{protocol = "TCP", port = 4222}]
}]
}
{
name = "allow-agent-to-database"
ingress = [{
from = [{podSelector = {matchLabels = {"app": "vapora-agents"}}}]
ports = [{protocol = "TCP", port = 8000}]
}]
egress = [{
to = [{podSelector = {matchLabels = {"app": "surrealdb"}}}]
ports = [{protocol = "TCP", port = 8000}]
}]
}
]
# ===== OUTPUT =====
output = {
agent_pools = agent_pools
hpas = hpas
pdbs = pod_disruption_budgets
network_policies = network_policies
}