@alexeiled/pi-model-router 0.5.0

package/extensions/ui.ts

@@ -0,0 +1,131 @@
+import type { ExtensionContext } from '@earendil-works/pi-coding-agent';
+import type {
+  RouterConfig,
+  RouterPinByProfile,
+  RouterThinkingByProfile,
+  RoutingDecision,
+} from './types';
+
+const getEffectiveThinking = (
+  thinkingByProfile: RouterThinkingByProfile,
+  profileName: string,
+  decision: RoutingDecision,
+) => thinkingByProfile[profileName]?.[decision.tier] ?? decision.thinking;
+
+const getDecisionFlags = (decision: RoutingDecision): string[] => {
+  const flags: string[] = [];
+  if (decision.isFallback) flags.push('fallback');
+  if (decision.isBudgetForced) flags.push('budget-limit');
+  if (decision.isRuleMatched) flags.push('rule');
+  return flags;
+};
+
+export const formatDecision = (decision: RoutingDecision): string => {
+  return `${decision.profile}: ${decision.tier} -> ${decision.targetProvider}/${decision.targetModelId} [${decision.thinking}] (${decision.reasoning})`;
+};
+
+export const formatPinSummary = (
+  pinnedTierByProfile: RouterPinByProfile,
+): string => {
+  const entries = Object.entries(pinnedTierByProfile)
+    .sort(([a], [b]) => a.localeCompare(b))
+    .map(([profile, tier]) => `${profile}:${tier}`);
+  return entries.length > 0 ? entries.join(', ') : 'none';
+};
+
+export const formatThinkingSummary = (
+  thinkingByProfile: RouterThinkingByProfile,
+): string => {
+  const entries = Object.entries(thinkingByProfile)
+    .sort(([a], [b]) => a.localeCompare(b))
+    .map(([profile, tierMap]) => {
+      const tiers = Object.entries(tierMap)
+        .sort(([a], [b]) => a.localeCompare(b))
+        .map(([tier, level]) => `${tier}:${level}`);
+      return `${profile}(${tiers.join(',')})`;
+    });
+  return entries.length > 0 ? entries.join(', ') : 'none';
+};
+
+export const formatModelRef = (ref: string | undefined): string => {
+  return ref ?? 'none';
+};
+
+export const updateStatus = (
+  ctx: ExtensionContext,
+  routerEnabled: boolean,
+  selectedProfile: string | undefined,
+  pinnedTierByProfile: RouterPinByProfile,
+  thinkingByProfile: RouterThinkingByProfile,
+  lastDecision: RoutingDecision | undefined,
+  lastNonRouterModel: string | undefined,
+  accumulatedCost: number,
+  widgetEnabled: boolean,
+  currentConfig: RouterConfig,
+) => {
+  const activeRouterProfile = routerEnabled ? selectedProfile : undefined;
+  const statusProfile = selectedProfile ?? 'none';
+  const activePin = selectedProfile
+    ? pinnedTierByProfile[selectedProfile]
+    : undefined;
+  const pinLabel = activePin ? ` [pin:${activePin}]` : '';
+
+  if (activeRouterProfile) {
+    const matchesProfile =
+      lastDecision && lastDecision.profile === activeRouterProfile;
+    const matchesPin = activePin ? lastDecision?.tier === activePin : true;
+
+    let statusText: string;
+    if (lastDecision && matchesProfile && matchesPin) {
+      const effectiveThinking = getEffectiveThinking(
+        thinkingByProfile,
+        activeRouterProfile,
+        lastDecision,
+      );
+      statusText = `router:${activeRouterProfile}${pinLabel} -> ${lastDecision.tier} -> ${lastDecision.targetProvider}/${lastDecision.targetModelId} (${effectiveThinking})`;
+    } else {
+      statusText = `router:${activeRouterProfile}${pinLabel} -> waiting`;
+    }
+    ctx.ui.setStatus('router', `🚥 ${statusText}`);
+  } else {
+    ctx.ui.setStatus('router', undefined);
+  }
+
+  if (!widgetEnabled) {
+    ctx.ui.setWidget('router', undefined);
+    return;
+  }
+
+  const widgetLines = [
+    `Router: ${routerEnabled ? 'enabled' : 'disabled'}`,
+    `Profile: ${statusProfile}${activeRouterProfile ? ' (active)' : ''}`,
+    `Pin: ${activePin ?? 'auto'}`,
+    `Cost: $${accumulatedCost.toFixed(4)}` +
+      (currentConfig.maxSessionBudget
+        ? ` / $${currentConfig.maxSessionBudget.toFixed(2)}`
+        : ''),
+  ];
+  if (lastDecision && lastDecision.profile === statusProfile) {
+    const effectiveThinking = getEffectiveThinking(
+      thinkingByProfile,
+      statusProfile,
+      lastDecision,
+    );
+    const flags = getDecisionFlags(lastDecision);
+    const flagsStr = flags.length > 0 ? ` [${flags.join(',')}]` : '';
+
+    widgetLines.push(
+      `Route: ${lastDecision.tier}${flagsStr} -> ${lastDecision.targetProvider}/${lastDecision.targetModelId} (${effectiveThinking})`,
+      `Phase: ${lastDecision.phase}`,
+    );
+  } else if (!routerEnabled && lastNonRouterModel) {
+    widgetLines.push(`Fallback: ${lastNonRouterModel}`);
+  }
+  if (Object.keys(pinnedTierByProfile).length > 1) {
+    widgetLines.push(`Pins: ${formatPinSummary(pinnedTierByProfile)}`);
+  }
+  ctx.ui.setWidget(
+    'router',
+    widgetLines.map((line) => ctx.ui.theme.fg('dim', line)),
+  );
+};

package/model-router.example.json

@@ -0,0 +1,72 @@
+{
+  "debug": false,
+  "classifierModel": "flash",
+  "phaseBias": 0.5,
+  "maxSessionBudget": 1.0,
+  "models": {
+    "gpt-pro": {
+      "model": "openai/gpt-5.4-pro",
+      "contextWindow": 256000,
+      "maxTokens": 64000,
+      "thinkingLevels": ["high", "medium", "low"]
+    },
+    "flash": {
+      "model": "google/gemini-flash-latest",
+      "contextWindow": 1048576
+    },
+    "nano": {
+      "model": "openai/gpt-5.4-nano",
+      "contextWindow": 128000,
+      "maxTokens": 8192,
+      "reasoning": false
+    },
+    "sonnet": {
+      "model": "anthropic/claude-3-5-sonnet-20241022",
+      "contextWindow": 200000
+    }
+  },
+  "rules": [
+    {
+      "matches": ["deploy", "production", "release"],
+      "tier": "high",
+      "reason": "Safety check for production tasks"
+    },
+    { "matches": "changelog", "tier": "low" }
+  ],
+  "profiles": {
+    "auto": {
+      "high": {
+        "model": "gpt-pro",
+        "thinking": "high",
+        "fallbacks": ["sonnet"]
+      },
+      "medium": { "model": "flash", "thinking": "medium" },
+      "low": { "model": "nano", "thinking": "low" }
+    },
+    "cheap": {
+      "high": { "model": "flash", "thinking": "low" },
+      "medium": { "model": "nano", "thinking": "off" },
+      "low": { "model": "google/gemini-flash-lite-latest", "thinking": "off" }
+    },
+    "deep": {
+      "high": {
+        "model": "openai/o1-preview",
+        "thinking": "xhigh",
+        "thinkingLevels": ["xhigh", "high", "medium", "low"]
+      },
+      "medium": { "model": "gpt-pro", "thinking": "medium" },
+      "low": { "model": "flash", "thinking": "low" }
+    },
+    "anthropic": {
+      "high": {
+        "model": "sonnet",
+        "thinking": "high"
+      },
+      "medium": {
+        "model": "sonnet",
+        "thinking": "medium"
+      },
+      "low": { "model": "anthropic/claude-3-haiku-20240307", "thinking": "low" }
+    }
+  }
+}

package/package.json

@@ -0,0 +1,77 @@
+{
+  "name": "@alexeiled/pi-model-router",
+  "version": "0.5.0",
+  "type": "module",
+  "files": [
+    "extensions",
+    "!extensions/*.test.ts",
+    "LICENSE",
+    "model-router.example.json",
+    "README.md",
+    "CHANGELOG.md"
+  ],
+  "description": "Smart per-turn model router for Pi that optimizes your AI budget and usage limits by dynamically switching LLM tiers based on task complexity, budget, and phase awareness.",
+  "keywords": [
+    "pi-package",
+    "pi",
+    "model-router",
+    "llm",
+    "coding-agent",
+    "extension"
+  ],
+  "license": "MIT",
+  "engines": {
+    "node": ">=22.19.0"
+  },
+  "packageManager": "npm@11.18.0",
+  "author": "Alexei Ledenev",
+  "contributors": [
+    "Ye Liu"
+  ],
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/alexei-led/pi-model-router.git"
+  },
+  "homepage": "https://github.com/alexei-led/pi-model-router#readme",
+  "bugs": {
+    "url": "https://github.com/alexei-led/pi-model-router/issues"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "exports": {
+    ".": "./extensions/index.ts"
+  },
+  "pi": {
+    "extensions": [
+      "./extensions/index.ts"
+    ]
+  },
+  "scripts": {
+    "tsc": "tsc --noEmit",
+    "build": "tsc",
+    "prepublishOnly": "npm run check && npm test",
+    "test": "vitest run",
+    "check": "biome check --error-on-warnings . && npm run tsc",
+    "lint": "biome lint --error-on-warnings .",
+    "lint:fix": "biome lint --write --error-on-warnings .",
+    "format": "biome format --write .",
+    "format:check": "biome format .",
+    "pack:dry": "npm pack --dry-run"
+  },
+  "peerDependencies": {
+    "@earendil-works/pi-agent-core": ">=0.86.0",
+    "@earendil-works/pi-ai": ">=0.86.0",
+    "@earendil-works/pi-coding-agent": ">=0.86.0",
+    "@earendil-works/pi-tui": ">=0.86.0"
+  },
+  "devDependencies": {
+    "@biomejs/biome": "2.5.14",
+    "@earendil-works/pi-agent-core": "0.86.0",
+    "@earendil-works/pi-ai": "0.86.0",
+    "@earendil-works/pi-coding-agent": "0.86.0",
+    "@earendil-works/pi-tui": "0.86.0",
+    "typescript": "^7.0.2",
+    "vitest": "^5.0.1"
+  }
+}