Abacus Claw — How To Documentation
Your guide to unlocking the full power of Abacus Claw.
Browse the sections below to explore everything Abacus Claw can do — and how to use it effortlessly.
Documentation Sections​
- What Is Abacus Claw?
- Features
- Channel Configurations
- What Abacus Claw Provides
- Billing
- Bring Your Own API Key (BYOK)
- Quick Start
What Is Abacus Claw?​
Abacus Claw is the hosted, managed version of OpenClaw – an open-source platform for building always-on AI agents that run in the cloud, connect to the tools and channels you already use, and keep working even when you close your browser.
Unlike one-shot AI chatbots that reset after every conversation, Claw gives your agent:
- Persistent memory: It remembers context across sessions and channels.
- A full cloud computer: A Linux environment with terminal access, a browser, and persistent file storage.
- Multi-channel presence: Your agent can talk to you (and your users) on WhatsApp, Telegram, and Slack simultaneously.
- Background automation: Schedule recurring tasks with cron jobs so your agent works while you sleep.
- Tool integrations: Connect Slack, Notion, GitHub, Gmail, Google Calendar, and more via built-in tools.
You can start using Claw right now at claw.abacus.ai — no local setup required.
Features​
Claw's power comes from five core capabilities: Persistent Memory, Persistent Operation, Cron Jobs, Channels, and Tools. Each one extends what your agent can do beyond a simple chat window.
Persistent Memory​
Your agent maintains continuous memory across all conversations, channels, and sessions. It remembers what you've told it before, tracks ongoing projects, and builds context over time — so you never have to repeat yourself.
This isn't just chat history. Your agent actively uses its memory to:
- Recall preferences and past decisions
- Track long-running tasks and follow up proactively
- Connect information across different conversations and channels
- Build a personalized understanding of your work and goals
Example prompts​
Remember that I'm working on a product launch for "EcoBottle" scheduled for April 15th.
Track all related tasks, deadlines, and decisions we discuss. Remind me a week before
launch to finalize the press kit.
Keep track of all the books I mention wanting to read. When I ask "what should I read next?"
suggest something from that list based on what I've been working on lately.
You're my personal knowledge assistant. Whenever I share an interesting article, insight,
or idea in our conversations, save it to your memory and organize it by topic. When I'm
working on something related, proactively surface relevant things I've mentioned before.
Persistent Operation​
Your agent runs 24/7 in the cloud — it doesn't sleep, doesn't need your laptop to be open, and doesn't stop working when you close the browser. This means it can:
- Respond to messages on WhatsApp, Telegram, or Slack even when you're offline
- Execute scheduled tasks at any time of day or night
- Monitor external services and trigger actions in real-time
- Maintain continuous workflows that span hours, days, or weeks
The "always on" nature makes your agent a true autonomous assistant, not just a chatbot you interact with during active sessions.
Example prompts​
Monitor my GitHub repository for new issues. Whenever someone opens an issue,
immediately post a summary in our #engineering Slack channel and add it to
our Notion project tracker — even if I'm asleep.
Act as my always-available personal assistant on WhatsApp. I should be able to
message you at any time — morning, night, weekends — and get immediate responses.
Remember our ongoing conversations and pick up where we left off.
Run a continuous monitoring loop: every 30 minutes, check if there are any new
emails from VIP clients in my Gmail. If you find one, immediately notify me on
Telegram with the sender, subject, and a one-sentence summary.
Cron Jobs​
Cron jobs let you schedule your agent to perform tasks automatically on a recurring basis — daily briefings, weekly reports, periodic data pulls, or anything else that benefits from automation.
You set up cron jobs by telling your agent what to do and when. It handles the scheduling natively.
Example prompts​
Every morning at 8 AM, check my Gmail for any new emails from clients,
summarize them, and send me a digest on Telegram.
Every Monday at 9 AM, pull the top 5 trending topics from Hacker News
and post a summary to the #engineering channel in Slack.
At 6 PM every day, compile a personalized newsfeed for me. Use my Slack
messages and starred Gmail threads as signals for what topics I care about,
then send the newsfeed to me on Telegram.
Channels​
Channels are the messaging platforms your agent can send and receive messages on. Once connected, your agent becomes reachable on that platform — it can respond to DMs, participate in group chats, and push proactive notifications.
Here are some common channels you can begin with:
| Channel | Use Case |
|---|---|
| Personal assistant, customer support bots, group agents | |
| Telegram | Personal workflows, team bots, forum-based agents |
| Slack | Workspace automation, team assistants, channel bots |
Example prompts​
Set up a WhatsApp customer support bot for my Airbnb rental business.
It should answer common questions about check-in times, house rules,
Wi-Fi passwords, and parking — using the FAQ document I'll upload.
When it can't answer something, it should collect the guest's name and
question and notify me on Telegram.
Connect to my Telegram account and act as my personal assistant.
When I message you on Telegram, remember the context of our past
conversations. Help me manage to-dos, set reminders, and answer
questions using the files stored in my Claw computer.
Become my always-on personal WhatsApp AI agent. Maintain persistent memory
of all our conversations. Help me brainstorm, keep track of ideas I mention
in passing, and proactively remind me about things I said I'd follow up on.
Tools​
Tools are integrations that give your agent access to external services — not as a chat channel, but as a data source or action target. Your agent can read from, write to, and interact with these services programmatically.
Here are some common tools you can begin with: Slack, Notion, GitHub, Gmail, Google Calendar, Google Drive, and more.
Example prompts​
Connect to my Notion workspace and GitHub. Whenever I merge a PR on GitHub,
automatically create a changelog entry in my Notion "Release Notes" database
with the PR title, description, and date.
Act as my content repurposing assistant. When I drop a blog post draft into
my Google Drive "Content" folder, automatically:
1. Turn it into a Twitter/X thread (punchy, conversational)
2. Rewrite it as a LinkedIn post (professional tone)
3. Save both versions to a Notion database called "Content Pipeline"
Connect to my Gmail and Slack. Monitor my Gmail for any emails from
@investors.com and immediately post a notification in the #fundraising
Slack channel with the sender, subject, and a one-line summary.
Channel Configurations​
Each channel has its own configuration options that control who can talk to your agent, how it behaves in groups, and how messages are routed. Below is a detailed breakdown for each supported channel.
WhatsApp​
WhatsApp integration turns your agent into a WhatsApp contact that can handle DMs and group conversations.
Key settings​
| Setting | Description |
|---|---|
| selfChatMode | When enabled, your agent operates in "self-chat" mode — it reads and responds in your personal WhatsApp chat with yourself. Useful for a private AI assistant that lives in your own chat. |
| dmPolicy | Controls who can DM your agent. Options: pairing (one-time pairing code), allowlist (only approved numbers), open (anyone can message), disabled (DMs off). |
| allowFrom | A list of phone numbers (with country code) allowed to DM the agent. Only used when dmPolicy is set to allowlist. |
| groupPolicy | Controls how the agent handles groups. Options: allowlist (only specified groups), open (all groups), disabled (no groups). |
| groupAllowFrom | A list of group JIDs (WhatsApp group identifiers) the agent is allowed to participate in. Used when groupPolicy is allowlist. |
| groups | Per-group configuration. Each group entry uses its JID as the key, and can have its own allowFrom list to restrict which group members can interact with the agent. |
| requireMention | When enabled, the agent only responds in groups when explicitly @mentioned. Prevents the agent from replying to every message. |
Tips​
- Finding group JIDs: Group JIDs look like
120363xxxxx@g.us. You can ask your agent to list the groups it's in after connecting. - Session isolation: Each group gets its own isolated conversation session, so context doesn't bleed between groups.
- Start with
allowlistfor both DM and group policies while testing, then open up access once you're confident in the agent's behavior.
Telegram​
Telegram integration works through a Telegram Bot created via @BotFather. Your agent controls the bot and responds to messages sent to it.
Key settings​
| Setting | Description |
|---|---|
| botToken | The API token from BotFather. This is required to connect your agent to Telegram. |
| dmPolicy | Controls who can DM the bot. Options: allowlist, open, disabled. |
| allowFrom | A list of numeric Telegram user IDs allowed to DM the bot. Used when dmPolicy is allowlist. (Note: these are numeric IDs, not usernames.) |
| groups | Configuration for group chats the bot participates in. Each group can have its own settings. |
| requireMention | When enabled, the bot only responds in groups when @mentioned by name. |
| Privacy mode | Telegram bots have a privacy mode that limits which messages they can see in groups. If your agent needs to read all group messages (not just commands and mentions), you must disable privacy mode in BotFather settings. |
Tips​
- Getting your Telegram user ID: Send a message to @userinfobot on Telegram — it will reply with your numeric ID.
- Forum topics: Telegram forum-style groups (with topics) are supported. Each topic can be treated as a separate conversation context.
- Bot setup steps: Open Telegram → message
@BotFather→ send/newbot→ follow prompts → copy the token → paste it into Claw's Telegram channel config.
Slack​
Slack integration connects your agent to a Slack workspace where it can respond in channels, DMs, and threads.
Claw supports two connection modes:
| Mode | Description |
|---|---|
| Socket Mode | Recommended for most users. Uses a WebSocket connection — no public URL needed. Requires an App-Level Token (xapp-...) and a Bot Token (xoxb-...). |
| HTTP Mode | Uses a public URL endpoint for Slack to send events to. More complex to set up but needed for certain enterprise configurations. Requires a Bot Token and a configured Request URL. |
Key settings​
| Setting | Description |
|---|---|
| appToken | App-Level Token (xapp-...). Required for Socket Mode. Generated in your Slack app's settings under "Basic Information → App-Level Tokens." |
| botToken | Bot User OAuth Token (xoxb-...). Required for both modes. Found in "OAuth & Permissions" in your Slack app settings. |
| channelPolicy | Controls which channels the bot operates in. Options: allowlist (only specified channels), open (all channels it's added to), disabled. |
| requireMention | When enabled, the bot only responds when @mentioned. Useful in busy channels. |
| Per-channel user allowlists | For fine-grained control, you can specify which users are allowed to interact with the bot on a per-channel basis. |
Tips​
- Threading: The agent responds in threads by default to keep channels tidy.
- Slash commands: You can configure custom slash commands (e.g.,
/ask) that route directly to your agent. - Socket Mode is easiest: Unless you have a specific reason to use HTTP mode, start with Socket Mode — it requires no public URL and works behind firewalls.
- Scopes: Make sure your Slack app has the necessary OAuth scopes:
chat:write,channels:history,groups:history,im:history,app_mentions:read, andconnections:write(for Socket Mode).
What Abacus Claw Provides​
When you subscribe to ChatLLM Teams Pro, you're not just getting Abacus Claw — you're getting access to the most powerful suite of AI agents on a single platform.
The Complete Agent Bundle​
With Pro subscription, you get:
DeepAgent by Abacus AI — one of the most capable AI agents available today, ready to:
- Build full-stack web applications and mobile apps from scratch
- Create professional presentations and reports
- Set up scheduled tasks and event-triggered workflows
- Write, debug, and refactor code across any programming language
- Conduct comprehensive research and data analysis
- Automate complex multi-step workflows
Abacus Claw — the secured, hosted version of OpenClaw with:
- A more user-friendly interface designed for ease of use
- Pre-configured AI models ready to use immediately — no separate API key purchase required
- Enterprise-grade security and reliability
- Seamless integration with your existing tools and channels
- All of this at a low introductory price of just $20/month
You don't need to choose between different agents or manage multiple subscriptions. Everything works together on one platform, with one login, and one simple billing structure.
When you create a session in Claw, you get access to a full suite of capabilities — not just a chat window.
Chat Interface​
The main way to interact with your agent. Type natural language instructions and your agent will carry them out — whether that's writing code, configuring channels, searching the web, or managing files.
The chat interface supports rich responses including formatted text, code blocks, images, and file links.
Sessions​
Each session is a self-contained conversation with its own context and memory.
- Active session: You can chat, give instructions, and your agent will act on them.
- Past sessions: You can open previous sessions to review past conversations, but you cannot continue chatting in them. All new interactions must happen in a new session.
Past sessions are read-only. If you need to pick up where you left off, start a new session — your agent's persistent memory carries forward automatically.
Terminal​
Full terminal access to a Linux environment running on your agent's cloud computer. You can:
- Run shell commands and scripts.
- Install packages with
apt,pip,npm, etc. - Manage files, run servers, and execute code.
Computer​
Your agent has access to a full cloud-hosted computer with a browser. This enables:
- Web browsing and navigation
- Interacting with web applications
- Running automated workflows that require a browser environment
- Executing tasks that need a persistent computing environment
File Storage​
Each agent gets persistent file storage that remains available across sessions. Files persist over time, so your agent can:
- Store documents, datasets, and configuration files.
- Reference uploaded files in future conversations.
- Build up a knowledge base over time.
OpenClaw Dashboard​
The OpenClaw Dashboard is your visual control center for managing everything your agent does. If you prefer a UI-based system over chat commands, the dashboard is where you'll spend your time.
What you can do in the dashboard​
Live Monitoring & Sessions:
- Live session streaming — watch your agent's conversations in real-time across all channels
- Active sessions view — see all currently running sessions with model usage, token counts, and context percentage
- Session history — browse past sessions, compare any two sessions side-by-side, and review conversation logs
- Real-time updates — the dashboard auto-refreshes every 5 seconds to show the latest activity
Cron Job Management:
- View all scheduled jobs with their status, schedule, last run, next run, and duration
- Enable/disable jobs with one click
- Manually trigger jobs for testing
- Monitor job execution history and costs
Channel Settings UI:
- Configure WhatsApp, Telegram, and Slack channels through a visual interface
- Set DM policies, allowlists, and group configurations without editing config files
- Test channel connections and view connection status
Tool Management:
- See all connected tools (Notion, GitHub, Gmail, Slack, etc.)
- Add new tool integrations through the UI
- Review tool usage and API call logs
- Manage authentication and permissions for each tool
The dashboard is your at-a-glance command center — always open it to check on your agent's activity, review what's running, and manage your setup visually.
Billing​
Get all of this for just $20/month — and that's not all. With your ChatLLM Teams Pro subscription, you get access to DeepAgent by Abacus AI, one of the most powerful AI agents available today, plus Abacus Claw in its most user-friendly and secure managed form.
What's Included​
| Details | |
|---|---|
| Price | $20 / user / month |
| Credits | 25,000 credits per month |
| What you get | Full access to DeepAgent + Abacus Claw + all ChatLLM Teams Pro features |
How Credits Work in Claw​
There are two types of credit usage:
-
Hosting costs — Your agent's cloud computer costs 1 credit per 5 minutes while it is actively running. The computer automatically shuts down after a period of inactivity to stop consuming credits.
-
AI model usage — Each LLM call (when your agent thinks, plans, or generates a response) is billed separately based on the model used and the number of tokens processed. This is the same billing model as other ChatLLM features.
The 25,000 monthly credits cover both hosting and model usage combined.
For full billing details, visit the Claw FAQ.
Bring Your Own API Key (BYOK)​
If you want to use your own LLM API key (e.g., from OpenAI, Anthropic, or another provider), you can tell your agent to use it instead of consuming Abacus credits for model calls.
How it works​
- Mention your API key in chat — tell your agent which model provider (e.g., "OpenAI", "Anthropic", "Google") and the corresponding API key to use.
- Claw routes LLM calls through your key — all model usage is billed to your own API account instead of deducting from your Abacus credits.
- Hosting credits still apply — the cloud computer still costs 1 credit per 5 minutes regardless of which API key is used for model calls.
Always specify the correct name of the model provider along with the key so it can be configured properly.
Quick Start​
Ready to get started? Here's the fastest path:
- Go to claw.abacus.ai and log in.
- Start a new session and tell your agent what you want it to do in plain English.
- Connect a channel: Try linking your Telegram or Slack first (they're the easiest to set up).
- Set up a cron job: automate something simple, like a daily weather briefing.
- Explore tools: Connect Notion, GitHub, or Gmail to unlock cross-app workflows.
Your agent learns and adapts as you use it. The more context you give it, the better it gets.
Need more help? Visit the OpenClaw documentation or check the Claw FAQ.