Component
core/ai-model/litellm
LiteLLM running in-cluster as a self-hosted, OpenAI-compatible model proxy. The published alias in values.model is the name the proxy serves under, and it forwards each request to the provider-qualified upstream model in values.upstream_model, so a consumer reaches one stable in-cluster endpoint and name while the target behind it changes independently. The proxy holds the upstream credential attached to its provider_key slot and presents it on a caller's behalf. Exports its host, port, endpoint, model and protocol, the contract every other model component publishes. Because it holds that credential, its Recipes forward only the request fields the proxy declares for the route, plus any the Release admits through infra.admitted_request_fields, and refuse the rest with an error naming the field. They answer only the model routes, the model listing, the health checks and key management, and refuse every other path, so its token-count, pass-through and other administrative routes, its console and model paths carrying the model name are not available through any. On Kubernetes, reach for helm/litellm, which synthesizes the proxy, where the build must fetch nothing, and for helm/litellm-chart, BerriAI's chart, where its registry is reachable; for local development, the compose Recipe. With `values.metrics_enabled` true each Recipe has the proxy serve its Prometheus metrics, its path allowlist forwarding a GET of /metrics alone to them from port 9090, and publishes that address as `metrics_endpoint`; false, that key is empty. The port checks no caller, and the metrics name the hashed keys, key aliases and user emails the proxy attributes spend to.
Provenienza
- Digest del contenuto
- sha256:c9e4f18a552bb4bfccc17282661ddd96f87aca6478c7f7a0dddcdda6a35a93b4
- Pubblicato
- 2026-11-10T10:09:16Z
- Identità di firma
- https://token.actions.githubusercontent.com/https://github.com/aggrega-ai/aggrega-hub/.github/workflows/catalog-publish-sign.yml@refs/heads/main
Firma
- Ufficialeapplication/vnd.dev.sigstore.bundle.v0.3+json
https://token.actions.githubusercontent.com/https://github.com/aggrega-ai/aggrega-hub/.github/workflows/catalog-publish-sign.yml@refs/heads/main
sha256:d6b38e6c33e02b29347f6d8447b229b897aedadd5cec8c00546dc926d8187ea1
Riferimento a questa versione
oci://hub.aggrega.ai/component/core/ai-model/litellm:1.0.0Parametri dello sviluppatore
Valori che lo sviluppatore imposta nell’Aggregate.
Estende BaseComponents#/schemas/BaseModelParameter
| Nome | Tipo | Predefinito | Obbligatorio | Descrizione |
|---|---|---|---|---|
| upstream_model | string | — | sì | The provider-qualified upstream model the proxy forwards to, in the proxy's own provider/model notation (for example openai/gpt-4o or anthropic/claude-sonnet-4). A deterministic pointer string the runtime resolves per request; the compiler never contacts the provider. |
| provider_key_variable | string | OPENAI_API_KEY | no | Name of the environment variable under which the upstream provider's credential is delivered, and the variable the proxy's configuration names in its place. The correct name follows the provider chosen in upstream_model — the default suits an OpenAI-notation target; an Anthropic target uses ANTHROPIC_API_KEY, and so on per the provider's own convention. |
| metrics_enabled | boolean | false | no | Has the proxy serve its own Prometheus metrics through its path allowlist on port 9090, where a GET of /metrics alone is forwarded, and publishes their address as `metrics_endpoint`; while false that key is empty. The port checks no caller, and the metrics name the hashed keys, key aliases and user emails the proxy attributes spend to, so anything that reaches the port reads them. |
Dimensionamento dell’operatore
Dimensionamento dell’infrastruttura che l’operatore regola per ogni ambiente.
Estende LiteLLMGuardSchemas#/schemas/RequestGuardInfra
| Nome | Tipo | Predefinito | Obbligatorio | Descrizione |
|---|---|---|---|---|
| port | integer (min 1024, max 65535) | 4000 | no | Server listen port for the proxy API. Most operators should omit this and let the Recipe assign the default (4000). |
| external_base_url | string | — | no | Externally reachable origin of the proxy, as assembled at the ingress tier — scheme, host and, where it is not the scheme's default, port, with no path; for example https://llm.example.com. The proxy composes from this the sign-in callback address it hands an identity provider, the address of its admin console it sends a browser to after sign-in, and the proxy address it writes into every alert it sends. The product documents this setting as an origin only, so the Recipes halt the build on an address carrying a path. Omitted, the proxy composes each of those addresses from the request that arrives, which is correct for a proxy nobody signs in to from outside the cluster. |
| engine_version | string | v1.103.1 | no | LiteLLM proxy image tag. Selects the image tag of the proxy container on every Recipe. |
| source_version | string | — | no | Version of BerriAI's litellm-helm chart the chart-backed Recipe pulls, defaulting to the version that Recipe was checked against. The workload-synthesis and compose Recipes pull no chart and read nothing here. |
| compute | object | — | no | — |
| overrides | object (open) | — | no | Provider-specific overrides. Accepts any nested structure. Each Recipe documents which keys it reads. |
Contratto di output
Valori che questo componente pubblica verso i consumatori attraverso la topologia uses.
| Nome | Tipo | Predefinito | Obbligatorio | Descrizione |
|---|---|---|---|---|
| host | string | — | sì | In-cluster DNS name of the Service fronting the proxy. A consumer that composes its own address dials this name; one that does not reads `endpoint` instead. |
| port | integer | — | sì | TCP port that Service accepts proxy requests on. It follows the port an operator set in the Release, so a consumer reads it rather than assuming a default. |
| endpoint | string | — | sì | Full base URL of the OpenAI-compatible API the proxy serves, with the host, the port and the API path already composed. A consumer hands it to a client library as its base URL. |
| model | string | — | sì | Name a request must ask the proxy for. The proxy maps it onto whichever upstream provider it is configured to forward to, so a consumer names this rather than an upstream model. |
| protocol | string | — | no | URL scheme to reach `host` and `port` over, for a consumer composing its own address rather than reading `endpoint`. |
| metrics_endpoint | string | — | no | The full address Prometheus scrapes this component's metrics at — scheme, host, port and path, `http://<host>:9090/metrics`, the host being the name of the Service the Recipe serves them on. Empty while values.metrics_enabled is false. |
Socket uses
I socket di dipendenza attraverso cui questo componente si connette. La destinazione di ogni connessione collegata attraverso un socket deve soddisfare il suo selettore di etichette, e ogni connessione può portare un aggancio negli slot che il socket dichiara. Uno slot obbligatorio deve essere riempito da ogni connessione che l’Aggregate o la Release collega attraverso il socket.
Nessun socket uses dichiarato.
Slot di aggancio
Punti di associazione con nome in cui una Feature si aggancia a questo componente. Il selettore di uno slot limita quale Feature può riempirlo e la sua lista inject indica come può essere consegnato il materiale della Feature. Uno slot obbligatorio deve essere associato su ogni istanza, da un aggancio sull’istanza o a un ambito superiore, e uno slot sensibile è dichiarato per gestire materiale segreto. Una Feature può anche agganciarsi con un nome che nessuno slot dichiara.
| Slot | Selettore | Obbligatorio | Iniezione | Sensibile | Descrizione |
|---|---|---|---|---|---|
| api_key | feature.aggrega.ai/kind=secret | no | env | sì | Optional administrative key guarding the proxy's own API, which LiteLLM calls its master key. When attached, the Recipe delivers it as LITELLM_MASTER_KEY, the environment variable the proxy reads at startup, so callers and the management API must present it. Unlike a server that checks one key, the proxy also accepts the keys it issues through its key-management API and keeps in its bookkeeping database: with database_url attached a caller may present such a key instead, and with none the proxy accepts this key alone. Unattached, the proxy starts with no administrative key and treats every caller as an unprivileged internal user, which suits a development Release and not a shared one. |
| provider_key | feature.aggrega.ai/kind=secret | no | env | sì | Optional credential the proxy presents to the upstream provider it routes to, delivered as the environment variable named in values.provider_key_variable. Unattached, the proxy runs and serves its API but holds no upstream credential, so a routed request fails at the provider; attach one for any Release that must actually reach a model. |
| database_url | feature.aggrega.ai/kind=secret | no | env | sì | Optional connection string for the proxy's own bookkeeping database, which backs key management, spend tracking and request logs. When attached, the Recipe delivers it as DATABASE_URL, the environment variable the proxy reads at startup — on helm referenced from the secret that holds it, so the proxy's Deployment carries the reference and never the string; on compose as the value its credential delivers under that name, which a Compose variable reference keeps out of the file. The compiler never connects to the database. Unattached, the proxy runs stateless — it serves and routes normally, and the features that need persistence are simply inactive. |
Elenco consentito di auto-iniezione
env
Relazioni
Archi tipizzati da e verso altre entità pubblicate. Un riferimento senza una destinazione pubblicata viene mostrato come arco interrotto anziché come link.
Usa
- $refestende lo schema (values)
BaseComponents#/schemas/BaseModelParameter
- $refestende lo schema (infra)
LiteLLMGuardSchemas#/schemas/RequestGuardInfra
Usato da
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «model»
usa il socket «target_model»
usa il socket «extraction_model»
usa il socket «embedding_model»
usa il socket «models»
usa il socket «models»
usa il socket «models»
usa il socket «models»
usa il socket «models»
usa il socket «models»
usa il socket «models»
usa il socket «upstream»
usa il socket «models»
usa il socket «forward_target»
usa il socket «scrape_target»
usa il socket «store»
usa il socket «content_safety»
si applica a
si applica a
si applica a
si applica a
si applica a
si applica a
si applica a
implementa
implementa
implementa