janex-ai

JANEX-AI Agent

The open-source AI agent that does not just chat — it opens browsers, edits repos, runs tools, solves verification widgets, hunts bugs, and finishes workflows.

JANEX-AI is a terminal-native autonomous workspace for builders, researchers, CTF players, and automation power users.
Bring your own model endpoint, launch the TUI, and let the agent operate inside your real environment.

Quick Install · Why JANEX-AI · Features · Bug Hunt · Social Research · Config

Quick Install

npm install -g janex-ai
janex setup
janex

That is the whole path: install the CLI, configure your model endpoint, then start the terminal workspace.

For development:

git clone https://github.com/JaneHax/Janex-ai.git
cd Janex-ai
npm install
npm run build
npm link
janex

Need a simple prompt-only mode?

janex --lite

Want JANEX-AI in chat apps?

janex gateway

Why JANEX-AI?

Most AI coding tools stop at suggestions. JANEX-AI is built to operate.

If you want...	JANEX-AI gives you...
An AI that can actually use your machine	Terminal execution, file editing, repo search, build/test loops, and tool calls.
Browser automation that can continue past verification prompts	CloakBrowser automation with screenshot observation, human-like input, CAPTCHA/grid/audio/slider flows.
Security and CTF help that is not generic	A full Bug Hunt skill pack for web, pwn, crypto, reverse, forensics, OSINT, malware, misc, AI/ML, and writeups.
Research that goes beyond one search result	Web search, scraping, forum/social research, synthesis, reports, and depth-based review/debate.
Trading and market analysis workflows	Stock/crypto/DeFi skills plus a trading tool for analysis, technical indicators, news, portfolio, comparison, and risk views.
Decisions that need disagreement	Council/debate patterns, supporter/skeptic/judge research agents, and native multi-agent routing.
An agent you can reach outside the terminal	Discord, Telegram, and WhatsApp gateway support.
No lock-in to one hosted model	OpenAI-compatible, Anthropic-compatible, and custom endpoint configuration.
Extensible tools	MCP server manager plus a large local skill system.

JANEX-AI is for people who want an agent that can see, click, type, read, edit, run, investigate, remember, and report back.

What Can It Do?

You:  Fix the failing auth tests and patch the bug.
JANEX-AI: Reads the repo → edits code → runs tests → reads errors → fixes again.

You:  Open this signup page and complete the verification step.
JANEX-AI: Opens CloakBrowser → fills fields → solves image/audio/slider challenge → submits.

You:  Solve this CTF web challenge.
JANEX-AI: Loads Bug Hunt → maps routes → checks auth/injection/upload/JWT paths → writes exploit notes.

You:  Research what people say about this product across Reddit, X, YouTube, TikTok, HN, GitHub, Polymarket, and the web.
JANEX-AI: Runs social research → clusters evidence → ranks signals → writes a source-backed brief.

You:  Analyze NVDA vs AMD with technicals, news, risk, and crypto/DeFi context.
JANEX-AI: Uses trading + finance skills → gathers market data → compares signals → highlights risk, not financial advice.

You:  This product decision is debatable — convene a council.
JANEX-AI: Splits voices into Architect, Skeptic, Pragmatist, and Critic → surfaces disagreement → gives a verdict.

You:  /depth xhigh — audit this project from Telegram.
JANEX-AI: Applies deeper routing for that gateway user and returns a reviewed result.

Standout Features

1. Stealth Browser + CAPTCHA / Verification Solving

JANEX-AI ships with a Playwright-powered browser tool backed by CloakBrowser. It is designed for realistic browser automation: persistent profiles, visual page observation, human-like mouse paths, natural typing, scrolling, drag/drop, and autonomous handling of interactive verification widgets.

Capability	What JANEX-AI does
Image grid challenges	Screenshots tiles, asks a vision model to identify targets, clicks selected tiles with human-like movement, and verifies.
Audio challenges	Opens audio mode, retrieves or listens to the challenge, transcribes it through the configured audio path, and submits the answer.
Hybrid mode	Tries image solving first, then falls back to audio when needed.
Sliders / drag widgets	Uses eased mouse paths, overshoot correction, micro-jitter, and hold/drag actions.
FunCaptcha / Arkose-style widgets	Detects puzzle style and applies rotation, drag/drop, image-match, or interaction flows.
Turnstile-style managed prompts	Observes the widget and performs the required click/interaction flow when possible.
Human behavior	Warmup movement, natural delays, typo correction, scroll easing, random pauses, and persistent browser profile.

Browser actions include:

navigate, click, fill, type, screenshot, snapshot,
signup-assist, signin-assist,
detect-captcha, solve-captcha, captcha-grid, click-tile, captcha-verify,
slider-analyze, drag-to, hold-click

Hybrid CAPTCHA configuration:

# ~/.janex/config.yaml
captchaAudio: "hybrid"      # image | audio | hybrid
visionModel: "your-vision-model"
visionBaseUrl: "https://your-openai-compatible-endpoint/v1"
visionApiKey: "..."

Best results come from a strong vision-capable model. If your main model is text-only, configure a separate visionModel for browser screenshots.

2. Bug Hunt Skills

JANEX-AI includes a real local Bug Hunt skill system for CTFs, authorized security testing, security labs, and defensive research. Instead of generic security advice, JANEX-AI can route into category-specific playbooks and technique notes.

Skill Area	What it covers
Web exploitation	SQLi, XSS, SSTI, SSRF, XXE, JWT/JWE, OAuth/OIDC, SAML, file upload bugs, auth bypass, request smuggling, prototype pollution, GraphQL, parser tricks, known CVE patterns.
Pwn / binary exploitation	Buffer overflow, ROP, heap exploitation, format strings, seccomp, shellcoding, ELF triage.
Crypto	RSA, AES modes, ECC, PRNG weakness, padding oracles, lattice/Coppersmith-style attacks, ZKP and challenge-specific math.
Reverse engineering	ELF/PE analysis, stripped binaries, VMs, WASM, obfuscation, game/client reversing.
Forensics	PCAP, disk images, memory dumps, steganography, logs, browser artifacts, timeline reconstruction.
OSINT	Username/email/phone/domain/IP pivots, geolocation, DNS/WHOIS, public web traces, social-media investigation.
Malware	C2 traffic analysis, packers, .NET, obfuscated scripts, shellcode triage.
AI/ML security	Prompt injection challenges, adversarial examples, model extraction concepts, jail/challenge analysis.
Writeups	Turn solved challenges into clean, reproducible writeups.

Pre-install CTF tooling when you want a ready lab box:

bash skills/bug-hunt/install_ctf_tools.sh all

Use only where you are allowed to test: CTFs, lab boxes, authorized bug bounty targets, internal audits, and defensive investigations.

3. Research + Social Intelligence

JANEX-AI research is not limited to generic web search. The skill tree includes social-researching for finding what people actually say across social platforms, plus deep research, fact-checking, citation checks, claim verification, summarization, and video analysis.

Research lane	Coverage
Social research	Reddit, X, YouTube, TikTok, Hacker News, NodeLoc, Xiaohongshu, Polymarket, GitHub, and the wider web.
China AI intelligence	china_ai_research crawls public NodeLoc discussions plus available China/social/web sources for recent Chinese AI news and community sentiment.
Deep research	Multi-source planning, source gathering, deep-reading, synthesis, citations, and reports.
Fact / claim checks	Claim extraction, citation verification, skeptical review, and source-backed verdicts.
Video research	Useful when the topic includes YouTube/TikTok/video claims or creator sentiment.
Market / product intelligence	Competitive research, user sentiment, investor/market signals, and source attribution.

The higher /depth levels activate more review behavior in the native research pipeline: supporter, skeptic, debate system, judge, citation guardian, logic critic, writer, and final reviewer.

4. Trading, Stocks, Crypto, and DeFi Analysis

JANEX-AI includes both a trading tool and finance skills. It can analyze tickers, market news, technical indicators, stock comparisons, crypto/on-chain context, DeFi protocol risks, and portfolio-style summaries.

Finance lane	What it does
Trading analysis	Symbol analysis, sentiment/news, technical indicators, comparison, portfolio, risk, and report actions.
Stocks	Fundamental + technical analysis, earnings/guidance context, valuation and sector risk framing.
Crypto	Project research, wallet/on-chain context, token risk notes, market data, contract caution.
DeFi	TVL, protocol metrics, audit status, smart-contract/admin/upgradeability risks.
Prediction markets / market intelligence	Non-advisory planning, oracle/venue research, and risk review workflows where the skill pack is available.

JANEX-AI should frame this as analysis and risk context, not financial advice.

5. Native Multi-Agent + Debatable Decisions

For bigger tasks, JANEX-AI can route work through specialist agents while keeping them inside the native Janex-ai runtime: same provider config, same tools, same timeout behavior, same progress events.

/multiagent        Toggle native multi-agent routing
/depth high        Prefer deeper routing for research/complex engineering tasks
/depth xhigh       Add stronger verification/debate for high-value tasks
/depth max         Force deep routing for research and complex builds/audits
/depth ultra       Maximum depth with final synthesis/review
/fast              Return to low-depth fast single-agent mode

JANEX-AI also ships skill patterns for council-style decisions and team orchestration:

Pattern	Useful when
Research debate	Claims need supporter/skeptic debate, judging, citation checking, and final review.
Council	A decision is genuinely debatable and needs Architect, Skeptic, Pragmatist, and Critic voices.
Team orchestration	Multi-agent work needs ownership, work items, merge gates, handoffs, and evidence.
Parallel execution	Independent lanes can run faster without losing correctness.
Autonomous loops	Long-running pipelines need sequential or DAG-like agent workflows.

What this means in practice:

• code tasks can get implementation + review style passes,
• research tasks can use deeper source gathering and synthesis,
• complex audits can route through verifier/judge-style flows,
• product decisions can expose dissent before choosing,
• tool progress is visible instead of hidden behind a black box.

6. Multi-Platform Gateway

JANEX-AI is not locked to one terminal window. Run it as a bot and keep working from chat apps:

Interface	Best for
Terminal TUI	Coding sessions, repo edits, browser automation, local tools.
Lite mode	Fast prompt-only sessions.
Discord	Server/team workflows and remote task control.
Telegram	Mobile control and personal automation.
WhatsApp	Always-on assistant access through a familiar chat app.

janex gateway

Gateway mode supports per-user sessions, context continuity, command handling, and tool status updates.

7. MCP + Local Skill System

JANEX-AI supports the Model Context Protocol and a large local skill directory.

/mcp              Open interactive MCP manager
/mcp presets      Browse built-in server presets
/mcp catalog      Search MCP server catalog
/mcp connect      Add a server
/mcp reload       Restart MCP servers
/skills           List loaded local skills
/addskills        Enable the skill_loader tool

The local skills/ tree currently contains 330+ skill files, including Bug Hunt, trading/finance, social research, deep research, debate/council workflows, multi-agent orchestration, engineering patterns, testing, deployment, research, and automation skills.

Command Map

CLI entrypoints

janex              # Launch full TUI
janex --lite       # Launch lite prompt mode
janex setup        # Configure provider/model/key
janex gateway      # Start Discord/Telegram/WhatsApp gateway
janex api          # Start local Reddit research API on :3001 (normal CLI/gateway auto-start it)
janex --continue   # Continue latest session
janex --resume ID  # Resume a specific session

Deploy helpers

scripts/deploy.sh        # Linux / generic bash
scripts/deploy-macos.sh  # macOS bash wrapper
pwsh scripts/deploy.ps1  # Windows PowerShell

In-app slash commands

Command	Purpose
/help	Show command help.
/status	Show current model/session/tool status.
/model	View or switch model.
/tools	List available tools.
/context	Show context usage.
/clear	Clear visible chat.
/reset	Reset the current agent context.
/theme	Change TUI theme.
/border	Change border style.
/browserui / /gui	Browser UI helpers.
/mcp	Manage MCP servers.
/github	GitHub integration helpers.
/gmail	Gmail/email integration helpers.
/depth / /effort	Change reasoning/research depth.
/deep-research	Run explicit deep research flow.
/multiagent	Toggle native multi-agent routing.
/fast	Switch back to low-depth fast mode.
/skills / /skill	Browse skills or adjust skill visibility.
/todo	Manage todo items.
/rules	Manage session/project rules.
/sessions	Browse/resume sessions.
/save / /export / /copy	Save, export, or copy output.
/reload-mcp	Restart MCP servers and reload tools.
/permissions	Inspect permission settings.
/exit	Quit.

Configuration

Main config file:

~/.janex/config.yaml

Example:

provider: custom              # openai | anthropic | custom | custom-anthropic
apiStyle: openai              # openai | anthropic | auto
baseUrl: "https://your-endpoint/v1"
apiKey: "..."
model: "your-model"

researchMode: low             # low | medium | high | xhigh | max | ultra
themeName: janex
captchaAudio: "hybrid"

browser:
  proxies:
    - "http://user:pass@host:port"

gateway:
  discord:
    enabled: false
    token: ""
  telegram:
    enabled: false
    token: ""
  whatsapp:
    enabled: false

Environment variables override config values.

Variable	Required	Description	Example
JANEX-AI_PROVIDER	No	Provider type.	openai, anthropic, custom
JANEX-AI_API_KEY	Yes	Main model API key.	sk-...
JANEX-AI_BASE_URL	No	Base URL for OpenAI-compatible or Anthropic-compatible endpoints.	https://your-endpoint/v1
JANEX-AI_API_STYLE	No	Force endpoint protocol if auto-detect fails.	auto, openai, anthropic
JANEX-AI_MODEL	No	Main model name.	gpt-4o, your-model
JANEX-AI_VISION_MODEL	No	Vision model for screenshots/browser tasks.	your-vision-model
SEARCH_API_KEY	No	Optional external search provider key.	...
JANEX-AI_REDDIT_RELAY_URL	No	First-party Reddit relay/API base URL for research_forums.	http://127.0.0.1:3001
JANEX-AI_REDDIT_BACKEND	No	Reddit backend selection.	auto, relay, keyless, scrapecreators, off
REDDIT_CLIENT_ID / REDDIT_CLIENT_SECRET	No	Optional OAuth credentials for the Reddit relay; otherwise it falls back to public RSS.	...
SPOTIFY_ACCESS_TOKEN / SPOTIFY_TOKEN	No	Optional music-control credentials.	...
DISCORD_TOKEN	No	Discord gateway token.	...
TELEGRAM_TOKEN	No	Telegram gateway token.	...

Built-in Tool Categories

Category	Capabilities
File & Code	Read/write/edit/search files, terminal execution, code execution.
Browser	CloakBrowser automation, screenshots, snapshots, CAPTCHA/verification solving, sliders, drag/drop.
Web & Research	Web search, scraping, social research across Reddit/X/YouTube/TikTok/HN/NodeLoc/Xiaohongshu/GitHub/Polymarket, China AI research, forum research, YouTube helpers, report generation.
Git & GitHub	Git operations, PR/issues, releases, CI status, merge-conflict helpers.
Office	PDF, Excel, PowerPoint, email.
DevOps	Docker, VPS, deploy, cloud provisioning.
Security / OSINT	Authorized security testing helpers, CTF skills, OSINT investigation.
Finance / Blockchain	Trading analysis, stocks, crypto, DeFi, prediction-market research, EVM/Solana tooling.
Creative	GIF search, text humanizer, diagrams.
Utility	Maps, notifier, music, todo, memory, archive reader.
MCP	External tool servers loaded through Model Context Protocol.

Architecture

src/
  agent/       Agent loop, context, config, memory, research pipeline, multi-agent runtime
  cli/         OpenTUI React interface, command palette, input/output panels
  gateway/     Discord / Telegram / WhatsApp gateway
  mcp/         MCP client, registry, catalog, tool adapter
  providers/   OpenAI-compatible, Anthropic-compatible, and custom provider adapters
  skills/      Local skill registry and loader
  tools/       Browser, terminal, file ops, git, research, OSINT, office, deploy, cloud, etc.
  utils/       Update check, terminal sanitization, base URL helpers, ASCII logo

skills/
  bug-hunt/    CTF and authorized security-testing skill pack

bin/
  janex.js     CLI launcher

scripts/
  build.cjs    Build script

Development

Command	Description
npm run build	Compile TypeScript with scripts/build.cjs.
npm run dev	Run src/index.tsx through tsx.
npm run start	Run the built app from dist/index.js.
npm run prepare	Build during package preparation.
npm run postinstall	Run package postinstall setup.
npm run lint	Run ESLint on src/**/*.ts.
npm run format	Format TypeScript files with Prettier.

Project facts:

• npm package: janex-ai
• CLI binary: janex
• Runtime target: Node.js/Bun-compatible TypeScript
• Build output: dist/
• Config directory: ~/.janex/
• Browser profile: ~/.janex-browser-profile
• License: MIT

Star the Repo

If JANEX-AI looks useful, give it a ⭐ on GitHub. It helps more developers find the project, try the agent, report bugs, and contribute new tools/skills.

License

MIT