import type { SubagentBackend, SubagentInput, SubagentResult } from './types.js'
import { logger } from '../logger.js'

const OLLAMA_URL = process.env.OLLAMA_URL ?? 'http://localhost:11434'
const DEFAULT_TIMEOUT_MS = 900_000

export class OllamaBackend implements SubagentBackend {
  readonly name: string
  private readonly model: string

  constructor(model: string) {
    this.model = model
    this.name = `ollama:${model}`
  }

  async run(input: SubagentInput): Promise<SubagentResult> {
    const start = Date.now()
    const result: SubagentResult = { backend: this.name, text: '', durationMs: 0 }
    const timeoutMs = (input.timeoutMs ?? Number.parseInt(process.env.OLLAMA_TIMEOUT_MS ?? '', 10)) || DEFAULT_TIMEOUT_MS

    const ac = new AbortController()
    const timer = setTimeout(() => ac.abort(), timeoutMs)
    timer.unref()

    try {
      const res = await fetch(`${OLLAMA_URL}/api/chat`, {
        method: 'POST',
        signal: ac.signal,
        headers: { 'content-type': 'application/json' },
        body: JSON.stringify({
          model: this.model,
          messages: [{ role: 'user', content: input.prompt }],
          stream: false,
          // Force full GPU offload; Ollama still falls back to CPU if the
          // model genuinely doesn't fit. Prevents silent CPU spills on
          // borderline-sized models.
          options: { num_gpu: -1 },
        }),
      })
      if (!res.ok) {
        result.error = `ollama HTTP ${res.status}`
      } else {
        const body = (await res.json()) as {
          message?: { content?: string }
          prompt_eval_count?: number
          eval_count?: number
          total_duration?: number
        }
        result.text = body.message?.content?.trim() ?? ''
        result.inputTokens = body.prompt_eval_count
        result.outputTokens = body.eval_count
      }
    } catch (err) {
      result.error = err instanceof Error ? err.message : String(err)
      logger.warn({ err: result.error, model: this.model }, 'ollama backend failed')
    } finally {
      clearTimeout(timer)
    }
    result.durationMs = Date.now() - start
    if (!result.text && !result.error) result.error = 'ollama: empty response'
    return result
  }
}