Add bots files handling logic

2026-10-21 17:56:09 +01:00
parent b0af4db1bf
commit d17a4a7bdd
97 changed files with 5843 additions and 41 deletions
@@ -0,0 +1,95 @@
+import { z } from "zod";
+import { AppData } from "@/src/data/app-data";
+import grabBotFile from "@/src/functions/backend/db/bot-files/grab-bot-file";
+import grabDirNames from "@/src/utils/grab-dir-names";
+import readMedia from "@/src/harness/media/read-media";
+import type { HarnessTool } from "../../types";
+import { fileRef, type StorageToolContext } from "./storage-tool-context";
+
+type Args = {
+    file_id?: number;
+    path?: string;
+    offset?: number;
+    scope?: string;
+};
+
+const TextExtensions = [
+    "txt",
+    "md",
+    "markdown",
+    "csv",
+    "json",
+    "html",
+    "htm",
+    "xml",
+    "yml",
+    "yaml",
+    "log",
+    "tsv",
+];
+
+/**
+ * Reads one of the bot's files: text as it is (a window at a time when it
+ * is long), an image or audio through the media harness, and a PDF or DOCX
+ * through the text extractor.
+ */
+export default function readFileTool(context: StorageToolContext): HarnessTool<Args> {
+    return {
+        name: "read_file",
+        description: `Read a file you saved, by its ref id or its path. Text comes back in a window of about ${Math.round(AppData.BotFileReadMaxChars / 1000)}k characters; use "offset" for the next part. Images and audio are read for you, and PDFs and Word documents are turned into text.`,
+        schema: z.object({
+            file_id: z.number().int().positive().optional(),
+            path: z.string().max(AppData.BotFilePathMaxChars).optional(),
+            offset: z.number().int().min(0).optional(),
+            scope: z.string().optional().describe(`"bot:<id>" or "workspace:<id>", with "path"`),
+        }),
+        run: async ({ args }) => {
+            const [scope_name, owner] = (args.scope || "").split(":");
+
+            const file = await grabBotFile({
+                scopes: context.scopes,
+                file_id: args.file_id,
+                scope: scope_name,
+                owner_id: owner,
+                path: args.path,
+            });
+
+            if (!file?.id || !file.media?.media_write_path) {
+                return `No file matches that. Use list_files to see what you have.`;
+            }
+
+            const { DATA_DIR } = grabDirNames();
+            const absolute = `${DATA_DIR}/${file.media.media_write_path}`;
+            const ext = (file.name || "").split(".").pop()?.toLowerCase() || "";
+
+            if (TextExtensions.includes(ext)) {
+                const text = await Bun.file(absolute).text();
+                const offset = args.offset || 0;
+                const window = text.slice(offset, offset + AppData.BotFileReadMaxChars);
+                const end = offset + window.length;
+
+                return [
+                    `${file.path} ${fileRef({ file_id: file.id })}`,
+                    window,
+                    end < text.length
+                        ? `\n[Read ${offset}-${end} of ${text.length} characters. Call read_file with offset ${end} for the rest.]`
+                        : "",
+                ]
+                    .filter(Boolean)
+                    .join("\n\n");
+            }
+
+            const bytes = new Uint8Array(await Bun.file(absolute).arrayBuffer());
+
+            const text = await readMedia({
+                data: bytes,
+                media_type: file.mime_type || "application/octet-stream",
+                name: file.name || undefined,
+                bot_id: context.bot_id,
+                workspace_id: context.workspace_id,
+            });
+
+            return `${file.path} ${fileRef({ file_id: file.id })}\n\n${text.slice(0, AppData.BotFileReadMaxChars)}`;
+        },
+    };
+}