@golba98/codexa 1.0.1

package/src/core/providerRuntime/contextMetadata.ts

@@ -0,0 +1,409 @@
+import type { ModelSpec } from "../models/modelSpecs.js";
+import type { ProviderId, ProviderWorkspaceOverride } from "../providerLauncher/types.js";
+
+export type ContextLengthSource = "api" | "cli" | "config" | "known-registry" | "lmstudio-api" | "unknown";
+export type ContextConfidence = "verified" | "configured" | "known" | "estimated" | "unknown";
+
+export interface ModelContextMetadata {
+  providerId: ProviderId;
+  modelId: string;
+  contextLength: number | null;
+  source: ContextLengthSource;
+  confidence: ContextConfidence;
+  raw?: unknown;
+  rawField?: string;
+  error?: string;
+}
+
+export interface ResolveModelContextLengthOptions {
+  providerId: ProviderId;
+  modelId: string;
+  providerConfig?: ProviderWorkspaceOverride | null;
+  rawMetadata?: unknown;
+}
+
+interface KnownContextRegistryEntry {
+  contextLength: number;
+  sourceUrl: string;
+  note: string;
+  estimated?: boolean;
+}
+
+const KNOWN_CONTEXT_REGISTRY: Record<string, KnownContextRegistryEntry> = {
+  // Exact documented Gemini API model IDs only.
+  // Source: https://ai.google.dev/gemini-api/docs/models
+  "google:gemini-2.5-pro": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini API documented input token limit for this exact model ID.",
+  },
+  "google:gemini-2.5-flash": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini API documented input token limit for this exact model ID.",
+  },
+  "google:gemini-2.5-flash-lite": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini API documented input token limit for this exact model ID.",
+  },
+  // Gemini 3 preview model IDs — verified route IDs from GEMINI_VERIFIED_MODEL_IDS.
+  // Source: https://ai.google.dev/gemini-api/docs/models
+  "google:gemini-3-flash-preview": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini 3 Flash Preview — 1M token input limit.",
+  },
+  "google:gemini-3.1-pro-preview": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini 3.1 Pro Preview — 1M token input limit.",
+  },
+  "google:gemini-3.1-flash-lite-preview": {
+    contextLength: 1_048_576,
+    sourceUrl: "https://ai.google.dev/gemini-api/docs/models",
+    note: "Gemini 3.1 Flash Lite Preview — 1M token input limit.",
+  },
+  // Exact documented Anthropic model IDs only. Provider aliases such as
+  // "haiku" remain unknown unless configured or discovered by CLI metadata.
+  // Source: https://docs.anthropic.com/en/docs/about-claude/models
+  "anthropic:claude-opus-4-7": {
+    contextLength: 200_000,
+    sourceUrl: "https://docs.anthropic.com/en/docs/about-claude/models",
+    note: "Anthropic documented context window for this exact model ID.",
+  },
+  "anthropic:claude-sonnet-4-6": {
+    contextLength: 200_000,
+    sourceUrl: "https://docs.anthropic.com/en/docs/about-claude/models",
+    note: "Anthropic documented context window for this exact model ID.",
+  },
+  "anthropic:claude-haiku-4-5": {
+    contextLength: 200_000,
+    sourceUrl: "https://docs.anthropic.com/en/docs/about-claude/models",
+    note: "Anthropic documented context window for this exact model ID.",
+  },
+  // OpenAI/Codex model IDs as used by the Codex CLI ("openai" provider).
+  // All values are estimated — Codex CLI effective context may differ from the
+  // OpenAI API. Keys are lowercase-normalised (resolveFromKnownRegistry normalises
+  // before lookup so "GPT-5 Codex" → "gpt-5-codex" matches here).
+  // Source: https://platform.openai.com/docs/models
+  "openai:gpt-5.5": {
+    contextLength: 1_048_576,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex CLI effective context unverified.",
+  },
+  "openai:gpt-5.5-mini": {
+    contextLength: 200_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated.",
+  },
+  "openai:gpt-5.4": {
+    contextLength: 400_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated.",
+  },
+  "openai:gpt-5.4-mini": {
+    contextLength: 200_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated.",
+  },
+  "openai:gpt-5.3-codex": {
+    contextLength: 400_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex variant.",
+  },
+  "openai:gpt-5.2": {
+    contextLength: 200_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated.",
+  },
+  "openai:gpt-5.1-codex": {
+    contextLength: 400_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex variant.",
+  },
+  "openai:gpt-5.1-codex-mini": {
+    contextLength: 200_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex mini variant.",
+  },
+  "openai:gpt-5-codex": {
+    contextLength: 400_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex variant.",
+  },
+  "openai:gpt-5-codex-mini": {
+    contextLength: 200_000,
+    estimated: true,
+    sourceUrl: "https://platform.openai.com/docs/models",
+    note: "Estimated — Codex mini variant.",
+  },
+};
+
+// Short alias IDs used by ANTHROPIC_FALLBACK_MODELS → canonical versioned IDs in the registry.
+const ANTHROPIC_MODEL_ALIAS_MAP: Record<string, string> = {
+  opus:   "claude-opus-4-7",
+  sonnet: "claude-sonnet-4-6",
+  haiku:  "claude-haiku-4-5",
+};
+
+// Normalise OpenAI/Codex model IDs to lowercase-dashed form before registry lookup.
+// Handles display-label variants like "GPT-5 Codex" → "gpt-5-codex".
+function normalizeOpenAIModelId(modelId: string): string {
+  return modelId.toLowerCase().replace(/\s+/g, "-");
+}
+
+const CONTEXT_FIELD_CANDIDATES = [
+  "loaded_context_length",
+  "context_length",
+  "contextLength",
+  "max_context_length",
+  "maxContextLength",
+  "n_ctx",
+  "ctx_len",
+  "max_tokens",
+  "max_input_tokens",
+] as const;
+
+const NESTED_METADATA_KEYS = [
+  "model_info",
+  "modelInfo",
+  "metadata",
+  "details",
+  "config",
+  "raw",
+] as const;
+
+const contextCache = new Map<string, ModelContextMetadata>();
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+
+function validContextLength(value: unknown): number | null {
+  if (typeof value !== "number" || !Number.isInteger(value) || value <= 0) {
+    return null;
+  }
+  return value;
+}
+
+function cacheKey(options: ResolveModelContextLengthOptions): string {
+  return JSON.stringify({
+    providerId: options.providerId,
+    modelId: options.modelId,
+    models: options.providerConfig?.models ?? null,
+  });
+}
+
+export function clearModelContextMetadataCache(): void {
+  contextCache.clear();
+}
+
+export function formatContextLength(value: number | null): string {
+  return value === null ? "Unknown" : value.toLocaleString("en-US");
+}
+
+export function formatContextCompact(n: number): string {
+  if (n >= 1_000_000) return `${(n / 1_000_000).toFixed(1)}M`;
+  if (n >= 100_000) return `${Math.round(n / 1_000)}K`;
+  if (n >= 1_000) return `${(n / 1_000).toFixed(1)}k`;
+  return `${n}`;
+}
+
+export function formatContextMeter(
+  usedTokens: number | null | undefined,
+  contextLimit: number | null,
+): string {
+  if (contextLimit === null || contextLimit === undefined) return "Unknown";
+  const used = usedTokens ?? 0;
+  const pct = contextLimit > 0
+    ? Math.min(100, Math.floor((used / contextLimit) * 100))
+    : 0;
+  return `${used.toLocaleString("en-US")} / ${contextLimit.toLocaleString("en-US")} · ${pct}%`;
+}
+
+export function contextMetadataToModelSpec(metadata: ModelContextMetadata): ModelSpec {
+  if (metadata.contextLength === null) {
+    return {
+      status: "unknown",
+      contextWindow: null,
+      maxOutputTokens: null,
+      sourceUrl: "",
+      verifiedAt: null,
+      error: metadata.error ?? null,
+    };
+  }
+
+  return {
+    status: "verified",
+    contextWindow: metadata.contextLength,
+    maxOutputTokens: metadata.contextLength,
+    sourceUrl: metadata.source,
+    verifiedAt: Date.now(),
+    isEstimated: metadata.confidence === "estimated",
+  };
+}
+
+function unknownMetadata(
+  providerId: ProviderId,
+  modelId: string,
+  error = "Context length metadata is unavailable for this provider/model.",
+): ModelContextMetadata {
+  return {
+    providerId,
+    modelId,
+    contextLength: null,
+    source: "unknown",
+    confidence: "unknown",
+    error,
+  };
+}
+
+function findContextField(raw: unknown): { value: number; field: string } | null {
+  if (!isRecord(raw)) return null;
+
+  for (const field of CONTEXT_FIELD_CANDIDATES) {
+    const value = validContextLength(raw[field]);
+    if (value !== null) return { value, field };
+  }
+
+  for (const nestedKey of NESTED_METADATA_KEYS) {
+    const nested = raw[nestedKey];
+    if (!isRecord(nested)) continue;
+    for (const field of CONTEXT_FIELD_CANDIDATES) {
+      const value = validContextLength(nested[field]);
+      if (value !== null) return { value, field: `${nestedKey}.${field}` };
+    }
+  }
+
+  return null;
+}
+
+function resolveFromRawMetadata(providerId: ProviderId, modelId: string, raw: unknown): ModelContextMetadata | null {
+  const found = findContextField(raw);
+  if (!found) return null;
+  const source: ContextLengthSource =
+    found.field === "loaded_context_length" || found.field.endsWith(".loaded_context_length")
+      ? "lmstudio-api"
+      : providerId === "local" ? "api" : "cli";
+  return {
+    providerId,
+    modelId,
+    contextLength: found.value,
+    source,
+    confidence: "verified",
+    raw,
+    rawField: found.field,
+  };
+}
+
+function resolveFromConfig(
+  providerId: ProviderId,
+  modelId: string,
+  providerConfig?: ProviderWorkspaceOverride | null,
+): ModelContextMetadata | null {
+  const value = providerConfig?.models?.[modelId]?.contextLength;
+  if (value === undefined) return null;
+  const contextLength = validContextLength(value);
+  if (contextLength === null) {
+    return {
+      providerId,
+      modelId,
+      contextLength: null,
+      source: "unknown",
+      confidence: "unknown",
+      error: `Invalid configured contextLength for ${modelId}. Expected a positive integer.`,
+    };
+  }
+  return {
+    providerId,
+    modelId,
+    contextLength,
+    source: "config",
+    confidence: "configured",
+  };
+}
+
+function resolveFromKnownRegistry(providerId: ProviderId, modelId: string): ModelContextMetadata | null {
+  const lookupId = providerId === "anthropic"
+    ? (ANTHROPIC_MODEL_ALIAS_MAP[modelId] ?? modelId)
+    : providerId === "openai"
+      ? normalizeOpenAIModelId(modelId)
+      : modelId;
+  const entry = KNOWN_CONTEXT_REGISTRY[`${providerId}:${lookupId}`];
+  if (!entry) return null;
+  return {
+    providerId,
+    modelId,
+    contextLength: entry.contextLength,
+    source: "known-registry",
+    confidence: entry.estimated ? "estimated" : "known",
+    raw: {
+      sourceUrl: entry.sourceUrl,
+      note: entry.note,
+    },
+  };
+}
+
+export async function resolveModelContextLength(
+  options: ResolveModelContextLengthOptions,
+): Promise<ModelContextMetadata> {
+  const key = cacheKey(options);
+  const cached = contextCache.get(key);
+  if (cached && !(options.rawMetadata !== undefined && cached.source === "unknown")) {
+    return cached;
+  }
+
+  const rawMetadata = options.rawMetadata ?? null;
+  const resolved =
+    resolveFromRawMetadata(options.providerId, options.modelId, rawMetadata)
+    ?? resolveFromConfig(options.providerId, options.modelId, options.providerConfig)
+    ?? resolveFromKnownRegistry(options.providerId, options.modelId)
+    ?? unknownMetadata(
+      options.providerId,
+      options.modelId,
+      options.providerId === "local"
+        ? "/v1/models did not include context length metadata."
+        : "Provider metadata did not include context length.",
+    );
+
+  contextCache.set(key, resolved);
+  return resolved;
+}
+
+export function resolveModelContextLengthCached(
+  options: ResolveModelContextLengthOptions,
+): ModelContextMetadata {
+  const key = cacheKey(options);
+  const cached = contextCache.get(key);
+  if (cached && !(options.rawMetadata !== undefined && cached.source === "unknown")) return cached;
+
+  const raw = resolveFromRawMetadata(options.providerId, options.modelId, options.rawMetadata);
+  if (raw) {
+    contextCache.set(key, raw);
+    return raw;
+  }
+
+  const config = resolveFromConfig(options.providerId, options.modelId, options.providerConfig);
+  if (config) {
+    contextCache.set(key, config);
+    return config;
+  }
+
+  const registry = resolveFromKnownRegistry(options.providerId, options.modelId);
+  if (registry) {
+    contextCache.set(key, registry);
+    return registry;
+  }
+
+  return unknownMetadata(options.providerId, options.modelId);
+}