A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.
Python artificial-intelligencedagdata-sciencedata-visualizationdataflowdeveloper-toolsmachine-learningnotebookspipelinepythonreactivesqlweb-app
/* Copyright 2026 Marimo. All rights reserved. */
import {
KNOWN_PROVIDERS,
type KnownProviderId,
type ProviderId,
} from "@/core/ai/ids/ids";
import { getKnownModelMaps } from "@/core/ai/model-registry";
import type { AiConfig, UserConfig } from "@/core/config/config-schema";
type CredentialChecker = (ai: AiConfig | undefined) => boolean;
/**
* Credential checkers for each known provider.
*/
const CREDENTIAL_CHECKERS: Record<KnownProviderId, CredentialChecker> = {
openai: (ai) => Boolean(ai?.open_ai?.api_key),
github: (ai) => Boolean(ai?.github?.api_key),
anthropic: (ai) => Boolean(ai?.anthropic?.api_key),
google: (ai) => Boolean(ai?.google?.api_key),
openrouter: (ai) => Boolean(ai?.openrouter?.api_key),
azure: (ai) => Boolean(ai?.azure?.api_key && ai?.azure?.base_url),
wandb: (ai) => Boolean(ai?.wandb?.api_key),
"opencode-go": (ai) => Boolean(ai?.opencode_go?.api_key),
bedrock: (ai) => Boolean(ai?.bedrock?.region_name),
ollama: (ai) => Boolean(ai?.ollama?.base_url),
// No user-configurable credentials in the UI.
deepseek: () => false,
// Hosted marimo models (marimo.app / molab) route through open_ai_compatible.
marimo: (ai) => Boolean(ai?.open_ai_compatible?.base_url),
};
/**
* Known providers with credentials, then custom providers with a base URL.
*/
export function listConfiguredProviders(
ai: AiConfig | undefined,
): ProviderId[] {
const knownProviders = KNOWN_PROVIDERS.filter((provider) =>
CREDENTIAL_CHECKERS[provider](ai),
);
const customProviders = Object.entries(ai?.custom_providers ?? {})
.filter(([, providerConfig]) => Boolean(providerConfig?.base_url))
.map(([name]) => name);
return [...knownProviders, ...customProviders];
}
/**
* Keep provider groups whose provider has credentials configured.
*/
export function filterToConfiguredProviders<T>(
entries: readonly [ProviderId, T][],
ai: AiConfig | undefined,
): [ProviderId, T][] {
const configured = new Set<string>(listConfiguredProviders(ai));
return entries.filter(([provider]) => configured.has(provider));
}
/**
* When provider setup is locked, only show models for configured providers.
*/
export function listModelsForAiSettings<T>(
entries: readonly [ProviderId, T][],
ai: AiConfig | undefined,
): [ProviderId, T][] {
if (ai?.allow_provider_config !== false) {
return [...entries];
}
return filterToConfiguredProviders(entries, ai);
}
export function getRecommendedModel(
config: UserConfig["ai"],
): string | undefined {
const provider = listConfiguredProviders(config)[0];
if (!provider) {
return undefined;
}
return getKnownModelMaps().defaultModelByProvider.get(provider);
}
export interface AutoPopulateResult {
chatModel: string | undefined;
editModel: string | undefined;
}
/**
* Determines which models to auto-populate based on configured credentials.
* Returns the recommended model for chat/edit if credentials are configured but models aren't set.
*
* @param values - The full form values
*/
export function autoPopulateModels(
values: UserConfig["ai"],
): AutoPopulateResult {
const result: AutoPopulateResult = {
chatModel: undefined,
editModel: undefined,
};
const needsChatModel = !values?.models?.chat_model;
const needsEditModel = !values?.models?.edit_model;
if (!needsChatModel && !needsEditModel) {
return result;
}
const recommendedModel = getRecommendedModel(values);
if (!recommendedModel) {
return result;
}
if (needsChatModel) {
result.chatModel = recommendedModel;
}
if (needsEditModel) {
result.editModel = recommendedModel;
}
return result;
}