trellis 2.0.7 → 2.0.10

package/dist/{index-8pce39mh.js → index-65z0xfjw.js}

@@ -1,8 +1,13 @@
 // @bun
 import {
   createVcsOp,
-  decisionEntityId
-} from "./index-fd4e26s4.js";
+  decisionEntityId,
+  init_ops,
+  init_types
+} from "./index-v9b4hqa7.js";
+import {
+  __esm
+} from "./index-a76rekgs.js";
 
 // src/decisions/hooks.ts
 class HookRegistry {
@@ -81,6 +86,8 @@ function matchesPattern(pattern, toolName) {
   const escaped = pattern.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*/g, ".*");
   return new RegExp(`^${escaped}$`).test(toolName);
 }
+var init_hooks = () => {};
+
 // src/decisions/auto-capture.ts
 function wrapToolHandler(toolName, handler, opts) {
   return async (params) => {
@@ -134,6 +141,7 @@ function summarize(result) {
   const str = typeof result === "string" ? result : JSON.stringify(result, null, 0);
   return str.length > 500 ? str.slice(0, 500) + "\u2026" : str;
 }
+var init_auto_capture = () => {};
 
 // src/decisions/index.ts
 import { existsSync, readFileSync, writeFileSync, mkdirSync } from "fs";
@@ -268,5 +276,11 @@ function getDecision(ctx, id) {
     return null;
   return buildDecision(ctx, eid);
 }
+var init_decisions = __esm(() => {
+  init_hooks();
+  init_auto_capture();
+  init_ops();
+  init_types();
+});
 
-export { HookRegistry, wrapToolHandler, recordDecision, queryDecisions, getDecisionChain, getDecision };
+export { HookRegistry, wrapToolHandler, recordDecision, queryDecisions, getDecisionChain, getDecision, init_decisions };

package/dist/{index-zf6htvnm.js → index-7gvjxt27.js}

@@ -15,12 +15,23 @@ var init_types = __esm(() => {
 });
 
 // src/embeddings/model.ts
+async function importTransformers() {
+  try {
+    return await import("./transformers.node-bx3q9d7k.js");
+  } catch {
+    try {
+      return await import("@xenova/transformers");
+    } catch {
+      throw new Error("No transformers library found. Install @huggingface/transformers (recommended) or @xenova/transformers.");
+    }
+  }
+}
 async function loadModel(config = DEFAULT_MODEL_CONFIG) {
   if (pipeline)
     return pipeline;
   if (!loadPromise) {
     loadPromise = (async () => {
-      const { pipeline: createPipeline } = await import("@xenova/transformers");
+      const { pipeline: createPipeline } = await importTransformers();
       const opts = {};
       if (config.cacheDir) {
         opts.cache_dir = config.cacheDir;
@@ -652,6 +663,156 @@ var init_search = __esm(() => {
   init_chunker();
 });
 
+// src/embeddings/auto-embed.ts
+function entitySummaryText(entityId, facts, links) {
+  const type = facts.find((f) => f.a === "type")?.v ?? "Entity";
+  const name = facts.find((f) => f.a === "name" || f.a === "title")?.v ?? entityId;
+  const parts = [`${type}: ${name} (${entityId})`];
+  const attrs = facts.filter((f) => !["type", "name", "title", "createdAt", "updatedAt"].includes(f.a));
+  if (attrs.length > 0) {
+    parts.push(attrs.map((f) => `  ${f.a} = ${f.v}`).join(`
+`));
+  }
+  if (links.length > 0) {
+    parts.push("Relations:");
+    parts.push(links.map((l) => `  ${l.a} \u2192 ${l.e2}`).join(`
+`));
+  }
+  return parts.join(`
+`);
+}
+function createAutoEmbedMiddleware(options) {
+  const store = new VectorStore(options.dbPath);
+  const embedFn = options.embedFn ?? embed;
+  const embedIndividual = options.embedIndividualFacts ?? false;
+  return {
+    name: "auto-embed",
+    handleOp: async (op, ctx, next) => {
+      await next(op, ctx);
+      try {
+        await _processOp(op, store, embedFn, embedIndividual);
+      } catch {}
+    },
+    close: () => {
+      store.close();
+    }
+  };
+}
+async function _processOp(op, store, embedFn, embedIndividual) {
+  const now = new Date().toISOString();
+  const entityIds = new Set;
+  if (op.facts)
+    for (const f of op.facts)
+      entityIds.add(f.e);
+  if (op.links)
+    for (const l of op.links) {
+      entityIds.add(l.e1);
+      entityIds.add(l.e2);
+    }
+  if (op.deleteFacts)
+    for (const f of op.deleteFacts)
+      entityIds.add(f.e);
+  if (op.deleteLinks)
+    for (const l of op.deleteLinks) {
+      entityIds.add(l.e1);
+      entityIds.add(l.e2);
+    }
+  if (op.deleteFacts || op.deleteLinks) {
+    for (const eid of entityIds) {
+      store.deleteByEntity(eid);
+    }
+  }
+  if (op.facts && op.facts.length > 0) {
+    const factsByEntity = new Map;
+    for (const f of op.facts) {
+      const existing = factsByEntity.get(f.e) ?? [];
+      existing.push(f);
+      factsByEntity.set(f.e, existing);
+    }
+    const linksByEntity = new Map;
+    if (op.links) {
+      for (const l of op.links) {
+        const existing = linksByEntity.get(l.e1) ?? [];
+        existing.push(l);
+        linksByEntity.set(l.e1, existing);
+      }
+    }
+    const records = [];
+    for (const [eid, facts] of factsByEntity) {
+      const links = linksByEntity.get(eid) ?? [];
+      const summaryText = entitySummaryText(eid, facts, links);
+      if (summaryText.trim()) {
+        try {
+          const vector = await embedFn(summaryText);
+          records.push({
+            id: `entity:${eid}:summary`,
+            entityId: eid,
+            content: summaryText,
+            chunkType: "summary_md",
+            updatedAt: now,
+            embedding: vector
+          });
+        } catch {}
+      }
+      if (embedIndividual) {
+        for (const fact of facts) {
+          if (["type", "createdAt", "updatedAt"].includes(fact.a))
+            continue;
+          const text = `${fact.a}: ${fact.v}`;
+          try {
+            const vector = await embedFn(text);
+            records.push({
+              id: `entity:${eid}:fact:${fact.a}`,
+              entityId: eid,
+              content: text,
+              chunkType: "doc_comment",
+              updatedAt: now,
+              embedding: vector
+            });
+          } catch {}
+        }
+      }
+    }
+    if (records.length > 0) {
+      store.upsertBatch(records);
+    }
+  }
+}
+async function buildRAGContext(query, vectorStore, embedFn = embed, options) {
+  const maxChunks = options?.maxChunks ?? 10;
+  const maxTokens = options?.maxTokens ?? 4000;
+  const minScore = options?.minScore ?? 0.1;
+  const queryVector = await embedFn(query);
+  const results = vectorStore.search(queryVector, {
+    limit: maxChunks * 2,
+    minScore
+  });
+  const chunks = [];
+  let totalChars = 0;
+  for (const r of results) {
+    if (chunks.length >= maxChunks)
+      break;
+    if (totalChars + r.chunk.content.length > maxTokens * 4)
+      break;
+    chunks.push({
+      content: r.chunk.content,
+      entityId: r.chunk.entityId,
+      score: r.score,
+      chunkType: r.chunk.chunkType
+    });
+    totalChars += r.chunk.content.length;
+  }
+  return {
+    query,
+    chunks,
+    estimatedTokens: Math.ceil(totalChars / 4)
+  };
+}
+var init_auto_embed = __esm(() => {
+  init_store();
+  init_model();
+});
+
 // src/embeddings/index.ts
 var exports_embeddings = {};
 __export(exports_embeddings, {
@@ -660,6 +821,7 @@ __export(exports_embeddings, {
   loadModel: () => loadModel,
   embedBatch: () => embedBatch,
   embed: () => embed,
+  createAutoEmbedMiddleware: () => createAutoEmbedMiddleware,
   cosineSimilarity: () => cosineSimilarity,
   chunkSummary: () => chunkSummary,
   chunkMilestone: () => chunkMilestone,
@@ -669,6 +831,7 @@ __export(exports_embeddings, {
   chunkDocComments: () => chunkDocComments,
   chunkDecision: () => chunkDecision,
   chunkCodeEntities: () => chunkCodeEntities,
+  buildRAGContext: () => buildRAGContext,
   VectorStore: () => VectorStore,
   EmbeddingManager: () => EmbeddingManager,
   DEFAULT_MODEL_CONFIG: () => DEFAULT_MODEL_CONFIG
@@ -678,7 +841,8 @@ var init_embeddings = __esm(() => {
   init_model();
   init_store();
   init_search();
+  init_auto_embed();
   init_chunker();
 });
 
-export { DEFAULT_MODEL_CONFIG, loadModel, embed, embedBatch, resetModel, VectorStore, cosineSimilarity, chunkIssue, chunkDecision, chunkMilestone, chunkMarkdown, chunkCodeEntities, chunkDocComments, chunkSummary, chunkFile, slidingWindow, EmbeddingManager, exports_embeddings, init_embeddings };
+export { DEFAULT_MODEL_CONFIG, loadModel, embed, embedBatch, resetModel, init_model, VectorStore, cosineSimilarity, init_store, chunkIssue, chunkDecision, chunkMilestone, chunkMarkdown, chunkCodeEntities, chunkDocComments, chunkSummary, chunkFile, slidingWindow, EmbeddingManager, createAutoEmbedMiddleware, buildRAGContext, init_auto_embed, exports_embeddings, init_embeddings };