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-llm-router: Configuration module
|
|
|
|
|
// Load and parse LLM router configuration from TOML
|
|
|
|
|
|
|
|
|
|
use serde::{Deserialize, Serialize};
|
|
|
|
|
use std::collections::HashMap;
|
|
|
|
|
use std::path::Path;
|
|
|
|
|
use thiserror::Error;
|
|
|
|
|
|
|
|
|
|
#[derive(Debug, Error)]
|
|
|
|
|
pub enum ConfigError {
|
|
|
|
|
#[error("Failed to read config file: {0}")]
|
|
|
|
|
ReadError(#[from] std::io::Error),
|
|
|
|
|
|
|
|
|
|
#[error("Failed to parse TOML: {0}")]
|
|
|
|
|
ParseError(#[from] toml::de::Error),
|
|
|
|
|
|
|
|
|
|
#[error("Invalid configuration: {0}")]
|
|
|
|
|
ValidationError(String),
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[derive(Debug, Clone, Serialize, Deserialize)]
|
|
|
|
|
pub struct LLMRouterConfig {
|
|
|
|
|
pub routing: RoutingConfig,
|
|
|
|
|
pub providers: HashMap<String, ProviderConfig>,
|
|
|
|
|
#[serde(default)]
|
|
|
|
|
pub routing_rules: Vec<RoutingRule>,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[derive(Debug, Clone, Serialize, Deserialize)]
|
|
|
|
|
pub struct RoutingConfig {
|
|
|
|
|
pub default_provider: String,
|
|
|
|
|
#[serde(default = "default_true")]
|
|
|
|
|
pub cost_tracking_enabled: bool,
|
|
|
|
|
#[serde(default = "default_true")]
|
|
|
|
|
pub fallback_enabled: bool,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
fn default_true() -> bool {
|
|
|
|
|
true
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[derive(Debug, Clone, Serialize, Deserialize)]
|
|
|
|
|
pub struct ProviderConfig {
|
|
|
|
|
#[serde(default = "default_true")]
|
|
|
|
|
pub enabled: bool,
|
|
|
|
|
pub api_key: Option<String>,
|
|
|
|
|
pub url: Option<String>,
|
|
|
|
|
pub model: String,
|
|
|
|
|
#[serde(default = "default_max_tokens")]
|
|
|
|
|
pub max_tokens: usize,
|
|
|
|
|
#[serde(default = "default_temperature")]
|
|
|
|
|
pub temperature: f32,
|
|
|
|
|
#[serde(default)]
|
|
|
|
|
pub cost_per_1m_input: f64,
|
|
|
|
|
#[serde(default)]
|
|
|
|
|
pub cost_per_1m_output: f64,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
fn default_max_tokens() -> usize {
|
|
|
|
|
4096
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
fn default_temperature() -> f32 {
|
|
|
|
|
0.7
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[derive(Debug, Clone, Serialize, Deserialize)]
|
|
|
|
|
pub struct RoutingRule {
|
|
|
|
|
pub name: String,
|
|
|
|
|
pub condition: HashMap<String, String>,
|
|
|
|
|
pub provider: String,
|
|
|
|
|
pub model_override: Option<String>,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
impl LLMRouterConfig {
|
|
|
|
|
/// Load configuration from TOML file
|
|
|
|
|
pub fn load<P: AsRef<Path>>(path: P) -> Result<Self, ConfigError> {
|
|
|
|
|
let content = std::fs::read_to_string(path)?;
|
|
|
|
|
let mut config: Self = toml::from_str(&content)?;
|
|
|
|
|
|
|
|
|
|
// Expand environment variables in API keys and URLs
|
|
|
|
|
config.expand_env_vars();
|
|
|
|
|
config.validate()?;
|
|
|
|
|
|
|
|
|
|
Ok(config)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/// Expand environment variables in configuration
|
|
|
|
|
fn expand_env_vars(&mut self) {
|
|
|
|
|
for (_, provider) in self.providers.iter_mut() {
|
|
|
|
|
if let Some(ref api_key) = provider.api_key {
|
|
|
|
|
provider.api_key = Some(expand_env_var(api_key));
|
|
|
|
|
}
|
|
|
|
|
if let Some(ref url) = provider.url {
|
|
|
|
|
provider.url = Some(expand_env_var(url));
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/// Validate configuration
|
|
|
|
|
fn validate(&self) -> Result<(), ConfigError> {
|
|
|
|
|
// Check that default provider exists
|
|
|
|
|
if !self.providers.contains_key(&self.routing.default_provider) {
|
|
|
|
|
return Err(ConfigError::ValidationError(format!(
|
|
|
|
|
"Default provider '{}' not found in providers",
|
|
|
|
|
self.routing.default_provider
|
|
|
|
|
)));
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Check that all routing rules reference valid providers
|
|
|
|
|
for rule in &self.routing_rules {
|
|
|
|
|
if !self.providers.contains_key(&rule.provider) {
|
|
|
|
|
return Err(ConfigError::ValidationError(format!(
|
|
|
|
|
"Routing rule '{}' references unknown provider '{}'",
|
|
|
|
|
rule.name, rule.provider
|
|
|
|
|
)));
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
Ok(())
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/// Get provider configuration by name
|
|
|
|
|
pub fn get_provider(&self, name: &str) -> Option<&ProviderConfig> {
|
|
|
|
|
self.providers.get(name)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/// Find routing rule matching conditions
|
|
|
|
|
pub fn find_rule(&self, conditions: &HashMap<String, String>) -> Option<&RoutingRule> {
|
|
|
|
|
self.routing_rules.iter().find(|rule| {
|
2026-01-11 21:32:56 +00:00
|
|
|
rule.condition
|
|
|
|
|
.iter()
|
|
|
|
|
.all(|(key, value)| conditions.get(key).map(|v| v == value).unwrap_or(false))
|
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
|
|
|
})
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/// Expand environment variables in format ${VAR} or ${VAR:-default}
|
|
|
|
|
fn expand_env_var(input: &str) -> String {
|
|
|
|
|
if !input.starts_with("${") || !input.ends_with('}') {
|
|
|
|
|
return input.to_string();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
let var_part = &input[2..input.len() - 1];
|
|
|
|
|
|
|
|
|
|
// Handle ${VAR:-default} format
|
|
|
|
|
if let Some(pos) = var_part.find(":-") {
|
|
|
|
|
let var_name = &var_part[..pos];
|
|
|
|
|
let default_value = &var_part[pos + 2..];
|
|
|
|
|
std::env::var(var_name).unwrap_or_else(|_| default_value.to_string())
|
|
|
|
|
} else {
|
|
|
|
|
// Handle ${VAR} format
|
|
|
|
|
std::env::var(var_part).unwrap_or_default()
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
|
mod tests {
|
|
|
|
|
use super::*;
|
|
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
|
fn test_expand_env_var() {
|
|
|
|
|
std::env::set_var("TEST_VAR", "test_value");
|
|
|
|
|
assert_eq!(expand_env_var("${TEST_VAR}"), "test_value");
|
|
|
|
|
assert_eq!(expand_env_var("plain_text"), "plain_text");
|
2026-01-11 21:32:56 +00:00
|
|
|
assert_eq!(expand_env_var("${NONEXISTENT:-default}"), "default");
|
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
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
|
fn test_config_validation() {
|
|
|
|
|
let config = LLMRouterConfig {
|
|
|
|
|
routing: RoutingConfig {
|
|
|
|
|
default_provider: "claude".to_string(),
|
|
|
|
|
cost_tracking_enabled: true,
|
|
|
|
|
fallback_enabled: true,
|
|
|
|
|
},
|
|
|
|
|
providers: {
|
|
|
|
|
let mut map = HashMap::new();
|
|
|
|
|
map.insert(
|
|
|
|
|
"claude".to_string(),
|
|
|
|
|
ProviderConfig {
|
|
|
|
|
enabled: true,
|
|
|
|
|
api_key: Some("test".to_string()),
|
|
|
|
|
url: None,
|
|
|
|
|
model: "claude-sonnet-4".to_string(),
|
|
|
|
|
max_tokens: 4096,
|
|
|
|
|
temperature: 0.7,
|
|
|
|
|
cost_per_1m_input: 3.0,
|
|
|
|
|
cost_per_1m_output: 15.0,
|
|
|
|
|
},
|
|
|
|
|
);
|
|
|
|
|
map
|
|
|
|
|
},
|
|
|
|
|
routing_rules: vec![],
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
assert!(config.validate().is_ok());
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
|
fn test_invalid_default_provider() {
|
|
|
|
|
let config = LLMRouterConfig {
|
|
|
|
|
routing: RoutingConfig {
|
|
|
|
|
default_provider: "nonexistent".to_string(),
|
|
|
|
|
cost_tracking_enabled: true,
|
|
|
|
|
fallback_enabled: true,
|
|
|
|
|
},
|
|
|
|
|
providers: HashMap::new(),
|
|
|
|
|
routing_rules: vec![],
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
assert!(config.validate().is_err());
|
|
|
|
|
}
|
|
|
|
|
}
|