Merge PR #578: feat — add configurable context length to model metadata (by @hijak)

open-sse/config/providerRegistry.ts

@@ -16,6 +16,8 @@ export interface RegistryModel {
   toolCalling?: boolean;
   targetFormat?: string;
   unsupportedParams?: readonly string[];
+  /** Maximum context window in tokens */
+  contextLength?: number;
 }
 
 // Reasoning models reject temperature, top_p, penalties, logprobs, n.
@@ -65,6 +67,8 @@ export interface RegistryEntry {
   chatPath?: string;
   clientVersion?: string;
   passthroughModels?: boolean;
+  /** Default context window for all models in this provider (can be overridden per-model) */
+  defaultContextLength?: number;
 }
 
 interface LegacyProvider {
@@ -137,6 +141,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     urlSuffix: "?beta=true",
     authType: "oauth",
     authHeader: "x-api-key",
+    defaultContextLength: 200000,
     headers: {
       "Anthropic-Version": "2023-06-01",
       "Anthropic-Beta":
@@ -180,6 +185,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     },
     authType: "apikey",
     authHeader: "x-goog-api-key",
+    defaultContextLength: 1000000,
     oauth: {
       clientIdEnv: "GEMINI_OAUTH_CLIENT_ID",
       clientIdDefault: "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com",
@@ -216,6 +222,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     },
     authType: "oauth",
     authHeader: "bearer",
+    defaultContextLength: 1000000,
     oauth: {
       clientIdEnv: "GEMINI_CLI_OAUTH_CLIENT_ID",
       clientIdDefault: "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com",
@@ -247,6 +254,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     baseUrl: "https://chatgpt.com/backend-api/codex/responses",
     authType: "oauth",
     authHeader: "bearer",
+    defaultContextLength: 400000,
     headers: {
       Version: "0.92.0",
       "Openai-Beta": "responses=experimental",
@@ -399,6 +407,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     responsesBaseUrl: "https://api.githubcopilot.com/responses",
     authType: "oauth",
     authHeader: "bearer",
+    defaultContextLength: 128000,
     headers: {
       "copilot-integration-id": "vscode-chat",
       "editor-version": "vscode/1.110.0",
@@ -447,6 +456,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     baseUrl: "https://codewhisperer.us-east-1.amazonaws.com/generateAssistantResponse",
     authType: "oauth",
     authHeader: "bearer",
+    defaultContextLength: 200000,
     headers: {
       "Content-Type": "application/json",
       Accept: "application/vnd.amazon.eventstream",
@@ -473,6 +483,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     chatPath: "/aiserver.v1.ChatService/StreamUnifiedChatWithTools",
     authType: "oauth",
     authHeader: "bearer",
+    defaultContextLength: 200000,
     headers: {
       "connect-accept-encoding": "gzip",
       "connect-protocol-version": "1",
@@ -501,6 +512,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     baseUrl: "https://api.openai.com/v1/chat/completions",
     authType: "apikey",
     authHeader: "bearer",
+    defaultContextLength: 128000,
     models: [
       { id: "gpt-4o", name: "GPT-4o" },
       { id: "gpt-4o-mini", name: "GPT-4o Mini" },
@@ -522,6 +534,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     urlSuffix: "?beta=true",
     authType: "apikey",
     authHeader: "x-api-key",
+    defaultContextLength: 200000,
     headers: {
       "Anthropic-Version": "2023-06-01",
     },
@@ -548,6 +561,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     authType: "apikey",
     authHeader: "Authorization",
     authPrefix: "Bearer",
+    defaultContextLength: 200000,
     models: [
       { id: "glm-5", name: "GLM-5" },
       { id: "kimi-k2.5", name: "Kimi K2.5" },
@@ -565,6 +579,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     authType: "apikey",
     authHeader: "Authorization",
     authPrefix: "Bearer",
+    defaultContextLength: 200000,
     models: [
       { id: "minimax-m2.5-free", name: "MiniMax M2.5 Free" },
       { id: "big-pickle", name: "Big Pickle" },
@@ -580,6 +595,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     baseUrl: "https://openrouter.ai/api/v1/chat/completions",
     authType: "apikey",
     authHeader: "bearer",
+    defaultContextLength: 128000,
     headers: {
       "HTTP-Referer": "https://endpoint-proxy.local",
       "X-Title": "Endpoint Proxy",
@@ -593,6 +609,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     format: "claude",
     executor: "default",
     baseUrl: "https://api.z.ai/api/anthropic/v1/messages",
+    defaultContextLength: 200000,
     urlSuffix: "?beta=true",
     authType: "apikey",
     authHeader: "x-api-key",

open-sse/services/contextManager.ts

@@ -5,14 +5,34 @@
  * 3 layers: trim tool messages, compress thinking, aggressive purification.
  */
 
-// Default token limits per provider (rough estimates based on model context windows)
-const DEFAULT_LIMITS = {
+import { REGISTRY } from "../config/providerRegistry.ts";
+
+// Default token limits per provider (fallbacks when not in registry)
+const DEFAULT_LIMITS: Record<string, number> = {
   claude: 200000,
   openai: 128000,
   gemini: 1000000,
+  codex: 400000,
   default: 128000,
 };
 
+// Environment variable overrides (highest priority)
+function getEnvOverride(provider: string): number | null {
+  const envKey = `CONTEXT_LENGTH_${provider.toUpperCase().replace(/[^A-Z0-9]/g, "_")}`;
+  const envValue = process.env[envKey];
+  if (envValue) {
+    const parsed = parseInt(envValue, 10);
+    if (!isNaN(parsed) && parsed > 0) return parsed;
+  }
+  // Global override
+  const globalValue = process.env.CONTEXT_LENGTH_DEFAULT;
+  if (globalValue) {
+    const parsed = parseInt(globalValue, 10);
+    if (!isNaN(parsed) && parsed > 0) return parsed;
+  }
+  return null;
+}
+
 // Rough chars-per-token ratio for quick estimation
 const CHARS_PER_TOKEN = 4;
 
@@ -27,9 +47,20 @@ export function estimateTokens(text) {
 
 /**
  * Get token limit for a provider/model combination
+ * Priority: Env override > Registry defaultContextLength > DEFAULT_LIMITS
  */
 export function getTokenLimit(provider, model = null) {
-  // Check if model has a known limit
+  // 1. Check environment variable override first
+  const envOverride = getEnvOverride(provider);
+  if (envOverride) return envOverride;
+
+  // 2. Check registry for provider default
+  const registryEntry = REGISTRY[provider];
+  if (registryEntry?.defaultContextLength) {
+    return registryEntry.defaultContextLength;
+  }
+
+  // 3. Check if model name hints at a known limit
   if (model) {
     const lower = model.toLowerCase();
     if (lower.includes("claude")) return DEFAULT_LIMITS.claude;
@@ -38,10 +69,13 @@ export function getTokenLimit(provider, model = null) {
       lower.includes("gpt") ||
       lower.includes("o1") ||
       lower.includes("o3") ||
-      lower.includes("o4")
+      lower.includes("o4") ||
+      lower.includes("codex")
     )
-      return DEFAULT_LIMITS.openai;
+      return DEFAULT_LIMITS.codex;
   }
+
+  // 4. Fallback to DEFAULT_LIMITS or default
   return DEFAULT_LIMITS[provider] || DEFAULT_LIMITS.default;
 }
 

src/app/api/v1/models/catalog.ts

@@ -16,6 +16,7 @@ import { getAllAudioModels } from "@omniroute/open-sse/config/audioRegistry.ts";
 import { getAllModerationModels } from "@omniroute/open-sse/config/moderationRegistry.ts";
 import { getAllVideoModels, getVideoProvider } from "@omniroute/open-sse/config/videoRegistry.ts";
 import { getAllMusicModels, getMusicProvider } from "@omniroute/open-sse/config/musicRegistry.ts";
+import { REGISTRY } from "@omniroute/open-sse/config/providerRegistry.ts";
 
 const FALLBACK_ALIAS_TO_PROVIDER = {
   ag: "antigravity",
@@ -190,6 +191,7 @@ export async function getUnifiedModelsResponse(
         permission: [],
         root: combo.name,
         parent: null,
+        ...(combo.context_length ? { context_length: combo.context_length } : {}),
       });
     }
 
@@ -206,8 +208,15 @@ export async function getUnifiedModelsResponse(
         continue;
       }
 
+      // Get default context length from registry (provider-level default)
+      const registryEntry = REGISTRY[alias] || REGISTRY[canonicalProviderId];
+      const defaultContextLength = registryEntry?.defaultContextLength;
+
       for (const model of providerModels) {
         const aliasId = `${alias}/${model.id}`;
+        // Model-level context length overrides provider default
+        const contextLength = model.contextLength || defaultContextLength;
+
         models.push({
           id: aliasId,
           object: "model",
@@ -216,6 +225,7 @@ export async function getUnifiedModelsResponse(
           permission: [],
           root: model.id,
           parent: null,
+          ...(contextLength ? { context_length: contextLength } : {}),
         });
 
         // Add provider-id prefix in addition to short alias (ex: kiro/model + kr/model).
@@ -229,6 +239,7 @@ export async function getUnifiedModelsResponse(
             permission: [],
             root: model.id,
             parent: aliasId,
+            ...(contextLength ? { context_length: contextLength } : {}),
           });
         }
       }

src/shared/validation/schemas.ts

@@ -107,6 +107,7 @@ export const createComboSchema = z.object({
   system_message: z.string().max(50000).optional(),
   tool_filter_regex: z.string().max(1000).optional(),
   context_cache_protection: z.boolean().optional(),
+  context_length: z.number().int().min(1000).max(2000000).optional(),
 });
 
 // ──── Auto-Combo Schemas ────