@rely-ai/caliber 1.5.6 → 1.6.0

package/dist/bin.js

@@ -520,6 +520,11 @@ var DEFAULT_MODELS = {
   cursor: "default",
   "claude-cli": "default"
 };
+var DEFAULT_FAST_MODELS = {
+  anthropic: "claude-haiku-4-5-20251001",
+  vertex: "claude-haiku-4-5-20251001",
+  openai: "gpt-4.1-mini"
+};
 function loadConfig() {
   const envConfig = resolveFromEnv();
   if (envConfig) return envConfig;
@@ -591,7 +596,12 @@ function getConfigFilePath() {
   return CONFIG_FILE;
 }
 function getFastModel() {
-  return process.env.CALIBER_FAST_MODEL || process.env.ANTHROPIC_SMALL_FAST_MODEL || void 0;
+  if (process.env.CALIBER_FAST_MODEL) return process.env.CALIBER_FAST_MODEL;
+  if (process.env.ANTHROPIC_SMALL_FAST_MODEL) return process.env.ANTHROPIC_SMALL_FAST_MODEL;
+  const config = loadConfig();
+  if (config?.fastModel) return config.fastModel;
+  if (config?.provider) return DEFAULT_FAST_MODELS[config.provider];
+  return void 0;
 }
 
 // src/llm/anthropic.ts
@@ -4361,6 +4371,7 @@ async function scoreWithLLM(candidates, toolDeps) {
     const vendorTag = c.vendor ? " [VENDOR/OFFICIAL]" : "";
     return `${i}. "${c.name}"${vendorTag} (${c.stars} stars) \u2014 ${c.description.slice(0, 100)}`;
   }).join("\n");
+  const fastModel = getFastModel();
   const scored = await llmJsonCall({
     system: SCORE_MCP_PROMPT,
     prompt: `TOOL DEPENDENCIES IN PROJECT:
@@ -4368,7 +4379,8 @@ ${toolDeps.join(", ")}
 
 MCP SERVER CANDIDATES:
 ${candidateList}`,
-    maxTokens: 4e3
+    maxTokens: 4e3,
+    ...fastModel ? { model: fastModel } : {}
   });
   if (!Array.isArray(scored)) return [];
   return scored.filter((s) => s.score >= 60 && s.index >= 0 && s.index < candidates.length).sort((a, b) => b.score - a.score).slice(0, 5).map((s) => ({
@@ -4401,13 +4413,15 @@ async function fetchReadme(repoFullName) {
 async function extractMcpConfig(readme, serverName) {
   try {
     const truncated = readme.length > 15e3 ? readme.slice(0, 15e3) : readme;
+    const fastModel = getFastModel();
     const result = await llmJsonCall({
       system: EXTRACT_CONFIG_PROMPT,
       prompt: `MCP Server: ${serverName}
 
 README:
 ${truncated}`,
-      maxTokens: 2e3
+      maxTokens: 2e3,
+      ...fastModel ? { model: fastModel } : {}
     });
     if (!result || !result.command) return null;
     return {
@@ -4943,6 +4957,7 @@ async function searchAllProviders(technologies, platform) {
 }
 async function scoreWithLLM2(candidates, projectContext, technologies) {
   const candidateList = candidates.map((c, i) => `${i}. "${c.name}" \u2014 ${c.reason || "no description"}`).join("\n");
+  const fastModel = getFastModel();
   const scored = await llmJsonCall({
     system: `You evaluate whether AI agent skills and tools are relevant to a specific software project.
 Given a project context and a list of candidates, score each one's relevance from 0-100 and provide a brief reason (max 80 chars).
@@ -4970,7 +4985,8 @@ ${technologies.join(", ")}
 
 CANDIDATES:
 ${candidateList}`,
-    maxTokens: 8e3
+    maxTokens: 8e3,
+    ...fastModel ? { model: fastModel } : {}
   });
   if (!Array.isArray(scored)) return [];
   return scored.filter((s) => s.score >= 60 && s.index >= 0 && s.index < candidates.length).sort((a, b) => b.score - a.score).slice(0, 20).map((s) => ({
@@ -6945,10 +6961,12 @@ ${skillsSummary}`);
   const prompt = `${contextParts.length ? contextParts.join("\n\n---\n\n") + "\n\n---\n\n" : ""}## Tool Events from Session (${fittedEvents.length} events)
 
 ${eventsText}`;
+  const fastModel = getFastModel();
   const raw = await llmCall({
     system: LEARN_SYSTEM_PROMPT,
     prompt,
-    maxTokens: 4096
+    maxTokens: 4096,
+    ...fastModel ? { model: fastModel } : {}
   });
   return parseAnalysisResponse(raw);
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rely-ai/caliber",
-  "version": "1.5.6",
+  "version": "1.6.0",
   "description": "Analyze your codebase and generate optimized AI agent configs (CLAUDE.md, .cursorrules, skills) — no API key needed",
   "type": "module",
   "bin": {