@velum-labs/routekit-config

packages/config publishes @velum-labs/routekit-config: reusable RouteKit router-config loading, provider validation, live-model selection, and atomic writes.

Resolution

loadRouterConfig() loads the canonical ~/.config/routekit/router.yaml, or one explicit complete document when an embedding supplies configPath. Configuration rejects inline credentials; providers obtain credential and optional base-URL environment-variable names from @velum-labs/routekit-registry.

import {
  assertModelsAvailable,
  configuredProviderIds,
  loadRouterConfig,
  resolveModelId,
  writeRouterConfig
} from "@velum-labs/routekit-config";

const loaded = loadRouterConfig();
const providers = configuredProviderIds(loaded.config);
const liveModels = ["openai/gpt-5.5"];
const model = resolveModelId(loaded.config, liveModels);
assertModelsAvailable([model], liveModels);

configuredProviderIds() preserves declaration order. missingModelIds() and assertModelsAvailable() compare configured model requirements with a discovered live catalog. resolveModelId() validates an explicit namespaced model, or selects the configured default/first live model only when none was requested. Explicit unknown models are rejected; they never fall through to the default. writeRouterConfig() and updateRouterConfig() validate before atomically writing mode-0600 YAML.

loadRepositoryRoutingProfile({ repositoryRoot, name }) resolves one read-only committed manifest from .routekit/routing/profiles/<name>.yaml. It verifies the filename/name binding, repository-contained activation and evidence paths, and the activation basis reference without writing global router or daemon state.

Docs

• Configuration reference: ../../docs/configuration.md
• Maintainer reference: ../../docs/typescript-reference.md