Your Team Gets an AI Workforce That Never Sleeps
We build AI agents that handle your repetitive work — triaging messages, updating systems, generating reports, following up on leads — so your team focuses on what actually matters.
What You Get
Custom AI Agents
Purpose-built agents that understand your business, your tools, and your workflows. Not a generic chatbot.
Always-On Automation
Agents run 24/7 on scheduled tasks, respond to messages instantly, and never miss a step.
Full Visibility
A real-time dashboard shows exactly what your agents are doing, what they've handled, and what needs attention.
How It Works
Every automation follows the same pattern — collect, decide, act:
1. Collect
Agents pull data from your existing tools — email, Discord, forms, CRMs, spreadsheets. No manual data entry.
2. Decide
AI classifies, prioritizes, and routes work based on rules you define. It reads context, not just keywords.
3. Act
Agents update your systems, send messages, create tickets, schedule follow-ups — whatever the workflow needs.
The Dashboard — Your Control Center
You get a live dashboard where you can see everything your AI agents are doing in real time. Monitor health, review sessions, manage tools, and adjust settings — all from one place.
Multiple Agents
Configure separate agents for different roles — each with their own personality, tools, and permissions.
Channel Integration
Connect Discord, Telegram, WhatsApp, Slack, iMessage, and more. Agents respond where your team already works.
Workflows We Build
| Workflow | What the Agent Does | What You Measure |
|---|---|---|
| Lead Intake | Auto-responds, qualifies, updates CRM, books meetings, alerts your team | Response time, lead-to-booked rate |
| Support Triage | Classifies tickets, drafts replies, escalates with full context | First-reply time, resolution speed |
| Operations Ticketing | Standardizes intake, enforces required fields, routes to the right team | Dropped requests, rework rate |
| Reporting | Scheduled rollups, summaries, dashboard refreshes — delivered on time, every time | Hours saved per week |
| Content Assembly | Drafts posts, summaries, and assets from source material | Cycle time per piece |
| Custom Workflows | If your team does it repeatedly, we can probably automate it | Defined together |
Expected Results
For Your Operations
Lower manual workload. More consistent handoffs. Fewer missed steps. Nothing falls through the cracks.
For Your Customers
Faster responses. More reliable experience. Issues get routed and resolved without human bottlenecks.
How We Work Together
Pick One Workflow
We identify the highest-impact repetitive process on your team and define what success looks like.
Build a Focused MVP
We connect your tools, configure the agent, and deploy a working automation — typically within days, not months.
Run, Measure, Optimize
The agent runs live for 2–4 weeks. We track KPIs together and fine-tune until the numbers prove it works.
Expand
Once results are validated, we roll out additional workflows. Each builds on the last.
Is This Right For You?
Great Fit
- Your team does the same tasks repeatedly
- You use tools with APIs (CRMs, ticketing, chat)
- You have clear process owners
- You can define what "done right" looks like
Needs Prep First
- No standardized process exists yet
- Tool access or permissions aren't sorted
- Process ownership is unclear
- Inputs are inconsistent or undocumented
Ready to Get Started?
Tell us the most repetitive workflow slowing your team down. We'll map a focused MVP with scope, timeline, and measurable targets.
Start a ConversationOpenClaw Gateway Dashboard
What is OpenClaw? It's the "brain hub" that connects your AI agents to your chats, files, tools, and schedule. Think of it as mission control for your personal AI workforce.
The Sidebar — Navigation Overview
The left sidebar is your main navigation. Everything is organized into three groups:
| Group | Purpose | Think of it as… |
|---|---|---|
| Control | What's running right now | The cockpit instruments |
| Agent | Your AI workers and their abilities | The crew roster |
| Settings | How everything is configured | The settings menu |
Control Group
Overview
What it is: Your dashboard's homepage — the first thing you see.
What it shows:
| Indicator | What it means |
|---|---|
| Gateway Status | Connected = everything's online. Disconnected = something's wrong. |
| Uptime | How long the gateway has been running without interruption. |
| Instances | How many Claude AI processes are alive right now. |
| Sessions | How many conversation threads are currently active. |
| Cron | Whether your scheduled tasks are enabled or paused. |
When to use it: This is your quick health check. If something seems broken, start here. Green = good. If the gateway shows disconnected, your agents can't do anything.
Channels
What it is: Where you connect messaging platforms so your agents can talk to you (and others).
What you'll see:
- A list of connected platforms (e.g., Discord)
- For each channel:
- Configured / Running — Is the bot actually live and responding?
- Accounts — You can run multiple bot accounts per platform (e.g., different Discord bots for different servers)
- Actions — Toggle what permissions your bot uses (read messages, send messages, moderation, etc.)
Available platforms: Discord, Telegram, WhatsApp, Slack, iMessage, Signal, and more. To add new ones, go to Config > Plugins and enable them.
When to use it: When you want to add a new chat platform, check if a bot is running, or change what your bot is allowed to do on a platform.
Instances
What it is: Shows active Claude AI process instances — the actual "brains" running behind the scenes.
In plain English: This tells you "is the AI engine actually running and connected right now?"
When to use it: Mostly diagnostic. You usually don't need to touch this unless you're troubleshooting why an agent isn't responding. If instances shows 0, no agent can process anything.
Sessions
What it is: Every conversation thread (with you, or triggered by a cron job) creates a session. This page lists them all.
What you can do here:
- See all active and recent sessions
- Inspect a session's full conversation history
- Manually reset a session (clear its memory/context)
- Wipe a session entirely
When to use it: If your agent gets stuck in a weird loop, keeps repeating itself, or seems confused — come here and reset its session. It's like rebooting the conversation without restarting the whole system.
Usage
What it is: Token and cost tracking for your AI API usage.
Why it matters: Every message your agent processes costs API tokens. This page helps you track:
- How many tokens have been used
- Cost breakdown over time
- Which agents or tasks are using the most resources
When to use it: Check this when you want to keep your API bill in check, or figure out if a runaway cron job is burning through tokens.
Cron Jobs
What it is: The scheduler — where you set up tasks that run automatically on a timer.
This is one of the most powerful features in OpenClaw. See the Practical Examples section for detailed setup walkthroughs.
What you can configure per job:
| Field | What it does |
|---|---|
| Name | A label for this job (e.g., "Morning Briefing") |
| Agent ID | Which agent runs it (default = your main agent) |
| Schedule | When it runs — every X minutes/hours, or specific days/times |
| Session | Isolated = clean slate each run (recommended). Shared = continues from last run. |
| Wake mode | Now = starts immediately at schedule time |
| Payload | Agent turn = sends a message/instruction to the agent |
| Agent message | The actual instruction (e.g., "Summarize my inbox") |
| Delivery | Where to send results — Announce summary pushes to your chat |
| Channel | Which messaging channel receives the results |
When to use it: Whenever you want something to happen automatically — daily briefings, weekly file audits, content research, health checks, etc.
Agent Group
Agents
What it is: The roster of your AI personalities/workers. Each agent is a separately-configured AI with its own personality, tools, and purpose.
Each agent has six tabs:
Tab 1: Overview
The agent's identity card.
| Setting | What it means |
|---|---|
| Workspace | The folder on disk this agent uses for its files (e.g., ~/.openclaw/workspace) |
| Primary Model | Which AI model it runs on (e.g., anthropic/claude-opus-4-6) |
| Identity Name | The name the agent knows itself by (e.g., "Ava") |
| Default | If true, this agent answers when no specific one is chosen |
| Identity Emoji | The icon shown next to this agent in the dashboard |
| Skills Filter | Whether this agent can use all skills or only a restricted set |
| Model Fallbacks | If the primary model fails (rate limit, outage), what model to try next |
Tab 2: Files
These are the core memory files that shape how the agent thinks and behaves. Think of them as the agent's "brain programming."
| File | Purpose | Analogy |
|---|---|---|
| IDENTITY.md | Who the agent is and how it presents itself | Its name tag and introduction |
| SOUL.md | The agent's personality, values, and behavioral guidelines | Its personality and moral compass |
| TOOLS.md | Guidance on when and how to use its tools | Its instruction manual |
| AGENTS.md | Defines sub-agents and routing rules | Its org chart |
| USER.md | Facts about you the agent should always know | Its cheat sheet about you |
| HEARTBEAT.md | What to do during scheduled check-ins (cron jobs) | Its standing orders |
| BOOTSTRAP.md | Startup context loaded when the agent first wakes up | Its morning briefing |
| MEMORY.md | Long-term memories accumulated over time | Its journal/diary |
You can click any file to edit it directly in the dashboard.
Tab 3: Tools
Toggle exactly which capabilities this agent has access to. Tools are organized by category:
| Category | Tools | What they do |
|---|---|---|
| Files | read, write, edit, apply_patch | Read and modify files on disk |
| Runtime | exec, process | Run shell commands and manage background tasks |
| Web | web_search, web_fetch | Search the internet and fetch web pages |
| Memory | memory_search, memory_get | Recall information from past conversations |
| Sessions | list, history, send, spawn, status | Manage conversation threads and spawn sub-agents |
| UI | browser, canvas | Control a web browser or use a drawing whiteboard |
| Messaging | message | Send messages to connected chat platforms |
| Automation | cron, gateway | Schedule tasks and control the gateway itself |
| Nodes | node | Access connected remote devices/servers |
| Agents | agents | List and interact with other agents |
| Media | image | Understand and analyze images |
Quick Presets (buttons at the top of the Tools tab):
| Preset | Best for |
|---|---|
| Minimal | Agents that should only read/respond, no file or web access |
| Coding | Development work — files, exec, process, no web/messaging |
| Messaging | Chat-focused — messaging, sessions, memory, no file editing |
| Full | Everything enabled — use for your main/power agent |
Tab 4: Skills
Which third-party skill packs this specific agent can use (out of the available pool). You can restrict certain agents to only certain skills — for example, your content agent gets summarize and web_search, but not exec or github.
Tab 5: Channels
Shows which messaging channels (Discord, Telegram, etc.) route messages to this agent, and the scheduler status for channel-specific settings.
Tab 6: Cron Jobs
Shows scheduled jobs assigned specifically to this agent. Same interface as the global Cron Jobs page, but filtered to just this agent's tasks.
Skills (Marketplace)
What it is: The global skills marketplace — pre-built capability packs that give your agents new superpowers.
Status indicators:
| Status | Meaning |
|---|---|
| Eligible | Requirements met, ready to use right now |
| Blocked | Something's missing — usually a CLI tool needs installing or an API key is needed |
Key skills worth knowing about:
| Skill | What it does | Requirements |
|---|---|---|
github | Create issues, PRs, run CI, manage repos | gh CLI |
gemini | One-shot Q&A using Google Gemini | gemini CLI |
summarize | Summarize YouTube videos, podcasts, URLs | various |
video-frames | Extract frames from videos | ffmpeg |
skill-creator | Build entirely new skills from inside the agent | none |
gog | Google Workspace — Gmail, Calendar, Drive, Sheets, Docs | Google API credentials |
notion | Read/write Notion pages and databases | Notion API key |
obsidian | Work with your Obsidian vault | local Obsidian install |
trello | Manage Trello boards | Trello API key |
weather | Get weather forecasts | none (free) |
tmux | Control terminal sessions | tmux |
openai-whisper-api | Transcribe audio files to text | OpenAI API key |
openai-image-gen | Generate images from text descriptions | OpenAI API key |
healthcheck | Security audit your server | system access |
blogwatcher | Monitor RSS feeds for new posts | none |
spotify-player | Control Spotify playback | Spotify account |
1password | Integrate with 1Password secrets | op CLI |
To install a blocked skill: Click the install button next to it — it'll show you exactly what command to run (e.g., brew install ffmpeg).
Nodes
What it is: Connects your agents to external machines where they can execute commands. This is for advanced setups where you want the AI to run code on a remote server or isolated environment.
Key security settings:
| Setting | Options | What it means |
|---|---|---|
| Security Mode | Deny (default) | Block all shell commands unless explicitly allowed |
| Ask Mode | On miss | If a command isn't on the allowlist, ask you before running it |
| Ask Fallback | Deny | If the UI isn't available to ask you, deny the command |
| Auto-allow Skill CLIs | On/Off | Automatically trust CLI tools that come with installed skills |
Exec Node Binding: Lets you pin specific agents to run their shell commands on specific machines. For example, your "server maintenance" agent only runs commands on your production server.
Settings Group
Config
What it is: The master control panel. Everything configurable lives here. It edits ~/.openclaw/openclaw.json through a safe UI.
The config is organized into these sections:
Updates
| Setting | What it does |
|---|---|
| Release channel | Choose stable, beta, or dev builds |
| Check on startup | Whether OpenClaw looks for updates when it launches |
Gateway (Networking)
| Setting | What it does |
|---|---|
| Port | What port the dashboard runs on |
| Bind | Who can access it: loopback (only this machine), lan (your network), tailnet (Tailscale VPN), or custom |
| Auth | Token or password protection, plus Tailscale integration |
| Control UI Base Path | Serve the dashboard at a custom URL path |
Agents (Global Defaults)
These apply to all agents unless overridden per-agent:
| Setting | What it does |
|---|---|
| Primary Model | Default AI model |
| System Prompt Mode | default or safeguard (extra safety guardrails) |
| Context Pruning | When conversations get too long, how to trim them |
| Heartbeat | Scheduled proactive check-ins; set a delivery target to push summaries to Discord/Telegram/etc. |
| Human Delay Mode | natural adds realistic typing delays to messages (feels less robotic) |
Authentication
| Setting | What it does |
|---|---|
| API Keys | Add keys for different AI providers (Anthropic, OpenAI, Google, etc.) |
| Auth Profiles | Named profiles so you can switch between accounts |
| Failover | If one API key fails (billing issue, rate limit), automatically try the next one |
Messages
How your agent handles incoming chat messages:
| Setting | Options | What it means |
|---|---|---|
| Queue Mode | steer / queue / collect / interrupt | Per-channel: process immediately, line up in order, batch together, or let new messages interrupt current work |
| Ack Reaction | Any emoji | The emoji the bot reacts with to confirm it received your message |
| Inbound Debounce | Milliseconds | Wait this long before processing, to batch rapid-fire messages |
Commands
Slash commands you can type in chat:
| Command | What it does | Default |
|---|---|---|
/config | Show/change config from chat | Off |
/debug | Show debug info | Off |
/bash | Run shell commands from chat | Off |
/restart | Restart the gateway | Off |
| Command Owners | Who can use owner-only commands | — |
Session (Conversation Memory)
| Setting | Options | What it means |
|---|---|---|
| DM Session Scope | main / per-peer | main = one continuous history. per-peer = each person gets their own isolated conversation. |
| Reset Triggers | Idle time / Daily hour | Automatically clear session context after inactivity or at a set time each day |
Tools (Global Policy)
| Setting | What it does |
|---|---|
| Exec Security | deny, allowlist, or full — global shell command policy |
| Exec Ask | off, on-miss, always — when to ask permission before executing |
| Web Search | Configure Brave Search or Perplexity API keys |
| Web Fetch | Lightweight HTTP fetching (cache TTL, max chars, timeout) |
Memory
| Setting | What it does |
|---|---|
| Enable Memory Search | Turns on semantic search over MEMORY.md files |
| Search Provider | openai, gemini, voyage, or local embeddings |
| Hybrid Search | Combines keyword search (BM25) with vector search for better recall |
| Session Indexing | Experimental: also search through past conversation transcripts |
Other Config Sections
| Section | What it does |
|---|---|
| Cron | Global scheduler on/off, max concurrent job runs |
| Logging | Console log level (silent through trace), file logging, sensitive data redaction |
| Browser | How the agent drives a web browser (headless mode, CDP URL, browser profiles) |
| TTS | Text-to-speech output via ElevenLabs, OpenAI TTS, or Edge TTS |
| Plugins | Enable/disable channel plugins (Telegram, Discord, WhatsApp, iMessage, LanceDB memory, voice calls, etc.) |
Debug
What it is: A live diagnostic view showing what's happening under the hood in real time.
When to use it: When something isn't working and you need to see the raw events, API calls, and internal state. This is your "look under the hood" tool.
Logs
What it is: A real-time log viewer showing system messages as they happen.
When to use it: For ongoing monitoring or troubleshooting. You can adjust how much detail appears here via Config > Logging (from silent to trace).
Practical Examples
Building a File Cleanup Agent
Goal: An agent that organizes, sorts, and archives files on a schedule.
Step 1 — Create the agent
Go to Agents and create a new agent workspace directory, or edit AGENTS.md in your main agent's Files tab to define a sub-agent.
Step 2 — Configure its Tools (Tools tab)
- Enable:
read,write,edit,exec,process - Disable:
web_search,web_fetch,browser,canvas,message, all session tools - Start from the Coding preset, then disable web tools
Step 3 — Write its personality (Files tab)
SOUL.md:
You are a file organization specialist. Your job is to clean up, sort, deduplicate, and organize files. You do not browse the web or send messages. You focus only on the local filesystem.
USER.md:
Work within these directories:
~/Downloads,~/Documents,~/Desktop. Archive destination:~/Archive/.
Step 4 — Schedule it (Cron Jobs tab)
- Schedule: Weekly, Mondays at 9am
- Session: Isolated
- Message: "Review ~/Downloads and move files older than 30 days to appropriate archive folders. Summarize what you moved."
- Delivery: Announce summary to your Discord channel
Building a YouTube Ideas Agent
Goal: An agent that researches trends and generates video ideas.
Step 1 — Create the agent
Create a new agent workspace, name it something like "content-brain."
Step 2 — Configure its Tools (Tools tab)
- Enable:
web_search,web_fetch,read,write,memory_search,memory_get - Install the summarize skill (for YouTube video summaries)
- Install the github skill if you want it to create issue cards for ideas
Step 3 — Write its personality (Files tab)
SOUL.md:
You are a YouTube content strategist. When asked for ideas, research trending topics in [your niche], check past ideas in memory to avoid repeats, and produce a formatted list of video concepts with titles, hooks, and outline points.
USER.md:
Channel niche: [your niche]. Audience: [your audience]. Style: [your style]. Past videos: [list or link]. Save ideas to ~/content-ideas.md.
Step 4 — Schedule it (Cron Jobs tab)
- Schedule: Weekly, Fridays at 10am
- Session: Isolated
- Message: "Research trending topics in [niche] this week and add 5 fresh video ideas to ~/content-ideas.md with titles, hooks, and outlines."
- Delivery: Announce summary to your preferred channel
Setting Up Cron Jobs
The power move. Cron jobs turn your agents from reactive (you ask, they answer) into proactive (they work on their own schedule).
Example jobs to create:
| Job Name | Schedule | Agent Message |
|---|---|---|
| Morning Briefing | Daily at 8:00 AM | "Give me a summary of what I should focus on today based on my notes and memory" |
| File Audit | Weekly, Monday | "List all files in ~/Downloads older than 2 weeks and suggest what to delete or archive" |
| Content Ideas | Weekly, Friday | "Generate 5 YouTube video ideas for my channel based on recent trends" |
| Weekly Review | Sunday at 6:00 PM | "Review what I accomplished this week from session history and suggest improvements" |
- Always use Isolated sessions — Prevents one run from contaminating the next with leftover context
- Write your HEARTBEAT.md — This file is what the agent reads during scheduled runs. Put standing instructions there like "always check ~/todo.md first"
- Be specific in your messages — "Summarize my inbox" is vague. "Check Gmail for unread emails from the last 24 hours, prioritize anything from work contacts, and list action items" is actionable.
- Set delivery channels — Always route results to a channel you'll actually see (Discord DM, Telegram, etc.)
Quick Reference
Concepts Cheat Sheet
| Term | Plain English |
|---|---|
| Gateway | The main OpenClaw server process that coordinates everything |
| Instance | A running Claude AI process |
| Session | A conversation thread with history |
| Channel | A connected messaging platform (Discord, Telegram, etc.) |
| Agent | A configured AI worker with its own personality and tools |
| Skill | A pre-built capability pack (GitHub integration, summarizer, etc.) |
| Node | A remote machine the agent can execute commands on |
| Tool | A low-level capability (read files, run commands, search web) |
| Cron Job | A scheduled task that runs automatically |
| SOUL.md | The file that defines an agent's personality |
| HEARTBEAT.md | The file an agent reads during scheduled/proactive check-ins |
Tools vs Skills — What's the Difference?
| Tools | Skills | |
|---|---|---|
| What | Low-level capabilities | Higher-level integrations |
| Examples | read, write, exec, web_search | GitHub, Notion, Summarize, Spotify |
| Built-in? | Yes, always available | Installed separately |
| Dependencies | None | Often need a CLI tool or API key |
| Configured where | Agent > Tools tab | Agent > Skills tab |
You need both layers working together for full capability. Tools are the hands; skills are the specialized instruments those hands can hold.
Troubleshooting Checklist
| Problem | Where to look |
|---|---|
| Agent not responding | Overview — Is the gateway connected? Instances — Are there any running? |
| Bot offline in Discord/Telegram | Channels — Is the channel showing "Running"? |
| Cron job not firing | Cron Jobs — Is it enabled? Check the schedule. Config > Cron — Is the global scheduler on? |
| Agent acting confused | Sessions — Reset the session to clear bad context |
| Skill not working | Skills — Check if it shows "Blocked" and install the missing dependency |
| High API costs | Usage — Check which agent/task is burning tokens |
| Commands not working in chat | Config > Commands — Are they enabled? Are you listed as an owner? |
| Agent can't run shell commands | Nodes — Check exec security mode. Is it set to Deny? |
Key File Locations
| File | Path |
|---|---|
| Main config | ~/.openclaw/openclaw.json |
| Agent configs | ~/.openclaw/agents/{name}/ |
| Workspace | ~/.openclaw/workspace/ |
| Memory data | ~/.openclaw/memory/ |
| Cron configs | ~/.openclaw/cron/ |
| Exec approvals | ~/.openclaw/exec-approvals.json |
Setting Up Model Fallbacks (Recommended)
In Config > Agents > Model Fallbacks, configure backup models so your agents don't fail silently when hitting rate limits:
- Primary:
anthropic/claude-opus-4-6 - Fallback 1:
anthropic/claude-sonnet-4-6 - Fallback 2:
google/gemini-2.5-pro
This way, if Claude hits a rate limit, the agent automatically switches to the next model and keeps working.