English | 한국어

tabyAgent

A more autonomous, more persistent, and easier alternative to OpenClaw/Hermes.

Give it a task and it will do it — even if it takes hours, even if things go wrong.

Memory, skills, self-improvement, scheduled tasks, todos, web browsing, GUI apps, and multiple agents work out of the box. No extra setup.

What it can do

• Everyday chat: Get answers in Telegram. Almost every format is supported — text, images, files, and more.
• Inference providers: Connect OpenAI, OpenRouter, Upstage, OrcaRouter, Synthetic, Ollama (local and Cloud), ZenMux, Codex OAuth, Grok OAuth, GitHub Copilot OAuth, or your own API endpoint.
• Skills, MCP: Add the capabilities and tools you want to the agent. Even if you don't install anything yourself, the agent finds and installs what it needs.
• Scheduled tasks: Recurring jobs run on a schedule, report when they finish, and skip when they are not needed.
• Todos: Manages todo lists for both you and the agents. You manage yours with /todo or the approve buttons in chat, while agents automatically watch the list, schedule work, and handle your tasks for you.
• Multiple agents: Create specialist agents for different roles — each with its own memory, persona, and Telegram topic — and have them work together.
• Run it anywhere: Docker container or local Node.js. Native support is macOS and Linux. Windows can work through Docker, but that is not guaranteed.
• Self-improvement: tabyAgent can improve itself. It learns from how problems were solved and from your corrections, and it gets sharper the more you use it. It also learns the things you regularly do, and can remind you when you forget.

Differences

• Differences between tabyAgent and tabyBot: (1) tabyAgent runs on Telegram while tabyBot runs in its own web UI. (2) Both share the same core features — pick whichever interface fits you better.

Feature	tabyAgent	Grok Bot	OpenClaw	Hermes	ChatGPT (Chat)
Everyday chat	✅ Yes	✅ Yes	✅ Yes	✅ Yes	✅ Yes
Multiple agents	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ No
Search	✅ Yes	✅ Yes	✅ Yes	✅ Yes	✅ Yes
Multiple providers	✅ Yes	❌ No	✅ Yes	✅ Yes	❌ No
Smart todo list	✅ User + agents	❌ No	❌ No	❌ No	❌ No
Proactive outreach	✅ Yes	❌ No	❌ No	❌ No	❌ No
Skills	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ No
MCP	✅ Yes	✅ Yes	✅ Yes	✅ Yes	✅ Dev mode
Scheduled tasks	✅ Yes	✅ Yes	✅ Yes	✅ Yes	✅ Yes
Self-improvement	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ No
Terminal	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ Sandbox only
Browser	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ No
GUI apps	✅ Yes	✅ Yes	✅ Yes	✅ Yes	❌ No
Local execution	✅ Yes	❌ No	✅ Yes	✅ Yes	❌ No
Persistence	✅ Disk	❌ No	❌ No	❌ No	❌ No
Refusal	✅ Never	✅ Never	❌ Sometimes	❌ Sometimes	❌ Sometimes
Memory usage	✅ ~800MB	✅ Unknown	❌ ~10GB	❌ ~2GB	✅ Cloud
NSFW level	✅ Allow, indirect only, block	❌ Not available	❌ Not available	❌ Not available	❌ Block
License	✅ AGPL-3.0	❌ Proprietary	✅ MIT	✅ MIT	❌ Proprietary

Tests used the Ollama Cloud provider and the Kimi-K2.6 model.

Example prompts

• "What's last month's DigitalOcean bill?"
• "Find every file in Documents that contains the word report and summarize them all."
• "Run gemma4 e4b and e2b on this machine yourself, then compare them."
• "Organize next week's schedule."
• "Delete the sales row in this Excel file and highlight the columns that have values."
• "Cancel my ChatGPT Plus subscription for me."
• "Is this actually real? (X/Reddit link)"
• "Summarize this (long pasted text)."
• "I'll give you my email account so you can use it later. Address is (), SMTP/POP3 password is ()."
• "Sign up for Spotify with the email account I told you about earlier."
• "Visualize this paper so it's easy to understand. (paper link)"
• "Compare privatestater analytics and privatestater captcha with Google Analytics and reCAPTCHA."
• "What's a privacy-respecting Gmail alternative?"
• "What should I eat for lunch in a bit? I have 4,328 won in my account."
• "Every morning at 8, send me today's weather and my todos."
• "Let me know when a new one drops. (manga/channel link)"
• "If there's anything on my todo list you can handle, just do it."
• "I study Japanese around 9pm every night — nudge me if I forget."
• "What was that thing we talked about before?"
• "Make a translation specialist bot and have it translate this whole document."
• "That thing you did earlier was wrong — do it this way from now on."
• "Take as long as you need — organize every photo in this folder by date."
• "Before I get off work, summarize what I did today."

Quick start

Option A: Installer script

1. Create a Telegram bot

1. Open Telegram and chat with @BotFather.
2. Send /newbot and follow the instructions.
3. Copy the bot token.

2. Install (Linux / macOS)

Paste the line below into a terminal and press Enter. Installation can take a while, so wait for it to finish.

You can choose Docker or a local run. Docker is recommended for security and isolation.

The automated script does not support Windows. If you are on Windows, consider switching your main OS to a Linux-based distribution. In many cases it is faster, more privacy-friendly, and more freedom-preserving. Requirements: If you pick Docker, everything you need is installed for you. If you pick local, Node.js 22 or later must already be installed.

curl -fsSL https://raw.githubusercontent.com/gpdir16/tabyAgent/main/scripts/install.sh | bash

To update tabyAgent later, run the same command again. Settings and memory are kept.

The installer also adds a tabyagent command to manage the instance: tabyagent start, stop, restart, status, logs, approve, and uninstall.

3. Set up in Telegram

1. Send /start to your new bot.
2. The setup wizard walks you through language, LLM provider, API key, and model.
3. When setup is done, you can start chatting.

You can change language, model, thinking level, and more with /config at any time. Use /todo for the shared todo list and /agents to manage additional agents.

Option B: Docker Compose (not recommended)

git clone https://github.com/gpdir16/tabyAgent.git
cd tabyAgent
cp .env.example .env # Edit .env and set TELEGRAM_BOT_TOKEN
docker compose up -d

Third-party licenses

Project	Version	License	Source
grammy	1.46.0	MIT	Repository
@modelcontextprotocol/sdk	1.30.0	MIT	Repository
js-tiktoken	1.0.21	MIT	Repository
node-cron	3.0.3	ISC	Repository
openai	4.104.0	Apache-2.0	Repository
prettier	3.8.3	MIT	Repository
camofox-browser	2.4.7	MIT	Repository
Camoufox	—	MPL-2.0	Repository
Playwright	—	Apache-2.0	Repository

Transitive npm dependencies

• Apache-2.0: openai
• BSD-2-Clause: json-schema-typed, webidl-conversions
• BSD-3-Clause: fast-uri, qs
• ISC: inherits, isexe, node-cron, once, setprototypeof, which, wrappy, zod-to-json-schema
• MIT: @grammyjs/types, @hono/node-server, @modelcontextprotocol/sdk, @types/node, @types/node-fetch, abort-controller, accepts, agentkeepalive, ajv, ajv-formats, asynckit, base64-js, body-parser, bytes, call-bind-apply-helpers, call-bound, combined-stream, content-disposition, content-type, cookie, cookie-signature, cors, cross-spawn, debug, delayed-stream, depd, dunder-proto, ee-first, encodeurl, es-define-property, es-errors, es-object-atoms, es-set-tostringtag, escape-html, etag, event-target-shim, eventsource, eventsource-parser, express, express-rate-limit, fast-deep-equal, finalhandler, form-data, form-data-encoder, formdata-node, forwarded, fresh, function-bind, get-intrinsic, get-proto, gopd, grammy, has-symbols, has-tostringtag, hasown, hono, http-errors, humanize-ms, iconv-lite, ip-address, ipaddr.js, is-promise, jose, js-tiktoken, json-schema-traverse, math-intrinsics, media-typer, merge-descriptors, mime-db, mime-types, ms, negotiator, node-domexception, node-fetch, object-assign, object-inspect, on-finished, parseurl, path-key, path-to-regexp, pkce-challenge, prettier, proxy-addr, range-parser, raw-body, require-from-string, router, safer-buffer, send, serve-static, shebang-command, shebang-regex, side-channel, side-channel-list, side-channel-map, side-channel-weakmap, statuses, toidentifier, tr46, type-is, undici-types, unpipe, uuid, vary, web-streams-polyfill, whatwg-url, zod

License

AGPL-3.0