RooCodeInc · mrubens · Sep 9, 2026
@@ -0,0 +1,5 @@
+---
+'@roomote/web': minor
+---
+
+Add a `browse` tool to Fast Sessions that gives each conversation a private cloud browser without a sandbox. Configure `R_FAST_BROWSER_PROVIDER=browseruse` and `R_BROWSER_USE_API_KEY` to enable it. Screenshots and recordings are saved as Session artifacts, shown in the web transcript, and attached to chat replies when the agent asks for delivery.
@@ -115,16 +115,16 @@
 ENV S3_ENDPOINT=http://minio:9000
 ENV S3_PRESIGN_ENDPOINT=http://minio:9000
 ENV S3_REGION=us-east-1
 ENV S3_ACCESS_KEY_ID=roomote
 ENV S3_SECRET_ACCESS_KEY=roomote-local-artifacts-password
 ENV S3_BUCKET_ARTIFACTS=roomote-artifacts
 ENV JOB_AUTH_PRIVATE_KEY=local-self-host-build-job-auth-private-key
 ENV JOB_AUTH_PUBLIC_KEY=local-self-host-build-job-auth-public-key
 ENV PREVIEW_AUTH_PRIVATE_KEY=local-self-host-build-preview-auth-private-key
 ENV PREVIEW_AUTH_PUBLIC_KEY=local-self-host-build-preview-auth-public-key
 ENV DASHBOARD_PASSWORD=roomote-local-admin
 ENV ENCRYPTION_KEY=local-roomote-encryption-key-0001
 ENV ARTIFACT_SIGNING_KEY=local-roomote-artifact-signing-key-1
 ENV PREVIEW_PROXY_BASE_URL=http://localhost:18081
 ENV PREVIEW_DOMAINS=localhost,127.0.0.1,roomotepreview.localhost

@@ -323,6 +323,13 @@
   apt -qq install -y git ffmpeg util-linux && \
   rm -rf /var/lib/apt/lists/*
 
+# The Fast `browse` tool drives a cloud browser through the agent-browser CLI
+# from the api process. Only the CLI ships here; no Chrome is installed, so the
+# tool stays off until R_FAST_BROWSER_PROVIDER names a provider.
+ARG AGENT_BROWSER_VERSION=0.37.1
+RUN npm install -g "agent-browser@${AGENT_BROWSER_VERSION}" && \
+  npm cache clean --force
+
 # The install and version check run as root; with the image's HOME=/tmp they
 # would bake root-owned dotdirs (npm cache, OpenCode's data/config/cache
 # dirs) into the layer. The runtime user shares that HOME and must be able to

@@ -28,6 +28,7 @@ import {
 } from '@roomote/communication';
 import {
   admitFastAgentHumanFollowUp,
+  buildFastAgentMediaArtifactCreator,
   createFastAgentConversationArtifact,
   persistFastAgentInlineHumanTurn,
   recordFastAgentConversationMessageBestEffort,
@@ -395,6 +396,7 @@ export async function processDiscordFastAgentMessage(
             fastConversationId: session.id,
             ...artifact,
           }),
+        createMediaArtifact: buildFastAgentMediaArtifactCreator(session.id),
         ...(durableTurnForResume
           ? {
               requestDurableResume: () =>

@@ -14,6 +14,7 @@ import {
 } from '@roomote/communication';
 import {
   buildFastAgentArtifactCreator,
+  buildFastAgentMediaArtifactCreator,
   findFastAgentSessionForProviderMessage,
   persistFastAgentInlineHumanTurn,
   recordFastAgentConversationMessageBestEffort,
@@ -164,6 +165,7 @@ async function processFastAgentReaction(params: {
             }
           : {}),
         createArtifact: buildFastAgentArtifactCreator(session.id),
+        createMediaArtifact: buildFastAgentMediaArtifactCreator(session.id),
         activity: createFastAgentSlackSessionActivity({
           slack: context.slack,
           workspaceId: context.teamId,

@@ -24,6 +24,7 @@ import {
 import { appendAttachmentTextsToPromptText } from '@roomote/cloud-agents';
 import {
   admitFastAgentHumanFollowUp,
+  buildFastAgentMediaArtifactCreator,
   createFastAgentConversationArtifact,
   persistFastAgentInlineHumanTurn,
   wakeFastAgentParentEventAt,
@@ -308,6 +309,7 @@ export async function processFastAgentMessage(params: {
             fastConversationId: session.id,
             ...artifact,
           }),
+        createMediaArtifact: buildFastAgentMediaArtifactCreator(session.id),
         ...(durableTurn
           ? {
               requestDurableResume: () =>

@@ -367,6 +367,8 @@ as per-task auth tokens or workspace paths.
 | `R_ALLOWED_EMAILS`             | Optional          | Comma-separated email allowlist for deployments that restrict sign-in by email.  |
 | `R_ELEVENLABS_API_KEY`         | Optional          | ElevenLabs API key for narrated feature-demo videos. The key stays on the control plane; sandboxes reach text-to-speech only through an authenticated Roomote endpoint. A key scoped to text-to-speech only is sufficient and recommended. |
 | `R_ELEVENLABS_VOICE_ID`        | Optional          | ElevenLabs voice ID used for feature-demo narration. Required alongside the API key for narration to be available.                                                       |
+| `R_FAST_BROWSER_PROVIDER`      | Optional          | Enables the Fast `browse` tool, which gives every Session a private browser without a sandbox. Set to `browseruse` to use Browser Use cloud browsers (requires `R_BROWSER_USE_API_KEY`). Unset disables the tool.                          |
+| `R_BROWSER_USE_API_KEY`        | Optional          | Browser Use API key for the Fast `browse` tool. The key stays on the control plane; the model only sees the browser command output.                                                                                                       |
 
 During Microsoft Teams setup, Roomote uses the Microsoft Entra app values for
 the Teams bot by default. Use **Show advanced config** after the Directory

@@ -253,11 +253,13 @@ function getVisibleToolInput(
         ? 'message'
         : toolName === 'inspect_images'
           ? 'question'
-          : toolName === 'post_to_channel'
-            ? 'text'
-            : toolName === 'send_chat_reaction_emoji'
-              ? 'name'
-              : null;
+          : toolName === 'browse'
+            ? 'command'
+            : toolName === 'post_to_channel'
+              ? 'text'
+              : toolName === 'send_chat_reaction_emoji'
+                ? 'name'
+                : null;
   if (!visibleField && toolName !== 'receive_task_report') {
     return null;
   }

@@ -10,6 +10,7 @@ import {
   FolderIcon,
   GalleryVerticalEnd,
   GitPullRequest,
+  Globe,
   HardDriveUpload,
   List,
   ListChecks,
@@ -40,6 +41,7 @@ export function toolIconForKey(key: ToolIconKey): LucideIcon {
   if (key === 'message') return MessageSquareText;
   if (key === 'memory') return BookOpenText;
   if (key === 'artifact') return HardDriveUpload;
+  if (key === 'globe') return Globe;
   if (key === 'widget') return GalleryVerticalEnd;
   if (key === 'roomote') return RoomoteR;
   if (key === 'video') return Video;

@@ -33,6 +33,7 @@ interface ResolvedToolPolicy {
 }
 
 const CONSEQUENTIAL_RECEIPTS = new Set([
+  'browse',
   'launch_task',
   'review_pull_request',
   'cancel_task',

@@ -43,6 +43,7 @@ export type ToolIconKey =
   | 'environment'
   | 'alert'
   | 'messages'
+  | 'globe'
   | 'tool';
 
 type ToolPresentationPhase = 'running' | 'completed' | 'failed';
@@ -106,6 +107,7 @@ const COMMUNICATION_TOOL_NAMES = new Set([
   'ignore_event',
 ]);
 const TOOL_ICON_OVERRIDES: Readonly<Partial<Record<string, ToolIconKey>>> = {
+  browse: 'globe',
   manage_custom_automations: 'task',
   manage_wakeups: 'task',
   get_about_me: 'roomote',
@@ -426,6 +428,16 @@ function resolveReceiptLanguage(
       verb: byPhase('Inspecting', 'Inspected', 'Failed to Inspect'),
       object: 'Images',
     };
+  if (toolName === 'browse') {
+    const command = stringArgument(args, 'command');
+    const summary = command
+      ? command.split(/\s+/u).slice(0, 2).join(' ')
+      : null;
+    return {
+      verb: byPhase('Browsing', 'Browsed', 'Failed to Browse'),
+      object: summary ?? 'browser',
+    };
+  }
   if (nativeToolName === 'skill' || nativeToolName === 'load_skill') {
     const name = stringArgument(args, 'name');
     return {

@@ -5,12 +5,15 @@ import { isSubagentToolPayload } from './subagent-tool';
 import type { AcpToolCallUiMessage, AcpToolResultUiMessage } from './types';
 
 const MANAGE_ARTIFACTS_TOOL_NAME = 'manage_artifacts';
+const BROWSE_TOOL_NAME = 'browse';
 
 type VisualProofUploadExtraction = {
   artifactId: string;
   artifactType: 'visual-proof';
   viewUrl: string;
   rawUrl?: string;
+  /** Set by the Fast `browse` tool so a recording's rawUrl renders as video. */
+  contentType?: string;
 };
 
 export type VisualProofDisplayMedia =
@@ -103,6 +106,7 @@ function parseVisualProofSuccessPayload(
   const viewUrl = asNonEmptyString(record.viewUrl);
   const artifactType = asNonEmptyString(record.artifactType);
   const rawUrl = asNonEmptyString(record.rawUrl) ?? undefined;
+  const contentType = asNonEmptyString(record.contentType) ?? undefined;
 
   if (!artifactId || !viewUrl || artifactType !== 'visual-proof') {
     return null;
@@ -113,6 +117,7 @@ function parseVisualProofSuccessPayload(
     artifactType: 'visual-proof',
     viewUrl,
     ...(rawUrl ? { rawUrl } : {}),
+    ...(contentType ? { contentType } : {}),
   };
 }
 
@@ -134,14 +139,22 @@ function extractVisualProofUploadFromToolMessage(
     return null;
   }
 
-  if (msg.data.isMcp !== true) {
-    return null;
-  }
-
   const toolName = getMcpToolName(msg.data);
+  // The Fast `browse` native tool saves its screenshots and recordings the
+  // same way task sandboxes upload visual proof.
+  const isBrowseCapture =
+    msg.data.isMcp === false &&
+    msg.data.isRoomoteNativeTool === true &&
+    toolName === BROWSE_TOOL_NAME;
+
+  if (!isBrowseCapture) {
+    if (msg.data.isMcp !== true) {
+      return null;
+    }
 
-  if (toolName !== MANAGE_ARTIFACTS_TOOL_NAME) {
-    return null;
+    if (toolName !== MANAGE_ARTIFACTS_TOOL_NAME) {
+      return null;
+    }
   }
 
   const output = asNonEmptyString(msg.data.output);
@@ -228,7 +241,20 @@ function resolveVisualProofDisplayMedia(
     }
   }
 
-  // Upload contract: rawUrl is only set for images.
+  // The Fast browse tool signs rawUrl for recordings too; the raw route
+  // serves WebM, so play it inline rather than treating it as an image.
+  if (extraction.rawUrl && extraction.contentType?.startsWith('video/')) {
+    return {
+      kind: 'video',
+      src: extraction.rawUrl,
+      viewUrl: extraction.viewUrl,
+      artifactId: extraction.artifactId,
+      path,
+      version,
+    };
+  }
+
+  // Upload contract: otherwise rawUrl is only set for images.
   if (extraction.rawUrl) {
     return {
       kind: 'image',

@@ -17,8 +17,10 @@ import {
 } from '@roomote/cloud-agents/server';
 import {
   buildFastAgentArtifactCreator,
+  buildFastAgentMediaArtifactCreator,
   buildFastAgentSurfaceReplyDelivery,
   createFastAgentSessionArtifact,
+  createFastAgentSessionMediaArtifact,
   persistFastAgentInlineHumanTurn,
   resolveUserMcpServerConfigs,
   wakeFastAgentParentEventAt,
@@ -454,6 +456,11 @@ export async function startFastSessionCommand(
               ...artifact,
             });
           },
+          createMediaArtifact: (artifact) =>
+            createFastAgentSessionMediaArtifact({
+              sessionId: unifiedSession.id,
+              ...artifact,
+            }),
           launchTask,
           postReply: async () => {},
         },
@@ -550,6 +557,7 @@ export async function startSetupFastSessionCommand(
         conversation,
         adapter: {
           createArtifact: buildFastAgentArtifactCreator(session.id),
+          createMediaArtifact: buildFastAgentMediaArtifactCreator(session.id),
           launchTask: createFastAgentWebTaskLauncher({
             userId: auth.userId,
           }),
@@ -852,6 +860,7 @@ export async function submitFastSessionUserInputCommand(
         conversation,
         adapter: {
           createArtifact: buildFastAgentArtifactCreator(session.id),
+          createMediaArtifact: buildFastAgentMediaArtifactCreator(session.id),
           launchTask: createFastAgentWebTaskLauncher({
             userId: auth.userId,
           }),

@@ -1,7 +1,10 @@
 import { createHash } from 'node:crypto';
 
 import { type FastAgentTurnAdapter } from '@roomote/cloud-agents/server';
-import { buildFastAgentArtifactCreator } from '@roomote/sdk/server';
+import {
+  buildFastAgentArtifactCreator,
+  buildFastAgentMediaArtifactCreator,
+} from '@roomote/sdk/server';
 import {
   and,
   db,
@@ -380,6 +383,9 @@ async function buildSetupPlatformEventTurn(
         createArtifact: buildFastAgentArtifactCreator(
           conversation.fastConversationId,
         ),
+        createMediaArtifact: buildFastAgentMediaArtifactCreator(
+          conversation.fastConversationId,
+        ),
         launchTask: (
           await import('@roomote/cloud-agents/server')
         ).createFastAgentWebTaskLauncher({