plano/crates/common/src/llm_providers.rs
Musa 78dc4edad9
Add first-class ChatGPT subscription provider support (#881)
* Add first-class ChatGPT subscription provider support

* Address PR feedback: move uuid import to top, reuse parsed config in up()

* Add ChatGPT token watchdog for seamless long-lived sessions

* Address PR feedback: error on stream=false for ChatGPT, fix auth file permissions

* Replace ChatGPT watchdog/restart with passthrough_auth

---------

Co-authored-by: Musa Malik <musam@uw.edu>
2026-04-23 15:34:44 -07:00

337 lines
13 KiB
Rust

use crate::configuration::LlmProvider;
use hermesllm::providers::ProviderId;
use std::collections::HashMap;
use std::sync::Arc;
#[derive(Debug)]
pub struct LlmProviders {
providers: HashMap<String, Arc<LlmProvider>>,
default: Option<Arc<LlmProvider>>,
/// Wildcard providers: maps provider prefix to base provider config
/// e.g., "openai" -> LlmProvider for "openai/*"
wildcard_providers: HashMap<String, Arc<LlmProvider>>,
}
impl LlmProviders {
pub fn iter(&self) -> std::collections::hash_map::Iter<'_, String, Arc<LlmProvider>> {
self.providers.iter()
}
pub fn default(&self) -> Option<Arc<LlmProvider>> {
self.default.clone()
}
/// Convert providers to OpenAI Models format for /v1/models endpoint
/// Filters out internal models and duplicate entries (backward compatibility aliases)
pub fn to_models(&self) -> hermesllm::apis::openai::Models {
use hermesllm::apis::openai::{ModelDetail, ModelObject, Models};
let data: Vec<ModelDetail> = self
.providers
.iter()
.filter(|(key, provider)| {
// Exclude internal models
provider.internal != Some(true)
// Only include canonical entries (key matches provider name)
// This avoids duplicates from backward compatibility short names
&& *key == &provider.name
})
.map(|(name, provider)| ModelDetail {
id: name.clone(),
object: Some("model".to_string()),
created: 0,
owned_by: provider.to_provider_id().to_string(),
})
.collect();
Models {
object: ModelObject::List,
data,
}
}
pub fn get(&self, name: &str) -> Option<Arc<LlmProvider>> {
// First try exact match
if let Some(provider) = self.providers.get(name).cloned() {
return Some(provider);
}
// If name contains '/', it could be:
// 1. A full model ID like "openai/gpt-4" that we need to lookup
// 2. A provider/model slug that should match a wildcard provider
if let Some((provider_prefix, model_name)) = name.split_once('/') {
// Try to find the expanded model entry (e.g., "openai/gpt-4")
let full_model_id = format!("{}/{}", provider_prefix, model_name);
if let Some(provider) = self.providers.get(&full_model_id).cloned() {
return Some(provider);
}
// Try to find just the model name (for expanded wildcard entries)
if let Some(provider) = self.providers.get(model_name).cloned() {
return Some(provider);
}
// Fall back to wildcard match (e.g., "openai/*")
if let Some(wildcard_provider) = self.wildcard_providers.get(provider_prefix) {
// Create a new provider with the specific model from the slug
let mut specific_provider = (**wildcard_provider).clone();
specific_provider.model = Some(model_name.to_string());
return Some(Arc::new(specific_provider));
}
}
None
}
}
#[derive(thiserror::Error, Debug)]
pub enum LlmProvidersNewError {
#[error("There must be at least one LLM Provider")]
EmptySource,
#[error("There must be at most one default LLM Provider")]
MoreThanOneDefault,
#[error("\'{0}\' is not a unique name")]
DuplicateName(String),
}
impl TryFrom<Vec<LlmProvider>> for LlmProviders {
type Error = LlmProvidersNewError;
fn try_from(llm_providers_config: Vec<LlmProvider>) -> Result<Self, Self::Error> {
if llm_providers_config.is_empty() {
return Err(LlmProvidersNewError::EmptySource);
}
let mut llm_providers = LlmProviders {
providers: HashMap::new(),
default: None,
wildcard_providers: HashMap::new(),
};
// Track specific (non-wildcard) provider names to detect true duplicates
let mut specific_provider_names = std::collections::HashSet::new();
// Track specific models that should be excluded from wildcard expansion
// Maps provider_prefix -> Set of model names (e.g., "anthropic" -> {"claude-sonnet-4-20250514"})
let mut specific_models_by_provider: HashMap<String, std::collections::HashSet<String>> =
HashMap::new();
// First pass: collect all specific model configurations
for llm_provider in &llm_providers_config {
let is_wildcard = llm_provider
.model
.as_ref()
.map(|m| m == "*" || m.ends_with("/*"))
.unwrap_or(false);
if !is_wildcard {
// Check if this is a provider/model format
if let Some((provider_prefix, model_name)) = llm_provider.name.split_once('/') {
specific_models_by_provider
.entry(provider_prefix.to_string())
.or_default()
.insert(model_name.to_string());
}
}
}
for llm_provider in llm_providers_config {
let llm_provider: Arc<LlmProvider> = Arc::new(llm_provider);
if llm_provider.default.unwrap_or_default() {
match llm_providers.default {
Some(_) => return Err(LlmProvidersNewError::MoreThanOneDefault),
None => llm_providers.default = Some(Arc::clone(&llm_provider)),
}
}
let name = llm_provider.name.clone();
// Check if this is a wildcard provider (model is "*" or ends with "/*")
let is_wildcard = llm_provider
.model
.as_ref()
.map(|m| m == "*" || m.ends_with("/*"))
.unwrap_or(false);
if is_wildcard {
// Extract provider prefix from name
// e.g., "openai/*" -> "openai"
let provider_prefix = name.trim_end_matches("/*").trim_end_matches('*');
// For wildcard providers, we:
// 1. Store the base config in wildcard_providers for runtime matching
// 2. Optionally expand to all known models if available
llm_providers
.wildcard_providers
.insert(provider_prefix.to_string(), Arc::clone(&llm_provider));
// Try to expand wildcard using ProviderId models
if let Ok(provider_id) = ProviderId::try_from(provider_prefix) {
let models = provider_id.models();
// Get the set of specific models to exclude for this provider
let models_to_exclude = specific_models_by_provider
.get(provider_prefix)
.cloned()
.unwrap_or_default();
if !models.is_empty() {
let excluded_count = models_to_exclude.len();
let total_models = models.len();
log::info!(
"Expanding wildcard provider '{}' to {} models{}",
provider_prefix,
total_models - excluded_count,
if excluded_count > 0 {
format!(" (excluding {} specifically configured)", excluded_count)
} else {
String::new()
}
);
// Create a provider entry for each model (except those specifically configured)
for model_name in models {
// Skip this model if it has a specific configuration
if models_to_exclude.contains(&model_name) {
log::debug!(
"Skipping wildcard expansion for '{}/{}' - specific configuration exists",
provider_prefix,
model_name
);
continue;
}
let full_model_id = format!("{}/{}", provider_prefix, model_name);
// Create a new provider with the specific model
let mut expanded_provider = (*llm_provider).clone();
expanded_provider.model = Some(model_name.clone());
expanded_provider.name = full_model_id.clone();
let expanded_rc = Arc::new(expanded_provider);
// Insert with full model ID as key
llm_providers
.providers
.insert(full_model_id.clone(), Arc::clone(&expanded_rc));
// Also insert with just model name for backward compatibility
llm_providers.providers.insert(model_name, expanded_rc);
}
}
} else {
log::warn!(
"Wildcard provider '{}' specified but no models found in registry. \
Will match dynamically at runtime.",
provider_prefix
);
}
} else {
// Non-wildcard provider - specific configuration
// Check for duplicate specific entries (not allowed)
if specific_provider_names.contains(&name) {
return Err(LlmProvidersNewError::DuplicateName(name));
}
specific_provider_names.insert(name.clone());
// This specific configuration takes precedence over any wildcard expansion
// The wildcard expansion already excluded this model (see first pass above)
log::debug!("Processing specific provider configuration: {}", name);
// Insert with the provider name as key
llm_providers
.providers
.insert(name.clone(), Arc::clone(&llm_provider));
// Also add model_id as key for provider lookup
if let Some(model) = llm_provider.model.clone() {
llm_providers.providers.insert(model, llm_provider);
}
}
}
Ok(llm_providers)
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::configuration::LlmProviderType;
fn create_test_provider(name: &str, model: Option<String>) -> LlmProvider {
LlmProvider {
name: name.to_string(),
model,
access_key: None,
endpoint: None,
cluster_name: None,
provider_interface: LlmProviderType::OpenAI,
default: None,
base_url_path_prefix: None,
port: None,
rate_limits: None,
usage: None,
internal: None,
stream: None,
passthrough_auth: None,
headers: None,
}
}
#[test]
fn test_static_provider_lookup() {
// Test 1: Statically defined provider - should be findable by model or provider name
let providers = vec![create_test_provider("my-openai", Some("gpt-4".to_string()))];
let llm_providers = LlmProviders::try_from(providers).unwrap();
// Should find by model name
let result = llm_providers.get("gpt-4");
assert!(result.is_some());
assert_eq!(result.unwrap().name, "my-openai");
// Should also find by provider name
let result = llm_providers.get("my-openai");
assert!(result.is_some());
assert_eq!(result.unwrap().name, "my-openai");
}
#[test]
fn test_wildcard_provider_with_known_model() {
// Test 2: Wildcard provider that expands to OpenAI models
let providers = vec![create_test_provider("openai/*", Some("*".to_string()))];
let llm_providers = LlmProviders::try_from(providers).unwrap();
// Should find via expanded wildcard entry
let result = llm_providers.get("openai/gpt-4");
let provider = result.unwrap();
assert_eq!(provider.name, "openai/gpt-4");
assert_eq!(provider.model.as_ref().unwrap(), "gpt-4");
// Should also be able to find by just model name (from expansion)
let result = llm_providers.get("gpt-4");
assert_eq!(result.unwrap().model.as_ref().unwrap(), "gpt-4");
}
#[test]
fn test_custom_wildcard_provider_with_full_slug() {
// Test 3: Custom wildcard provider with full slug offered
let providers = vec![create_test_provider(
"custom-provider/*",
Some("*".to_string()),
)];
let llm_providers = LlmProviders::try_from(providers).unwrap();
// Should match via wildcard fallback and extract model name from slug
let result = llm_providers.get("custom-provider/custom-model");
let provider = result.unwrap();
assert_eq!(provider.model.as_ref().unwrap(), "custom-model");
// Wildcard should be stored
assert!(llm_providers
.wildcard_providers
.contains_key("custom-provider"));
}
}