feat: add bash/browser agent tools + Docker support
Some checks are pending
CI / build-check-test (push) Waiting to run

- bash-tool.ts: execute shell commands via tool-server HTTP API
- browser-tool.ts: Playwright browser automation (navigate, click, type, screenshot)
- tool-server.mjs: Node.js HTTP server for bash exec + Playwright control (port 7700)
- Dockerfile + docker-compose.yml for containerised deployment
- Register tools in agent toolchain (main.ts, index.ts)
- Add dev:all script to run Vite + tool-server concurrently
This commit is contained in:
JAE 2026-03-26 23:36:29 +00:00
parent 92eea4ea61
commit 00e9816e57
9 changed files with 448 additions and 5 deletions

View file

@ -0,0 +1,29 @@
FROM node:20-slim
ENV DEBIAN_FRONTEND=noninteractive
ENV PLAYWRIGHT_BROWSERS_PATH=/ms-playwright
RUN apt-get update && apt-get install -y --no-install-recommends \
git curl ca-certificates \
libnss3 libnspr4 libatk1.0-0 libatk-bridge2.0-0 libcups2 \
libdrm2 libdbus-1-3 libxkbcommon0 libatspi2.0-0 \
libxcomposite1 libxdamage1 libxfixes3 libxrandr2 libgbm1 libpango-1.0-0 \
libasound2 libwayland-client0 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /app
COPY package*.json ./
RUN npm ci --production
RUN npx playwright install chromium
COPY . .
RUN npm run build 2>/dev/null || true
EXPOSE 5173 7700
ENV TOOL_SERVER_PORT=7700
CMD ["sh", "-c", "node server/tool-server.mjs & npx vite --host 0.0.0.0 --port 5173"]

View file

@ -0,0 +1,16 @@
services:
jae-web:
build: .
ports:
- "5173:5173"
- "7700:7700"
environment:
- OPENAI_API_KEY=${OPENAI_API_KEY}
- OPENAI_BASE_URL=${OPENAI_BASE_URL:-https://api.venice.ai/api/v1}
- TOOL_SERVER_PORT=7700
volumes:
- workspace:/workspace
restart: unless-stopped
volumes:
workspace:

View file

@ -11,7 +11,10 @@
"clean": "shx rm -rf dist", "clean": "shx rm -rf dist",
"terminal-server": "node server/terminal-server.mjs", "terminal-server": "node server/terminal-server.mjs",
"browser-server": "node server/browser-server.mjs", "browser-server": "node server/browser-server.mjs",
"dev:all": "concurrently -k -n VITE,TERM,BROWSER -c cyan,green,magenta \"npm run dev\" \"npm run terminal-server\" \"npm run browser-server\"" "dev:all": "concurrently \"npm run dev\" \"npm run dev:tools\"",
"dev:tools": "node server/tool-server.mjs",
"docker:build": "docker build -t jae-web .",
"docker:run": "docker run -p 5173:5173 -p 7700:7700 -e OPENAI_API_KEY=$OPENAI_API_KEY jae-web"
}, },
"dependencies": { "dependencies": {
"@jaeswift/jae-ai": "file:../../ai", "@jaeswift/jae-ai": "file:../../ai",

View file

@ -0,0 +1,155 @@
import http from 'http';
import { exec } from 'child_process';
import { chromium } from 'playwright';
const PORT = parseInt(process.env.TOOL_SERVER_PORT || '7700');
let browser = null;
let context = null;
let page = null;
const cors = {
'Access-Control-Allow-Origin': '*',
'Access-Control-Allow-Methods': 'POST, GET, OPTIONS',
'Access-Control-Allow-Headers': 'Content-Type',
'Content-Type': 'application/json',
};
function parseBody(req) {
return new Promise((resolve) => {
let body = '';
req.on('data', c => body += c);
req.on('end', () => {
try { resolve(JSON.parse(body || '{}')); } catch { resolve({}); }
});
});
}
async function launchBrowser() {
if (!browser) {
browser = await chromium.launch({
headless: true,
args: ['--no-sandbox', '--disable-setuid-sandbox', '--disable-dev-shm-usage']
});
}
return browser;
}
async function getPage() {
if (!page || page.isClosed()) {
const b = await launchBrowser();
if (context) await context.close().catch(() => {});
context = await b.newContext({ viewport: { width: 1280, height: 800 } });
page = await context.newPage();
}
return page;
}
async function snap() {
const p = await getPage();
const buf = await p.screenshot({ type: 'jpeg', quality: 70, fullPage: false });
return { screenshot: buf.toString('base64'), url: p.url(), title: await p.title() };
}
async function handleBash(body) {
const { command, timeout = 30000 } = body;
if (!command) return { error: 'No command provided' };
return new Promise((resolve) => {
exec(command, {
timeout,
maxBuffer: 10 * 1024 * 1024,
cwd: process.env.WORKSPACE || process.env.HOME || '/root',
env: { ...process.env, TERM: 'dumb', COLUMNS: '200' },
}, (error, stdout, stderr) => {
resolve({
stdout: stdout || '',
stderr: stderr || '',
exitCode: error ? (error.code ?? 1) : 0,
output: (stdout || '') + (stderr ? '\nSTDERR: ' + stderr : ''),
});
});
});
}
async function handleNavigate(body) {
const { url } = body;
if (!url) return { error: 'No URL' };
const p = await getPage();
const target = url.startsWith('http') ? url : 'https://' + url;
await p.goto(target, { timeout: 30000, waitUntil: 'domcontentloaded' });
return snap();
}
async function handleClick(body) {
const p = await getPage();
await p.mouse.click(body.x || 0, body.y || 0);
await p.waitForTimeout(500);
return snap();
}
async function handleType(body) {
const p = await getPage();
if (body.selector) await p.fill(body.selector, body.text || '');
else await p.keyboard.type(body.text || '');
await p.waitForTimeout(300);
return snap();
}
async function handleScroll(body) {
const p = await getPage();
await p.mouse.wheel(0, body.dy || 300);
await p.waitForTimeout(300);
return snap();
}
async function handleBack() {
const p = await getPage();
await p.goBack({ timeout: 10000 }).catch(() => {});
return snap();
}
async function handleText() {
const p = await getPage();
const text = await p.evaluate(() => document.body.innerText);
return { url: p.url(), title: await p.title(), text: text.slice(0, 8000) };
}
async function handleEval(body) {
const p = await getPage();
const result = await p.evaluate(body.script || 'null');
const ss = await snap();
return { ...ss, evalResult: String(result) };
}
const routes = {
'/api/bash': handleBash,
'/api/browser/navigate': handleNavigate,
'/api/browser/click': handleClick,
'/api/browser/type': handleType,
'/api/browser/scroll': handleScroll,
'/api/browser/back': handleBack,
'/api/browser/screenshot': () => snap(),
'/api/browser/text': handleText,
'/api/browser/eval': handleEval,
};
http.createServer(async (req, res) => {
if (req.method === 'OPTIONS') { res.writeHead(204, cors); res.end(); return; }
if (req.url === '/health') { res.writeHead(200, cors); res.end(JSON.stringify({ ok: true })); return; }
const handler = routes[req.url];
if (req.method === 'POST' && handler) {
try {
const body = await parseBody(req);
const result = await handler(body);
res.writeHead(200, cors);
res.end(JSON.stringify(result));
} catch (err) {
res.writeHead(500, cors);
res.end(JSON.stringify({ error: String(err) }));
}
return;
}
res.writeHead(404, cors);
res.end(JSON.stringify({ error: 'Not found' }));
}).listen(PORT, () => {
console.log(`[tool-server] listening on :${'PORT'}`);
});

View file

@ -22,7 +22,7 @@ import {
import { html, render } from "lit"; import { html, render } from "lit";
import { Brain, Download, History, Keyboard, Plus, Settings } from "lucide"; import { Brain, Download, History, Keyboard, Plus, Settings } from "lucide";
import "./app.css"; import "./app.css";
import { createImageGenTool, createMemoryTools, createTTSTool, createWebSearchTool } from "@jaeswift/jae-web-ui"; import { createImageGenTool, createMemoryTools, createTTSTool, createWebSearchTool, createBashTool, createBrowserTool } from "@jaeswift/jae-web-ui";
import { icon } from "@mariozechner/mini-lit"; import { icon } from "@mariozechner/mini-lit";
import { Button } from "@mariozechner/mini-lit/dist/Button.js"; import { Button } from "@mariozechner/mini-lit/dist/Button.js";
import { Input } from "@mariozechner/mini-lit/dist/Input.js"; import { Input } from "@mariozechner/mini-lit/dist/Input.js";
@ -351,7 +351,9 @@ IMPORTANT RULES:
createTools: async (runtimeProvidersFactory: any) => { createTools: async (runtimeProvidersFactory: any) => {
const replTool = createJavaScriptReplTool(); const replTool = createJavaScriptReplTool();
replTool.runtimeProvidersFactory = runtimeProvidersFactory; replTool.runtimeProvidersFactory = runtimeProvidersFactory;
return [replTool, createWebSearchTool(), createImageGenTool(), createTTSTool(), ...createMemoryTools()]; return [replTool, createWebSearchTool(),
createBashTool(),
createBrowserTool(), createImageGenTool(), createTTSTool(), ...createMemoryTools()];
}, },
}); });
costTracker.bindAgent(agent); costTracker.bindAgent(agent);

View file

@ -106,6 +106,7 @@ export { createImageGenTool, imageGenTool } from "./tools/image-gen.js";
// Tools // Tools
export { getToolRenderer, registerToolRenderer, renderTool, setShowJsonMode } from "./tools/index.js"; export { getToolRenderer, registerToolRenderer, renderTool, setShowJsonMode } from "./tools/index.js";
export { createJavaScriptReplTool, javascriptReplTool } from "./tools/javascript-repl.js"; export { createJavaScriptReplTool, javascriptReplTool } from "./tools/javascript-repl.js";
export { createMemoryTools, recallMemoryTool, saveMemoryTool } from "./tools/memory-tool.js";
export { renderCollapsibleHeader, renderHeader } from "./tools/renderer-registry.js"; export { renderCollapsibleHeader, renderHeader } from "./tools/renderer-registry.js";
export { BashRenderer } from "./tools/renderers/BashRenderer.js"; export { BashRenderer } from "./tools/renderers/BashRenderer.js";
export { CalculateRenderer } from "./tools/renderers/CalculateRenderer.js"; export { CalculateRenderer } from "./tools/renderers/CalculateRenderer.js";
@ -118,7 +119,8 @@ export type { ToolRenderer, ToolRenderResult } from "./tools/types.js";
export { createTTSTool, ttsTool } from "./tools/voice-tts.js"; export { createTTSTool, ttsTool } from "./tools/voice-tts.js";
// Venice / community tools // Venice / community tools
export { createWebSearchTool, webSearchTool } from "./tools/web-search.js"; export { createWebSearchTool, webSearchTool } from "./tools/web-search.js";
export { createMemoryTools, saveMemoryTool, recallMemoryTool } from "./tools/memory-tool.js"; export { createBashTool, bashTool } from "./tools/bash-tool.js";
export { createBrowserTool, browserTool } from "./tools/browser-tool.js";
export type { Attachment } from "./utils/attachment-utils.js"; export type { Attachment } from "./utils/attachment-utils.js";
// Utils // Utils
export { loadAttachment } from "./utils/attachment-utils.js"; export { loadAttachment } from "./utils/attachment-utils.js";

View file

@ -0,0 +1,89 @@
import type { AgentTool } from "@jaeswift/jae-agent-core";
import type { ToolResultMessage } from "@jaeswift/jae-ai";
import { Type } from "@sinclair/typebox";
import { html } from "lit";
import { Terminal } from "lucide";
import { registerToolRenderer, renderHeader } from "./renderer-registry.js";
import type { ToolRenderer, ToolRenderResult } from "./types.js";
const TOOL_SERVER = typeof window !== "undefined"
? (window as any).__JAE_TOOL_SERVER__ || "http://localhost:7700"
: "http://localhost:7700";
const bashSchema = Type.Object({
command: Type.String({ description: "Shell command to execute" }),
timeout: Type.Optional(Type.Number({ description: "Timeout in ms (default: 30000)" })),
});
export interface BashDetails {
stdout: string;
stderr: string;
exitCode: number;
command: string;
}
export const bashTool: AgentTool<typeof bashSchema, BashDetails> = {
name: "bash",
label: "Terminal",
description: "Execute a shell command on the server. Use for file operations, installing packages, running scripts, git commands, etc.",
parameters: bashSchema,
async execute(toolCallId, params, signal) {
const { command, timeout = 30000 } = params;
try {
const res = await fetch(TOOL_SERVER + "/api/bash", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ command, timeout }),
signal,
});
const data = await res.json() as any;
if (data.error) {
return {
content: [{ type: "text" as const, text: "Error: " + data.error }],
details: { stdout: "", stderr: data.error, exitCode: 1, command },
};
}
const output = data.output || data.stdout || "";
return {
content: [{ type: "text" as const, text: output.slice(0, 10000) || "(no output)" }],
details: { stdout: data.stdout, stderr: data.stderr, exitCode: data.exitCode, command },
};
} catch (err: any) {
return {
content: [{ type: "text" as const, text: "Bash tool error: " + err.message }],
details: { stdout: "", stderr: err.message, exitCode: 1, command },
};
}
},
};
class BashToolRenderer implements ToolRenderer<{ command: string }, BashDetails> {
render(
params: { command: string } | undefined,
result: ToolResultMessage<BashDetails> | undefined,
): ToolRenderResult {
const state = result ? (result.isError ? "error" : "complete") : "inprogress";
const cmd = result?.details?.command || params?.command || "...";
const exitCode = result?.details?.exitCode;
const icon = state === "error" ? "text-red-500" : exitCode === 0 ? "text-green-500" : "";
if (result?.details) {
const d = result.details;
return {
content: html`
<div class="flex flex-col gap-2">
${renderHeader(state, Terminal, "$ " + cmd)}
<pre class="text-xs bg-black text-green-400 p-3 rounded overflow-auto max-h-64 whitespace-pre-wrap font-mono">${d.stdout}${d.stderr ? "\nSTDERR: " + d.stderr : ""}</pre>
<span class="text-xs ${icon}">Exit code: ${d.exitCode}</span>
</div>`,
isCustom: false,
};
}
return { content: renderHeader(state, Terminal, "$ " + cmd), isCustom: false };
}
}
registerToolRenderer("bash", new BashToolRenderer());
export function createBashTool(): AgentTool<typeof bashSchema, BashDetails> {
return bashTool;
}

View file

@ -0,0 +1,142 @@
import type { AgentTool } from "@jaeswift/jae-agent-core";
import type { ToolResultMessage } from "@jaeswift/jae-ai";
import { Type } from "@sinclair/typebox";
import { html } from "lit";
import { Globe } from "lucide";
import { registerToolRenderer, renderHeader } from "./renderer-registry.js";
import type { ToolRenderer, ToolRenderResult } from "./types.js";
const TOOL_SERVER = typeof window !== "undefined"
? (window as any).__JAE_TOOL_SERVER__ || "http://localhost:7700"
: "http://localhost:7700";
const browserSchema = Type.Object({
action: Type.Union([
Type.Literal("navigate"),
Type.Literal("click"),
Type.Literal("type"),
Type.Literal("scroll"),
Type.Literal("back"),
Type.Literal("screenshot"),
Type.Literal("text"),
Type.Literal("eval"),
], { description: "Browser action to perform" }),
url: Type.Optional(Type.String({ description: "URL to navigate to" })),
x: Type.Optional(Type.Number({ description: "Click X coordinate" })),
y: Type.Optional(Type.Number({ description: "Click Y coordinate" })),
text: Type.Optional(Type.String({ description: "Text to type" })),
selector: Type.Optional(Type.String({ description: "CSS selector to type into" })),
dy: Type.Optional(Type.Number({ description: "Scroll delta Y pixels" })),
script: Type.Optional(Type.String({ description: "JavaScript to evaluate in page" })),
});
export interface BrowserDetails {
action: string;
url?: string;
title?: string;
screenshot?: string;
text?: string;
evalResult?: string;
error?: string;
}
export const browserTool: AgentTool<typeof browserSchema, BrowserDetails> = {
name: "browser",
label: "Browser",
description: "Control a headless browser. Actions: navigate (url), click (x,y), type (text, optional selector), scroll (dy), back, screenshot, text (get page text), eval (run JS).",
parameters: browserSchema,
async execute(toolCallId, params, signal) {
const { action, ...rest } = params;
const endpoint = action === "navigate" ? "/api/browser/navigate"
: action === "click" ? "/api/browser/click"
: action === "type" ? "/api/browser/type"
: action === "scroll" ? "/api/browser/scroll"
: action === "back" ? "/api/browser/back"
: action === "screenshot" ? "/api/browser/screenshot"
: action === "text" ? "/api/browser/text"
: action === "eval" ? "/api/browser/eval"
: null;
if (!endpoint) {
return {
content: [{ type: "text" as const, text: "Unknown action: " + action }],
details: { action, error: "Unknown action" },
};
}
try {
const res = await fetch(TOOL_SERVER + endpoint, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify(rest),
signal,
});
const data = await res.json() as any;
if (data.error) {
return {
content: [{ type: "text" as const, text: "Browser error: " + data.error }],
details: { action, error: data.error },
};
}
// Build text response for LLM
let textParts: string[] = [];
if (data.url) textParts.push("URL: " + data.url);
if (data.title) textParts.push("Title: " + data.title);
if (data.text) textParts.push("Page text:\n" + data.text);
if (data.evalResult) textParts.push("Eval result: " + data.evalResult);
if (data.screenshot) textParts.push("[Screenshot captured]");
if (textParts.length === 0) textParts.push("Action completed.");
// Include screenshot as image content if available
const content: any[] = [{ type: "text" as const, text: textParts.join("\n") }];
if (data.screenshot) {
content.push({ type: "image" as const, mimeType: "image/jpeg", data: data.screenshot });
}
return {
content,
details: { action, url: data.url, title: data.title, screenshot: data.screenshot, text: data.text, evalResult: data.evalResult },
};
} catch (err: any) {
return {
content: [{ type: "text" as const, text: "Browser tool error: " + err.message }],
details: { action, error: err.message },
};
}
},
};
class BrowserToolRenderer implements ToolRenderer<any, BrowserDetails> {
render(params: any | undefined, result: ToolResultMessage<BrowserDetails> | undefined): ToolRenderResult {
const state = result ? (result.isError ? "error" : "complete") : "inprogress";
const action = result?.details?.action || params?.action || "...";
const url = result?.details?.url || params?.url || "";
const label = url ? action + ": " + url : action;
if (result?.details?.screenshot) {
return {
content: html`
<div class="flex flex-col gap-2">
${renderHeader(state, Globe, label)}
<img src="data:image/jpeg;base64,${result.details.screenshot}"
class="rounded border border-border max-w-full" style="max-height:400px"
alt="Browser screenshot" />
${result.details.title ? html`<span class="text-xs text-muted-foreground">${result.details.title}</span>` : html``}
</div>`,
isCustom: false,
};
}
if (result?.details?.text) {
return {
content: html`
<div class="flex flex-col gap-2">
${renderHeader(state, Globe, label)}
<pre class="text-xs p-3 rounded border border-border overflow-auto max-h-48 whitespace-pre-wrap">${result.details.text}</pre>
</div>`,
isCustom: false,
};
}
return { content: renderHeader(state, Globe, label), isCustom: false };
}
}
registerToolRenderer("browser", new BrowserToolRenderer());
export function createBrowserTool(): AgentTool<typeof browserSchema, BrowserDetails> {
return browserTool;
}

View file

@ -2,6 +2,8 @@ import type { ToolResultMessage } from "@jaeswift/jae-ai";
import "./javascript-repl.js"; // Auto-registers the renderer import "./javascript-repl.js"; // Auto-registers the renderer
import "./extract-document.js"; import "./extract-document.js";
import "./memory-tool.js"; // Auto-registers the renderer import "./memory-tool.js"; // Auto-registers the renderer
import "./bash-tool.js"; // Auto-registers bash tool renderer
import "./browser-tool.js"; // Auto-registers browser tool renderer
import { getToolRenderer, registerToolRenderer } from "./renderer-registry.js"; import { getToolRenderer, registerToolRenderer } from "./renderer-registry.js";
import { BashRenderer } from "./renderers/BashRenderer.js"; import { BashRenderer } from "./renderers/BashRenderer.js";
import { DefaultRenderer } from "./renderers/DefaultRenderer.js"; import { DefaultRenderer } from "./renderers/DefaultRenderer.js";
@ -47,6 +49,9 @@ export function renderTool(
export { getToolRenderer, registerToolRenderer }; export { getToolRenderer, registerToolRenderer };
export { createImageGenTool, type ImageGenDetails, imageGenTool } from "./image-gen.js"; export { createImageGenTool, type ImageGenDetails, imageGenTool } from "./image-gen.js";
export { createMemoryTools, recallMemoryTool, saveMemoryTool } from "./memory-tool.js";
export { createTTSTool, type TTSDetails, ttsTool } from "./voice-tts.js"; export { createTTSTool, type TTSDetails, ttsTool } from "./voice-tts.js";
export { createWebSearchTool, type WebSearchDetails, type WebSearchResult, webSearchTool } from "./web-search.js"; export { createWebSearchTool, type WebSearchDetails, type WebSearchResult, webSearchTool } from "./web-search.js";
export { createMemoryTools, saveMemoryTool, recallMemoryTool } from "./memory-tool.js";
export { createBashTool, type BashDetails, bashTool } from "./bash-tool.js";
export { createBrowserTool, type BrowserDetails, browserTool } from "./browser-tool.js";