boardroom-map/startos/actions/configure-models.ts

import { sdk } from '../sdk'
import { configFile } from '../file-models/config'

const { InputSpec, Value, List } = sdk

const inputSpec = InputSpec.of({
  models: Value.list(
    List.obj(
      {
        name: 'Model Catalog',
        description:
          'The local models this service can serve on your Sparks. Each grader ' +
          'references one of these by its alias. The job runner loads models in ' +
          'waves so you can run a panel across more models than fit in GPU memory ' +
          'at once.',
        default: [],
        minLength: 1,
        maxLength: 16,
      },
      {
        uniqueBy: 'alias',
        displayAs: '{{alias}} → {{hfModel}}',
        spec: InputSpec.of({
          alias: Value.text({
            name: 'Alias',
            description: 'Short name graders use to pick this model (e.g. "qwen-32b").',
            required: true,
            default: null,
            placeholder: 'grader-a',
            patterns: [
              { regex: '^[a-z0-9][a-z0-9-]{0,30}$',
                description: 'Lowercase letters, numbers, dashes (max 31 chars).' },
            ],
          }),
          hfModel: Value.text({
            name: 'Hugging Face Model ID',
            description: 'The model vLLM serves. Must be present in the Spark HF cache for air-gapped mode.',
            required: true,
            default: null,
            placeholder: 'Qwen/Qwen3-32B-FP8',
          }),
          spark: Value.select({
            name: 'Served On',
            description:
              'Which Spark serves this model. Air-gapped grading mode requires the ' +
              'head (primary) Spark; the secondary is used only in local-services mode.',
            default: 'primary',
            values: { primary: 'Primary (head) Spark', secondary: 'Secondary Spark' },
          }),
          port: Value.number({
            name: 'vLLM Port',
            description: 'Host port the vLLM container for this model listens on. Unique per Spark.',
            required: true,
            default: 8001,
            integer: true,
            min: 1,
            max: 65535,
          }),
        }),
      },
    ),
  ),
  gpuMemoryUtilization: Value.text({
    name: 'GPU Memory Utilization',
    description: 'vLLM --gpu-memory-utilization (0–1). Lower it if you co-resident multiple models per Spark.',
    required: true,
    default: '0.85',
  }),
  maxModelLen: Value.number({
    name: 'Max Model Length',
    description: 'vLLM --max-model-len (context window). Deck text is chunked to fit.',
    required: true,
    default: 32768,
    integer: true,
    min: 2048,
  }),
  toolCallParser: Value.text({
    name: 'Tool-Call Parser',
    description:
      'vLLM tool-call parser for the grader\'s read-file tool loop. Match the ' +
      'served model family (Qwen3 → "hermes"). Empty disables native tool-calling.',
    required: false,
    default: 'hermes',
  }),
  maxConcurrentModels: Value.number({
    name: 'Max Co-resident Models (head Spark)',
    description:
      'How many distinct models may load on the head Spark at once. The job runner ' +
      'loads models in waves so it never exceeds this. 1 is safest.',
    required: true,
    default: 1,
    integer: true,
    min: 1,
    max: 8,
  }),
  proxyPort: Value.number({
    name: 'Model Proxy Port',
    description: 'Port for the on-Spark LiteLLM router that exposes every model alias on one endpoint.',
    required: true,
    default: 4000,
    integer: true,
    min: 1,
    max: 65535,
  }),
})

export const configureModels = sdk.Action.withInput(
  'configure-models',

  async ({ effects }) => ({
    name: 'Configure Models',
    description: 'Define the local model catalog served on your Sparks and the serving knobs.',
    warning: null,
    allowedStatuses: 'any',
    group: 'Setup',
    visibility: 'enabled',
  }),

  inputSpec,

  async ({ effects }) => {
    const cfg = await configFile.read().const(effects)
    if (!cfg) return {}
    return {
      models: cfg.models,
      gpuMemoryUtilization: cfg.gpuMemoryUtilization,
      maxModelLen: cfg.maxModelLen,
      toolCallParser: cfg.toolCallParser,
      maxConcurrentModels: cfg.maxConcurrentModels,
      proxyPort: cfg.proxyPort,
    }
  },

  async ({ effects, input }) => {
    await configFile.merge(effects, {
      models: input.models,
      gpuMemoryUtilization: input.gpuMemoryUtilization,
      maxModelLen: input.maxModelLen,
      toolCallParser: input.toolCallParser ?? '',
      maxConcurrentModels: input.maxConcurrentModels,
      proxyPort: input.proxyPort,
    })

    return {
      version: '1',
      title: 'Models Configured',
      message:
        'Saved ' + input.models.length + ' model(s). Make sure each is present in ' +
        'the Spark HF cache for air-gapped runs, then set "Configure Graders".',
      result: {
        type: 'single',
        value: input.models.map((m) => m.alias).join(', '),
        copyable: false,
        qr: false,
        masked: false,
      },
    }
  },
)