fix(copilot): surface upstream model limits

2026-07-15 18:21:09 +00:00 · 2026-03-07 09:26:09 +07:00
parent 8403284cd2
commit 8a5ed656ee
6 changed files with 300 additions and 14 deletions
@@ -13,6 +13,7 @@ import {
  getAvailableModels,
  isCopilotApiInstalled,
 } from '../copilot';
+import type { CopilotModel } from '../copilot';
 import { loadOrCreateUnifiedConfig, saveUnifiedConfig } from '../config/unified-config-loader';
 import { DEFAULT_COPILOT_CONFIG } from '../config/unified-config-types';
 import { ok, fail, info, color } from '../utils/ui';
@@ -195,14 +196,56 @@ async function handleModels(): Promise<number> {
    const defaultMark = model.isDefault ? ' (default)' : '';
    console.log(`  ${model.id}${current}${defaultMark}`);
    console.log(`    Provider: ${model.provider}`);
+    const limits = formatModelLimits(model);
+    if (limits) {
+      console.log(`    Limits:   ${limits}`);
+    }
  }

+  console.log('');
+  if (models.some((model) => formatModelLimits(model))) {
+    console.log('Live limits above come from GitHub Copilot model metadata.');
+  } else {
+    console.log('Live Copilot limits were unavailable. Start the daemon and rerun this command.');
+  }
+  console.log(
+    'CCS can switch Copilot models, but it cannot raise GitHub Copilot prompt/context caps.'
+  );
  console.log('');
  console.log('To change model: ccs config (Copilot section)');

  return 0;
 }

+function formatCompactTokens(value: number): string {
+  if (value >= 1_000_000) {
+    const millions = value / 1_000_000;
+    return millions % 1 === 0 ? `${millions}M` : `${millions.toFixed(1)}M`;
+  }
+  if (value >= 1_000) {
+    const thousands = value / 1_000;
+    return thousands % 1 === 0 ? `${thousands}K` : `${thousands.toFixed(1)}K`;
+  }
+  return `${value}`;
+}
+
+function formatModelLimits(model: CopilotModel): string | null {
+  if (!model.limits) return null;
+
+  const parts: string[] = [];
+  if (model.limits.maxPromptTokens) {
+    parts.push(`prompt ${formatCompactTokens(model.limits.maxPromptTokens)}`);
+  }
+  if (model.limits.maxContextWindowTokens) {
+    parts.push(`context ${formatCompactTokens(model.limits.maxContextWindowTokens)}`);
+  }
+  if (model.limits.maxOutputTokens) {
+    parts.push(`output ${formatCompactTokens(model.limits.maxOutputTokens)}`);
+  }
+
+  return parts.length > 0 ? parts.join(' | ') : null;
+}
+
 function formatQuotaLine(
  label: string,
  snapshot: {
@@ -9,6 +9,27 @@
 import * as http from 'http';
 import { CopilotModel } from './types';

+const DEFAULT_COPILOT_MODEL_ID = 'gpt-4.1';
+const MAX_MODELS_BODY_SIZE = 1024 * 1024; // 1MB
+
+interface CopilotDaemonModelLimits {
+  max_context_window_tokens?: unknown;
+  max_output_tokens?: unknown;
+  max_prompt_tokens?: unknown;
+}
+
+interface CopilotDaemonModel {
+  id?: unknown;
+  name?: unknown;
+  capabilities?: {
+    limits?: CopilotDaemonModelLimits;
+  };
+}
+
+interface CopilotModelsResponse {
+  data?: CopilotDaemonModel[];
+}
+
 /**
 * Default models available through copilot-api.
 * Used as fallback when API is not reachable.
@@ -154,6 +175,13 @@ export const DEFAULT_COPILOT_MODELS: CopilotModel[] = [
 */
 export async function fetchModelsFromDaemon(port: number): Promise<CopilotModel[]> {
  return new Promise((resolve) => {
+    let resolved = false;
+    const safeResolve = (models: CopilotModel[]) => {
+      if (resolved) return;
+      resolved = true;
+      resolve(models);
+    };
+
    const req = http.request(
      {
        // Use 127.0.0.1 instead of localhost for more reliable local connections
@@ -168,36 +196,37 @@ export async function fetchModelsFromDaemon(port: number): Promise<CopilotModel[

        res.on('data', (chunk) => {
          data += chunk;
+          if (data.length > MAX_MODELS_BODY_SIZE) {
+            req.destroy();
+            safeResolve(DEFAULT_COPILOT_MODELS);
+          }
        });

        res.on('end', () => {
          try {
-            const response = JSON.parse(data) as { data?: Array<{ id: string }> };
-            if (response.data && Array.isArray(response.data)) {
-              const models: CopilotModel[] = response.data.map((m) => ({
-                id: m.id,
-                name: formatModelName(m.id),
-                provider: detectProvider(m.id),
-                isDefault: m.id === 'gpt-4.1', // Free tier default
-              }));
-              resolve(models.length > 0 ? models : DEFAULT_COPILOT_MODELS);
+            const response = JSON.parse(data) as CopilotModelsResponse;
+            if (Array.isArray(response.data)) {
+              const models = response.data
+                .map(mapDaemonModel)
+                .filter((model): model is CopilotModel => model !== null);
+              safeResolve(models.length > 0 ? models : DEFAULT_COPILOT_MODELS);
            } else {
-              resolve(DEFAULT_COPILOT_MODELS);
+              safeResolve(DEFAULT_COPILOT_MODELS);
            }
          } catch {
-            resolve(DEFAULT_COPILOT_MODELS);
+            safeResolve(DEFAULT_COPILOT_MODELS);
          }
        });
      }
    );

    req.on('error', () => {
-      resolve(DEFAULT_COPILOT_MODELS);
+      safeResolve(DEFAULT_COPILOT_MODELS);
    });

    req.on('timeout', () => {
      req.destroy();
-      resolve(DEFAULT_COPILOT_MODELS);
+      safeResolve(DEFAULT_COPILOT_MODELS);
    });

    req.end();
@@ -216,7 +245,7 @@ export async function getAvailableModels(port: number): Promise<CopilotModel[]>
 * Uses gpt-4.1 as it's available on free tier.
 */
 export function getDefaultModel(): string {
-  return 'gpt-4.1';
+  return DEFAULT_COPILOT_MODEL_ID;
 }

 /**
@@ -246,3 +275,40 @@ function formatModelName(modelId: string): string {
    .map((word) => word.charAt(0).toUpperCase() + word.slice(1))
    .join(' ');
 }
+
+function normalizeLimitValue(value: unknown): number | undefined {
+  return typeof value === 'number' && Number.isFinite(value) && value > 0 ? value : undefined;
+}
+
+function extractLimits(limits?: CopilotDaemonModelLimits): CopilotModel['limits'] | undefined {
+  if (!limits) return undefined;
+
+  const normalized = {
+    maxContextWindowTokens: normalizeLimitValue(limits.max_context_window_tokens),
+    maxOutputTokens: normalizeLimitValue(limits.max_output_tokens),
+    maxPromptTokens: normalizeLimitValue(limits.max_prompt_tokens),
+  };
+
+  return Object.values(normalized).some((value) => value !== undefined) ? normalized : undefined;
+}
+
+function mapDaemonModel(entry: CopilotDaemonModel): CopilotModel | null {
+  if (typeof entry.id !== 'string') return null;
+  const id = entry.id.trim();
+  if (!id) return null;
+
+  const defaultMeta = DEFAULT_COPILOT_MODELS.find((model) => model.id === id);
+  const limits = extractLimits(entry.capabilities?.limits);
+
+  return {
+    ...defaultMeta,
+    id,
+    name:
+      typeof entry.name === 'string' && entry.name.trim().length > 0
+        ? entry.name.trim()
+        : (defaultMeta?.name ?? formatModelName(id)),
+    provider: defaultMeta?.provider ?? detectProvider(id),
+    isDefault: id === DEFAULT_COPILOT_MODEL_ID,
+    ...(limits ? { limits } : {}),
+  };
+}
@@ -41,6 +41,15 @@ export interface CopilotStatus {
 */
 export type CopilotPlanTier = 'free' | 'pro' | 'pro+' | 'business' | 'enterprise';

+export interface CopilotModelLimits {
+  /** Maximum total context window exposed by GitHub Copilot */
+  maxContextWindowTokens?: number;
+  /** Maximum output/completion tokens exposed by GitHub Copilot */
+  maxOutputTokens?: number;
+  /** Maximum prompt/input tokens exposed by GitHub Copilot */
+  maxPromptTokens?: number;
+}
+
 /**
 * Copilot model information.
 */
@@ -57,6 +66,8 @@ export interface CopilotModel {
  multiplier?: number;
  /** Whether this model is in preview */
  preview?: boolean;
+  /** Live model limits returned by GitHub Copilot metadata, when available */
+  limits?: CopilotModelLimits;
 }

 /**
@@ -0,0 +1,96 @@
+import { describe, expect, it } from 'bun:test';
+import * as http from 'http';
+import { DEFAULT_COPILOT_MODELS, fetchModelsFromDaemon } from '../../../src/copilot/copilot-models';
+
+describe('fetchModelsFromDaemon', () => {
+  it('falls back to defaults when daemon is unreachable', async () => {
+    const models = await fetchModelsFromDaemon(9999);
+    expect(models).toEqual(DEFAULT_COPILOT_MODELS);
+  });
+
+  it('falls back to defaults when daemon returns invalid JSON', async () => {
+    const server = http.createServer((_req, res) => {
+      res.writeHead(200, { 'Content-Type': 'application/json' });
+      res.end('{not-valid-json');
+    });
+
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    const address = server.address();
+    if (!address || typeof address === 'string') {
+      throw new Error('Unable to resolve test server port');
+    }
+
+    try {
+      const models = await fetchModelsFromDaemon(address.port);
+      expect(models).toEqual(DEFAULT_COPILOT_MODELS);
+    } finally {
+      await new Promise<void>((resolve) => server.close(() => resolve()));
+    }
+  });
+
+  it('parses live limits from daemon metadata and preserves known model metadata', async () => {
+    const server = http.createServer((_req, res) => {
+      res.writeHead(200, { 'Content-Type': 'application/json' });
+      res.end(
+        JSON.stringify({
+          data: [
+            {
+              id: 'claude-sonnet-4.5',
+              name: 'Claude Sonnet 4.5',
+              capabilities: {
+                limits: {
+                  max_context_window_tokens: 128000,
+                  max_prompt_tokens: 128000,
+                  max_output_tokens: 64000,
+                },
+              },
+            },
+            {
+              id: 'claude-sonnet-4.6',
+              name: 'Claude Sonnet 4.6',
+              capabilities: {
+                limits: {
+                  max_context_window_tokens: 128000,
+                  max_prompt_tokens: 128000,
+                  max_output_tokens: 64000,
+                },
+              },
+            },
+            {
+              id: '',
+              name: 'invalid-entry',
+            },
+          ],
+        })
+      );
+    });
+
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    const address = server.address();
+    if (!address || typeof address === 'string') {
+      throw new Error('Unable to resolve test server port');
+    }
+
+    try {
+      const models = await fetchModelsFromDaemon(address.port);
+      expect(models).toHaveLength(2);
+
+      const knownModel = models.find((model) => model.id === 'claude-sonnet-4.5');
+      expect(knownModel?.provider).toBe('anthropic');
+      expect(knownModel?.minPlan).toBe('pro');
+      expect(knownModel?.multiplier).toBe(1);
+      expect(knownModel?.limits).toEqual({
+        maxContextWindowTokens: 128000,
+        maxPromptTokens: 128000,
+        maxOutputTokens: 64000,
+      });
+
+      const liveOnlyModel = models.find((model) => model.id === 'claude-sonnet-4.6');
+      expect(liveOnlyModel?.name).toBe('Claude Sonnet 4.6');
+      expect(liveOnlyModel?.provider).toBe('anthropic');
+      expect(liveOnlyModel?.limits?.maxPromptTokens).toBe(128000);
+    } finally {
+      await new Promise<void>((resolve) => server.close(() => resolve()));
+    }
+  });
+});
@@ -14,6 +14,35 @@ import { FREE_PRESETS, PAID_PRESETS } from './presets';
 import { FlexibleModelSelector } from './model-selector';
 import type { ModelPreset } from './types';

+function formatCompactTokens(value: number): string {
+  if (value >= 1_000_000) {
+    const millions = value / 1_000_000;
+    return millions % 1 === 0 ? `${millions}M` : `${millions.toFixed(1)}M`;
+  }
+  if (value >= 1_000) {
+    const thousands = value / 1_000;
+    return thousands % 1 === 0 ? `${thousands}K` : `${thousands.toFixed(1)}K`;
+  }
+  return `${value}`;
+}
+
+function formatModelLimits(model?: CopilotModel): string | null {
+  if (!model?.limits) return null;
+
+  const parts: string[] = [];
+  if (model.limits.maxPromptTokens) {
+    parts.push(`prompt ${formatCompactTokens(model.limits.maxPromptTokens)}`);
+  }
+  if (model.limits.maxContextWindowTokens) {
+    parts.push(`context ${formatCompactTokens(model.limits.maxContextWindowTokens)}`);
+  }
+  if (model.limits.maxOutputTokens) {
+    parts.push(`output ${formatCompactTokens(model.limits.maxOutputTokens)}`);
+  }
+
+  return parts.length > 0 ? parts.join(' | ') : null;
+}
+
 interface ModelConfigTabProps {
  currentModel: string;
  opusModel: string;
@@ -41,6 +70,19 @@ export function ModelConfigTab({
  onUpdateSonnetModel,
  onUpdateHaikuModel,
 }: ModelConfigTabProps) {
+  const mappedModelLimits = [
+    { label: 'Default', id: currentModel },
+    { label: 'Opus', id: opusModel || currentModel },
+    { label: 'Sonnet', id: sonnetModel || currentModel },
+    { label: 'Haiku', id: haikuModel || currentModel },
+  ]
+    .map(({ label, id }) => {
+      const model = models.find((entry) => entry.id === id);
+      const limits = formatModelLimits(model);
+      return limits ? { label, id, limits } : null;
+    })
+    .filter((entry): entry is { label: string; id: string; limits: string } => entry !== null);
+
  return (
    <TabsContent
      value="config"
@@ -125,6 +167,26 @@ export function ModelConfigTab({
            <p className="text-xs text-muted-foreground mb-4">
              Configure which models to use for each tier
            </p>
+            <div className="mb-4 rounded-lg border border-amber-500/30 bg-amber-500/10 p-3 text-xs text-amber-900 dark:text-amber-200">
+              <p className="font-medium">GitHub Copilot controls prompt/context limits upstream.</p>
+              <p className="mt-1">
+                CCS can switch Copilot models, but it cannot increase the provider&apos;s max prompt
+                or context window.
+              </p>
+              {mappedModelLimits.length > 0 ? (
+                <div className="mt-2 space-y-1 text-[11px] font-mono">
+                  {mappedModelLimits.map((entry) => (
+                    <p key={`${entry.label}-${entry.id}`}>
+                      {entry.label}: {entry.id} ({entry.limits})
+                    </p>
+                  ))}
+                </div>
+              ) : (
+                <p className="mt-2 text-[11px] font-mono">
+                  Start the daemon to inspect live model limits from GitHub Copilot metadata.
+                </p>
+              )}
+            </div>
            <div className="space-y-4">
              <FlexibleModelSelector
                label="Default Model"
@@ -54,6 +54,12 @@ export interface CopilotConfig {
 /** GitHub Copilot plan tiers */
 export type CopilotPlanTier = 'free' | 'pro' | 'pro+' | 'business' | 'enterprise';

+export interface CopilotModelLimits {
+  maxContextWindowTokens?: number;
+  maxOutputTokens?: number;
+  maxPromptTokens?: number;
+}
+
 export interface CopilotModel {
  id: string;
  name: string;
@@ -66,6 +72,8 @@ export interface CopilotModel {
  multiplier?: number;
  /** Whether this model is in preview */
  preview?: boolean;
+  /** Live model limits returned by GitHub Copilot metadata, when available */
+  limits?: CopilotModelLimits;
 }

 export interface CopilotRawSettings {