moonpi 0.4.2

package/src/synthetic-models.ts

@@ -0,0 +1,438 @@
+import type { ProviderModelConfig } from "@mariozechner/pi-coding-agent";
+import { existsSync, readFileSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+import { getAgentDir } from "@mariozechner/pi-coding-agent";
+
+const SYNTHETIC_REASONING_EFFORT_MAP = {
+  minimal: "low",
+  low: "low",
+  medium: "medium",
+  high: "high",
+  xhigh: "high",
+} as const;
+
+export const SYNTHETIC_MODELS_FALLBACK: ProviderModelConfig[] = [
+  {
+    id: "hf:zai-org/GLM-4.7",
+    name: "zai-org/GLM-4.7",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 0.45,
+      output: 2.19,
+      cacheRead: 0.45,
+      cacheWrite: 0,
+    },
+    contextWindow: 202752,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:zai-org/GLM-5",
+    name: "zai-org/GLM-5",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 1,
+      output: 3,
+      cacheRead: 1,
+      cacheWrite: 0,
+    },
+    contextWindow: 196608,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:zai-org/GLM-5.1",
+    name: "zai-org/GLM-5.1",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+      supportsDeveloperRole: false,
+    },
+    input: ["text"],
+    cost: {
+      input: 1,
+      output: 3,
+      cacheRead: 1,
+      cacheWrite: 0,
+    },
+    contextWindow: 196608,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:zai-org/GLM-4.7-Flash",
+    name: "zai-org/GLM-4.7-Flash",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 0.1,
+      output: 0.5,
+      cacheRead: 0.1,
+      cacheWrite: 0,
+    },
+    contextWindow: 196608,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:meta-llama/Llama-3.3-70B-Instruct",
+    name: "meta-llama/Llama-3.3-70B-Instruct",
+    reasoning: true,
+    input: ["text"],
+    cost: {
+      input: 0.88,
+      output: 0.88,
+      cacheRead: 0.88,
+      cacheWrite: 0,
+    },
+    contextWindow: 131072,
+    maxTokens: 32768,
+  },
+  {
+    id: "hf:deepseek-ai/DeepSeek-R1-0528",
+    name: "deepseek-ai/DeepSeek-R1-0528",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 3,
+      output: 8,
+      cacheRead: 3,
+      cacheWrite: 0,
+    },
+    contextWindow: 131072,
+    maxTokens: 128000,
+  },
+  {
+    id: "hf:deepseek-ai/DeepSeek-V3.2",
+    name: "deepseek-ai/DeepSeek-V3.2",
+    reasoning: true,
+    input: ["text"],
+    cost: {
+      input: 0.56,
+      output: 1.68,
+      cacheRead: 0.56,
+      cacheWrite: 0,
+    },
+    contextWindow: 162816,
+    maxTokens: 8000,
+  },
+  {
+    id: "hf:openai/gpt-oss-120b",
+    name: "openai/gpt-oss-120b",
+    reasoning: true,
+    input: ["text"],
+    cost: {
+      input: 0.1,
+      output: 0.1,
+      cacheRead: 0.1,
+      cacheWrite: 0,
+    },
+    contextWindow: 131072,
+    maxTokens: 32768,
+  },
+  {
+    id: "hf:Qwen/Qwen3-Coder-480B-A35B-Instruct",
+    name: "Qwen/Qwen3-Coder-480B-A35B-Instruct",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 2,
+      output: 2,
+      cacheRead: 2,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:moonshotai/Kimi-K2.5",
+    name: "moonshotai/Kimi-K2.5",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text", "image"],
+    cost: {
+      input: 0.45,
+      output: 3.4,
+      cacheRead: 0.45,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:nvidia/Kimi-K2.5-NVFP4",
+    name: "nvidia/Kimi-K2.5-NVFP4",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text", "image"],
+    cost: {
+      input: 0.45,
+      output: 3.4,
+      cacheRead: 0.45,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:deepseek-ai/DeepSeek-V3",
+    name: "deepseek-ai/DeepSeek-V3",
+    reasoning: true,
+    input: ["text"],
+    cost: {
+      input: 1.25,
+      output: 1.25,
+      cacheRead: 1.25,
+      cacheWrite: 0,
+    },
+    contextWindow: 131072,
+    maxTokens: 128000,
+  },
+  {
+    id: "hf:Qwen/Qwen3-235B-A22B-Thinking-2507",
+    name: "Qwen/Qwen3-235B-A22B-Thinking-2507",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 0.65,
+      output: 3,
+      cacheRead: 0.65,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 32000,
+  },
+  {
+    id: "hf:Qwen/Qwen3.5-397B-A17B",
+    name: "Qwen/Qwen3.5-397B-A17B",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text", "image"],
+    cost: {
+      input: 0.6,
+      output: 3.6,
+      cacheRead: 0.6,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 65536,
+  },
+  {
+    id: "hf:MiniMaxAI/MiniMax-M2.5",
+    name: "MiniMaxAI/MiniMax-M2.5",
+    reasoning: true,
+    input: ["text"],
+    cost: {
+      input: 0.4,
+      output: 2,
+      cacheRead: 0.4,
+      cacheWrite: 0,
+    },
+    contextWindow: 191488,
+    maxTokens: 65536,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+      maxTokensField: "max_completion_tokens",
+    },
+  },
+  {
+    id: "hf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
+    name: "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
+    reasoning: true,
+    compat: {
+      supportsReasoningEffort: true,
+      reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+    },
+    input: ["text"],
+    cost: {
+      input: 0.3,
+      output: 1,
+      cacheRead: 0.3,
+      cacheWrite: 0,
+    },
+    contextWindow: 262144,
+    maxTokens: 65536,
+  },
+];
+
+interface SyntheticModelResponse {
+  id: string;
+  name: string;
+  input_modalities: string[];
+  output_modalities: string[];
+  context_length: number;
+  max_output_length: number;
+  pricing: {
+    prompt: string;
+    completion: string;
+    image: string;
+    request: string;
+    input_cache_reads: string;
+    input_cache_writes: string;
+  };
+  supported_features: string[];
+  quantization?: string;
+}
+
+const MODEL_COMPAT_OVERRIDES: Record<string, Partial<NonNullable<ProviderModelConfig["compat"]>>> = {
+  "hf:zai-org/GLM-5.1": {
+    supportsDeveloperRole: false,
+  },
+  "hf:MiniMaxAI/MiniMax-M2.5": {
+    maxTokensField: "max_completion_tokens",
+  },
+};
+
+function parsePricingValue(value: string): number {
+  if (!value || value === "0") return 0;
+  const num = parseFloat(value.replace(/^\$/, ""));
+  if (Number.isNaN(num)) return 0;
+  return num * 1_000_000;
+}
+
+// =========================================================================
+// Model cache – persists live API models to disk for fast startup
+// =========================================================================
+
+const CACHE_FILE_NAME = "synthetic-models-cache.json";
+
+/** Maximum age in milliseconds before cached models are considered stale (12h). */
+const CACHE_STALE_MS = 12 * 60 * 60 * 1000;
+
+interface SyntheticModelsCache {
+  /** ISO timestamp of when the cache was last written. */
+  updatedAt: string;
+  /** Model configs from the live Synthetic API. */
+  models: ProviderModelConfig[];
+}
+
+/** Resolve the cache file path (always inside the agent config dir). */
+export function getCacheFilePath(): string {
+  return join(getAgentDir(), CACHE_FILE_NAME);
+}
+
+/**
+ * Read cached synthetic models from disk.
+ * Returns `undefined` when the cache is missing, corrupt, or older than `maxAgeMs`.
+ */
+export function readCachedModels(maxAgeMs: number = CACHE_STALE_MS): ProviderModelConfig[] | undefined {
+  const filePath = getCacheFilePath();
+  if (!existsSync(filePath)) return undefined;
+
+  try {
+    const raw = readFileSync(filePath, "utf-8");
+    const cache = JSON.parse(raw) as SyntheticModelsCache;
+    if (!Array.isArray(cache.models) || cache.models.length === 0) return undefined;
+
+    // Check staleness
+    const updated = new Date(cache.updatedAt).getTime();
+    if (Number.isNaN(updated) || Date.now() - updated > maxAgeMs) return undefined;
+
+    return cache.models;
+  } catch {
+    return undefined;
+  }
+}
+
+/** Write synthetic models to the cache file. Failures are silently ignored. */
+export function writeCachedModels(models: ProviderModelConfig[]): void {
+  const filePath = getCacheFilePath();
+  const cache: SyntheticModelsCache = {
+    updatedAt: new Date().toISOString(),
+    models,
+  };
+  try {
+    writeFileSync(filePath, JSON.stringify(cache), "utf-8");
+  } catch {
+    // Cache writes are best-effort; don't block startup on permission errors, etc.
+  }
+}
+
+/**
+ * Merge cached models with the hardcoded fallback list.
+ *
+ * Cached models win on id collisions (they're fresher from the API).
+ * Models present only in the fallback list are kept so nothing disappears
+ * if the cache is partial.
+ */
+export function mergeWithFallback(cached: ProviderModelConfig[], fallback: ProviderModelConfig[]): ProviderModelConfig[] {
+  const byId = new Map<string, ProviderModelConfig>();
+  for (const m of fallback) byId.set(m.id, m);
+  for (const m of cached) byId.set(m.id, m); // cached wins on collision
+  return [...byId.values()];
+}
+
+// =========================================================================
+// API response parsing
+// =========================================================================
+
+export function parseSyntheticModels(data: SyntheticModelResponse[]): ProviderModelConfig[] {
+  return data.map((model) => {
+    const hasReasoning = model.supported_features?.includes("reasoning") ?? false;
+    const input: ("text" | "image")[] = [];
+    if (model.input_modalities?.includes("text")) input.push("text");
+    if (model.input_modalities?.includes("image")) input.push("image");
+    if (input.length === 0) input.push("text");
+
+    const config: ProviderModelConfig = {
+      id: model.id,
+      name: model.name,
+      reasoning: hasReasoning,
+      input,
+      cost: {
+        input: parsePricingValue(model.pricing?.prompt ?? "0"),
+        output: parsePricingValue(model.pricing?.completion ?? "0"),
+        cacheRead: parsePricingValue(model.pricing?.input_cache_reads ?? "0"),
+        cacheWrite: parsePricingValue(model.pricing?.input_cache_writes ?? "0"),
+      },
+      contextWindow: model.context_length ?? 128000,
+      maxTokens: model.max_output_length ?? 4096,
+    };
+
+    const overrides = MODEL_COMPAT_OVERRIDES[model.id];
+    if (hasReasoning) {
+      config.compat = {
+        supportsReasoningEffort: true,
+        reasoningEffortMap: SYNTHETIC_REASONING_EFFORT_MAP,
+        ...overrides,
+      };
+    } else if (overrides) {
+      config.compat = { ...overrides };
+    }
+
+    return config;
+  });
+}