feat: repo readme embeddings (setup_instructions)

package.json

@@ -91,4 +91,4 @@
     ]
   },
   "packageManager": "[email protected]"
-}
+}

src/adapters/supabase/helpers/embeddings.ts

@@ -88,7 +88,7 @@ export class Embeddings extends Super {
       source_id: sourceId,
       type,
       plaintext: htmlToPlainText(markdownToPlainText(markdown)).trim(),
-      embedding: await this._embedWithVoyage(markdown),
+      embedding: await this._embedWithVoyage(markdown, "document"),
       metadata,
       created_at: new Date().toISOString(),
       modified_at: new Date().toISOString(),
@@ -111,8 +111,13 @@ export class Embeddings extends Super {
     if (!body) {
       throw new Error(this.context.logger.error("Markdown not found", { sourceId })?.logMessage.raw);
     }
+    const embeddingData = await this.getEmbedding(sourceId);
+
+    if (!embeddingData) {
+      return await this.createEmbedding(sourceId, type, body, metadata);
+    }
 
-    const embedding = await this._embedWithVoyage(body);
+    const embedding = await this._embedWithVoyage(body, "document");
 
     const toStore: Omit<CommentType, "created_at"> = {
       source_id: sourceId,
@@ -123,12 +128,6 @@ export class Embeddings extends Super {
       modified_at: new Date().toISOString(),
     };
 
-    const embeddingData = await this.getEmbedding(sourceId);
-
-    if (!embeddingData) {
-      return await this.createEmbedding(sourceId, type, body, metadata);
-    }
-
     const { error } = await this.supabase.from("content").update(toStore).eq("source_id", sourceId);
 
     if (error) {
@@ -158,7 +157,7 @@ export class Embeddings extends Super {
   // Working with embeddings
 
   async findSimilarIssues(markdown: string, threshold: number, currentId: string): Promise<IssueSimilaritySearchResult[]> {
-    const embedding = await this._embedWithVoyage(markdown);
+    const embedding = await this._embedWithVoyage(markdown, "query");
     const { data, error } = await this.supabase.rpc("find_similar_issues", {
       current_id: currentId,
       query_embedding: embedding,
@@ -173,14 +172,15 @@ export class Embeddings extends Super {
 
   // Helpers
 
-  private async _embedWithVoyage(text: string | null): Promise<number[]> {
+  async _embedWithVoyage(text: string | null, inputType: "document" | "query"): Promise<number[]> {
     try {
       if (text === null) {
         return new Array(VECTOR_SIZE).fill(0);
       } else {
         const response = await this._voyageClient.embed({
           input: text,
           model: "voyage-large-2-instruct",
+          inputType: inputType
         });
         return (response.data && response.data[0]?.embedding) || [];
       }
@@ -189,7 +189,7 @@ export class Embeddings extends Super {
     }
   }
 
-  private _getMetadata(payload: Context["payload"]) {
+  private _getMetadata(payload: Context<"issue_comment.edited" | "issue_comment.deleted" | "issues.edited" | "issues.deleted" | "issue_comment.created" | "issues.opened">["payload"]) {
     const {
       repository: { private: isPrivate, node_id: repoNodeId },
       issue: { node_id: issueNodeId },

src/handlers/create-comment-embedding.ts → src/handlers/comments/create-comment-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 export async function createCommentEmbedding(context: Context<"issue_comment.created">): Promise<CallbackResult> {
   const {

src/handlers/delete-comment-embedding.ts → src/handlers/comments/delete-comment-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 export async function deleteCommentEmbedding(context: Context<"issue_comment.deleted">): Promise<CallbackResult> {
   const {

src/handlers/update-comment-embedding.ts → src/handlers/comments/update-comment-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 /**
  * Updates embeddings for comments.

src/handlers/onboarding/create-setup-instructions.ts

@@ -0,0 +1,77 @@
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
+
+export async function createSetupInstructions(context: Context<"push">): Promise<CallbackResult> {
+    const {
+        logger,
+        octokit,
+        adapters: { supabase },
+        payload: { repository, commits, sender, pusher }
+    } = context;
+
+    const docs = []
+
+    for (const commit of commits) {
+        const { added, modified } = commit;
+        const files = []
+
+        if (added && added.length > 0) {
+            files.push(...added)
+        }
+        if (modified && modified.length > 0) {
+            files.push(...modified)
+        }
+
+        for (const file of files) {
+            if (file.endsWith(".md")) {
+                docs.push(file)
+            }
+        }
+    }
+
+    if (docs.length === 0) {
+        return { status: 200, reason: "no markdown files found" };
+    }
+
+    logger.info(`Found ${docs.length} markdown files`);
+    if (!repository.owner || !repository.name) {
+        return { status: 200, reason: "no repository owner or name found" };
+    }
+
+    /**
+     * voyageai use a special encoding schema and we cannot easily
+     * use their encoder so we will just have to play it by ear for now.
+     */
+    for (const doc of docs) {
+        const sourceId = repository.full_name + "/" + doc;
+        const docContent = await octokit.repos.getContent({
+            owner: repository.owner.login,
+            repo: repository.name,
+            path: doc,
+            mediaType: {
+                format: "raw",
+            }
+        });
+
+        if (!docContent.data) {
+            return { status: 200, reason: "no content found" };
+        }
+
+        const text = docContent.data as unknown as string;
+
+        const uploaded = await supabase.embeddings.createEmbedding(sourceId, "setup_instructions", text, {
+            author_association: "OWNER",
+            author_id: sender?.id,
+            isPrivate: repository.private,
+            repo_node_id: repository.node_id,
+            repo_full_name: repository.full_name,
+            fileChunkIndex: 0,
+        });
+
+        logger.info("Uploaded markdown file", { ...uploaded, embedding: "removed for brevity" });
+    }
+
+    logger.ok("Successfully uploaded setup instructions", { repository: repository.full_name });
+
+    return { status: 200, reason: "success" };
+}

src/handlers/create-task-embedding.ts → src/handlers/tasks/create-task-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 export async function addTaskEmbedding(context: Context<"issues.opened">): Promise<CallbackResult> {
   const {

src/handlers/delete-task-embedding.ts → src/handlers/tasks/delete-task-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 export async function deleteTaskEmbedding(context: Context<"issues.deleted">): Promise<CallbackResult> {
   const {

src/handlers/task-deduplication.ts → src/handlers/tasks/task-deduplication.ts

@@ -1,5 +1,5 @@
-import { Context } from "../types";
-import { IssueSimilaritySearchResult } from "../types/embeddings";
+import { Context } from "../../types";
+import { IssueSimilaritySearchResult } from "../../types/embeddings";
 
 export interface IssueGraphqlResponse {
   node: {
@@ -88,6 +88,10 @@ async function handleSimilarIssuesComment(context: Context, issueNumber: number,
   const commentBody = issueList.map((issue) => `- [${issue.node.title}](${issue.node.url}) Similarity: ${issue.similarity}`).join("\n");
   const body = `This issue seems to be similar to the following issue(s):\n\n${commentBody}`;
 
+  if (!payload.repository.owner || !payload.repository.name) {
+    return;
+  }
+
   const existingComments = await context.octokit.issues.listComments({
     owner: payload.repository.owner.login,
     repo: payload.repository.name,
@@ -98,6 +102,10 @@ async function handleSimilarIssuesComment(context: Context, issueNumber: number,
     (comment) => comment.body && comment.body.includes("This issue seems to be similar to the following issue(s)")
   );
 
+  if (!payload.repository.owner || !payload.repository.name) {
+    return;
+  }
+
   if (existingComment) {
     await context.octokit.issues.updateComment({
       owner: payload.repository.owner.login,

src/handlers/update-task-embedding.ts → src/handlers/tasks/update-task-embedding.ts

@@ -1,5 +1,5 @@
-import { CallbackResult } from "../proxy-callbacks";
-import { Context } from "../types";
+import { CallbackResult } from "../../proxy-callbacks";
+import { Context } from "../../types";
 
 export async function updateTaskEmbedding(context: Context<"issues.edited">): Promise<CallbackResult> {
   const {

src/proxy-callbacks.ts

@@ -1,11 +1,12 @@
-import { createCommentEmbedding } from "./handlers/create-comment-embedding";
-import { addTaskEmbedding } from "./handlers/create-task-embedding";
-import { deleteCommentEmbedding } from "./handlers/delete-comment-embedding";
-import { deleteTaskEmbedding } from "./handlers/delete-task-embedding";
-import { taskSimilaritySearch } from "./handlers/task-deduplication";
-import { updateCommentEmbedding } from "./handlers/update-comment-embedding";
-import { updateTaskEmbedding } from "./handlers/update-task-embedding";
+import { createCommentEmbedding } from "./handlers/comments/create-comment-embedding";
+import { addTaskEmbedding } from "./handlers/tasks/create-task-embedding";
+import { deleteCommentEmbedding } from "./handlers/comments/delete-comment-embedding";
+import { deleteTaskEmbedding } from "./handlers/tasks/delete-task-embedding";
+import { taskSimilaritySearch } from "./handlers/tasks/task-deduplication";
+import { updateCommentEmbedding } from "./handlers/comments/update-comment-embedding";
+import { updateTaskEmbedding } from "./handlers/tasks/update-task-embedding";
 import { Context, SupportedEvents, SupportedEventsU } from "./types";
+import { createSetupInstructions } from "./handlers/onboarding/create-setup-instructions";
 
 export type CallbackResult = { status: 200 | 201 | 204 | 404 | 500; reason: string; content?: string | Record<string, unknown> };
 
@@ -45,6 +46,8 @@ const callbacks = {
   "issues.opened": [addTaskEmbedding, taskSimilaritySearch],
   "issues.edited": [updateTaskEmbedding],
   "issues.deleted": [deleteTaskEmbedding],
+
+  "push": [createSetupInstructions]
 } as ProxyCallbacks;
 
 /**

src/types/context.ts

@@ -16,7 +16,8 @@ export type SupportedEventsU =
   | "issue_comment.edited"
   | "issues.opened"
   | "issues.edited"
-  | "issues.deleted";
+  | "issues.deleted"
+  | "push"
 
 export type SupportedEvents = {
   [K in SupportedEventsU]: K extends WebhookEventName ? WebhookEvent<K> : never;

src/types/embeddings.ts

@@ -14,6 +14,7 @@ export interface CommentMetadata {
   issue_node_id: string;
   repo_node_id: string;
   isPrivate: boolean;
+  [key: string]: any;
 }
 
 export interface IssueSimilaritySearchResult {

tests/main.test.ts

@@ -199,11 +199,11 @@ function createContext(
   okSpy: jest.SpyInstance;
   verboseSpy: jest.SpyInstance;
   repo: Context["payload"]["repository"];
-  issue1: Context["payload"]["issue"];
+  issue1: Context<"issue_comment.created">["payload"]["issue"];
 } {
   const repo = db.repo.findFirst({ where: { id: { equals: repoId } } }) as unknown as Context["payload"]["repository"];
   const sender = db.users.findFirst({ where: { id: { equals: payloadSenderId } } }) as unknown as Context["payload"]["sender"];
-  const issue1 = db.issue.findFirst({ where: { id: { equals: issueOne } } }) as unknown as Context["payload"]["issue"];
+  const issue1 = db.issue.findFirst({ where: { id: { equals: issueOne } } }) as unknown as Context<"issue_comment.created">["payload"]["issue"];
 
   createComment(commentBody, commentId, nodeId); // create it first then pull it from the DB and feed it to _createContext
   const comment = db.issueComments.findFirst({
@@ -237,7 +237,7 @@ function createContext(
 function createContextInner(
   repo: Context["payload"]["repository"],
   sender: Context["payload"]["sender"],
-  issue: Context["payload"]["issue"],
+  issue: Context<"issue_comment.created">["payload"]["issue"],
   comment: SupportedEvents["issue_comment.created"]["payload"]["comment"],
   eventName: SupportedEventsU
 ): Context<"issue_comment.created"> {