systemprompt_models/services/providers/
mod.rs

1//! Provider registry: the single source of upstream connectivity.
2//!
3//! [`ProviderRegistry`] is the `providers:` list of the services tree
4//! (`services/ai/providers.yaml` by convention, merged across includes). It is
5//! implementation configuration shipped with the deployment, not a
6//! per-environment profile section: the same catalog boots every environment
7//! and only the credentials it names differ. Each [`ProviderEntry`] declares
8//! one upstream exactly once — its
9//! [`WireProtocol`], endpoint, credential ([`SecretName`]), extra headers, and
10//! the model catalog it serves. The two policy layers reference entries by
11//! [`ProviderId`] and never re-declare connectivity: the gateway policy
12//! (`services.gateway`) routes external model names to a provider, and the AI
13//! policy (`services/ai/config.yaml`) selects an agent default and per-provider
14//! overrides.
15//!
16//! Validation here is the authority for connectivity: unique provider names,
17//! SSRF-guarded endpoints, and globally-unique model ids/aliases. The gateway
18//! and AI layers validate only their references *into* this registry.
19//!
20//! Copyright (c) systemprompt.io — Business Source License 1.1.
21//! See <https://systemprompt.io> for licensing details.
22
23mod discovery_report;
24mod error;
25mod protocol;
26mod rate_card;
27mod surface;
28
29use std::collections::{HashMap, HashSet};
30
31use serde::{Deserialize, Serialize};
32use systemprompt_identifiers::{ModelId, ProviderId, SecretName};
33
34use crate::services::ai::{ModelCapabilities, ModelGovernance, ModelLimits, ModelPricing};
35
36pub use discovery_report::DiscoveryReport;
37pub use error::{ProviderRegistryError, ProviderRegistryResult};
38pub use protocol::WireProtocol;
39pub use rate_card::{
40    DocumentedLaunchStage, RETIREMENT_NOTICE_DAYS, VertexRateCard, VertexRateCardEntry,
41};
42pub use surface::ApiSurface;
43
44const DEFAULT_CATALOG_YAML: &str = include_str!("default_catalog.yaml");
45
46#[derive(Deserialize)]
47struct DefaultCatalogFile {
48    providers: Vec<ProviderEntry>,
49}
50
51#[derive(Debug, Clone, Serialize, Deserialize, schemars::JsonSchema)]
52#[serde(deny_unknown_fields)]
53pub struct ProviderModel {
54    pub id: ModelId,
55
56    #[serde(default, skip_serializing_if = "Vec::is_empty")]
57    pub aliases: Vec<ModelId>,
58
59    #[serde(default, skip_serializing_if = "Option::is_none")]
60    pub upstream_model: Option<String>,
61
62    #[serde(default)]
63    pub pricing: ModelPricing,
64
65    #[serde(default)]
66    pub capabilities: ModelCapabilities,
67
68    #[serde(default)]
69    pub limits: ModelLimits,
70
71    #[serde(default, skip_serializing_if = "Option::is_none")]
72    pub governance: Option<ModelGovernance>,
73}
74
75impl ProviderModel {
76    #[must_use]
77    pub fn matches(&self, requested: &str) -> bool {
78        self.id.as_str() == requested || self.aliases.iter().any(|a| a.as_str() == requested)
79    }
80
81    #[must_use]
82    pub fn effective_upstream_model<'a>(&'a self, requested: &'a str) -> &'a str {
83        self.upstream_model.as_deref().unwrap_or(requested)
84    }
85}
86
87#[derive(Debug, Clone, Serialize, Deserialize, schemars::JsonSchema)]
88#[serde(deny_unknown_fields)]
89pub struct ProviderEntry {
90    pub name: ProviderId,
91
92    pub wire: WireProtocol,
93
94    pub surface: ApiSurface,
95
96    pub endpoint: String,
97
98    pub api_key_secret: SecretName,
99
100    #[serde(default, skip_serializing_if = "HashMap::is_empty")]
101    pub extra_headers: HashMap<String, String>,
102
103    #[serde(default, skip_serializing_if = "Vec::is_empty")]
104    pub models: Vec<ProviderModel>,
105
106    #[serde(default)]
107    pub governance: ModelGovernance,
108}
109
110impl ProviderEntry {
111    #[must_use]
112    pub fn find_model(&self, requested: &str) -> Option<&ProviderModel> {
113        self.models.iter().find(|m| m.matches(requested))
114    }
115
116    #[must_use]
117    pub fn upstream_model_for<'a>(
118        &'a self,
119        route_override: Option<&'a str>,
120        requested: &'a str,
121    ) -> &'a str {
122        route_override.unwrap_or_else(|| {
123            self.find_model(requested)
124                .map_or(requested, |model| model.effective_upstream_model(requested))
125        })
126    }
127
128    #[must_use]
129    pub fn effective_governance(&self, requested: &str) -> ModelGovernance {
130        self.find_model(requested)
131            .and_then(|m| m.governance)
132            .unwrap_or(self.governance)
133    }
134}
135
136#[derive(Debug, Clone, Default, Serialize, Deserialize, schemars::JsonSchema)]
137#[serde(transparent)]
138pub struct ProviderRegistry {
139    pub providers: Vec<ProviderEntry>,
140}
141
142impl ProviderRegistry {
143    pub fn default_seed() -> ProviderRegistryResult<Self> {
144        let file: DefaultCatalogFile = serde_yaml::from_str(DEFAULT_CATALOG_YAML)
145            .map_err(|e| ProviderRegistryError::InvalidDefaultCatalog(e.to_string()))?;
146        Ok(Self {
147            providers: file.providers,
148        })
149    }
150
151    #[must_use]
152    pub fn find_provider(&self, name: &str) -> Option<&ProviderEntry> {
153        self.providers.iter().find(|p| p.name.as_str() == name)
154    }
155
156    #[must_use]
157    pub fn contains_model(&self, requested: &str) -> bool {
158        self.providers
159            .iter()
160            .any(|p| p.find_model(requested).is_some())
161    }
162
163    pub fn advertised_providers(&self) -> impl Iterator<Item = &ProviderEntry> {
164        self.providers
165            .iter()
166            .filter(|entry| entry.surface.is_advertised())
167    }
168
169    #[must_use]
170    pub fn advertised_model_ids(&self, surfaces: &[ApiSurface]) -> Vec<String> {
171        self.advertised_providers()
172            .filter(|entry| surfaces.is_empty() || surfaces.contains(&entry.surface))
173            .flat_map(|entry| {
174                entry.models.iter().flat_map(|m| {
175                    std::iter::once(m.id.as_str().to_owned())
176                        .chain(m.aliases.iter().map(|a| a.as_str().to_owned()))
177                })
178            })
179            .collect()
180    }
181
182    pub fn validate(&self) -> ProviderRegistryResult<()> {
183        let trusted = crate::net::trusted_http_hosts_from_env();
184        let mut seen_providers: HashSet<&str> = HashSet::with_capacity(self.providers.len());
185        let mut seen_models: HashSet<&str> = HashSet::new();
186
187        for provider in &self.providers {
188            if !seen_providers.insert(provider.name.as_str()) {
189                return Err(ProviderRegistryError::DuplicateProvider {
190                    name: provider.name.as_str().to_owned(),
191                });
192            }
193            if provider.endpoint.is_empty() {
194                return Err(ProviderRegistryError::EmptyEndpoint {
195                    name: provider.name.as_str().to_owned(),
196                });
197            }
198            crate::net::validate_outbound_url_with_trust(&provider.endpoint, &trusted).map_err(
199                |e| ProviderRegistryError::BlockedEndpoint {
200                    provider: provider.name.as_str().to_owned(),
201                    endpoint: provider.endpoint.clone(),
202                    reason: e.to_string(),
203                },
204            )?;
205            if names_a_project_literally(&provider.endpoint) {
206                return Err(ProviderRegistryError::LiteralProjectInEndpoint {
207                    provider: provider.name.as_str().to_owned(),
208                    endpoint: provider.endpoint.clone(),
209                });
210            }
211
212            for model in &provider.models {
213                if model.id.as_str().is_empty() {
214                    return Err(ProviderRegistryError::EmptyModelId {
215                        id: provider.name.as_str().to_owned(),
216                    });
217                }
218                if !seen_models.insert(model.id.as_str()) {
219                    return Err(ProviderRegistryError::DuplicateModel {
220                        id: model.id.as_str().to_owned(),
221                    });
222                }
223                for alias in &model.aliases {
224                    if !seen_models.insert(alias.as_str()) {
225                        return Err(ProviderRegistryError::DuplicateModel {
226                            id: alias.as_str().to_owned(),
227                        });
228                    }
229                }
230            }
231        }
232        Ok(())
233    }
234}
235
236pub const PROJECT_PLACEHOLDER: &str = "{project}";
237
238// Why: the placeholder names live here, beside the registry that validates
239// endpoints, so that the credential layer that fills them and the validator
240// that polices them can never disagree about their spelling. `{region}` has
241// no filler today — no shipped credential type carries a region — and an
242// endpoint using it is refused until one does, which is the intended shape:
243// a coordinate is served by the credential or not at all.
244pub const REGION_PLACEHOLDER: &str = "{region}";
245
246// Why: a Google Cloud project id is a tenant identifier, and Vertex reports it
247// verbatim in every IAM error it returns, which the gateway relays to the
248// caller. A catalog that names one literally therefore ships that id to every
249// installation of the image and every client that trips a 403. The id lives in
250// exactly one place, the service-account key, and the endpoint says
251// `{project}` instead.
252#[must_use]
253pub fn names_a_project_literally(endpoint: &str) -> bool {
254    let Ok(url) = url::Url::parse(endpoint) else {
255        return false;
256    };
257    let on_vertex = url.host_str().is_some_and(|host| {
258        host.eq_ignore_ascii_case("aiplatform.googleapis.com")
259            || host
260                .to_ascii_lowercase()
261                .ends_with("-aiplatform.googleapis.com")
262    });
263    if !on_vertex {
264        return false;
265    }
266    let mut segments = url.path_segments().into_iter().flatten();
267    while let Some(segment) = segments.next() {
268        if segment == "projects" {
269            return segments
270                .next()
271                .is_some_and(|id| id != "%7Bproject%7D" && id != PROJECT_PLACEHOLDER);
272        }
273    }
274    false
275}
systemprompt_models/services/providers/mod.rs

systemprompt_models/services/providers/
mod.rs
