Tips & Help

Everything you need to get the most out of your AI agent. From setup to advanced cost-saving strategies.

🚀

Getting Started

🤖

Choosing Models

💰

Saving Money

💬

Talking to Your AI

Command Reference

FAQ

Getting Started

New to OCP? Here's everything you need to know.

What is OCP?

OCP gives you a personal AI agent that runs on its own dedicated server. You chat with it through Telegram, and it's available 24/7. Think of it as your own private ChatGPT that never sleeps, can remember your conversations, run tasks in the background, browse the web, and much more.

Setup Overview

The whole process takes about 15 minutes:

1

Create an account and pick a plan

Choose the server size that fits your needs.

2

Connect your AI providers (OpenAI + Anthropic)

Your agent uses these to think. You pay the AI providers directly for usage.

3

Create a Telegram bot

This is how you'll talk to your AI. Free, takes 60 seconds.

4

Connect your bot and start chatting

Enter the connection code and you're live.

Getting Your AI Access Keys

Your AI agent needs access to AI models. Here's how to get your keys:

OpenAI

  1. 1. Go to platform.openai.com/api-keys
  2. 2. Create a free account if you don't have one
  3. 3. Add a payment method (Settings > Billing)
  4. 4. Click "Create new secret key"
  5. 5. Copy the key (starts with sk-)

Most people spend $5-15/month on OpenAI usage.

Anthropic

  1. 1. Go to console.anthropic.com
  2. 2. Create a free account if you don't have one
  3. 3. Add a payment method (Settings > Plans)
  4. 4. Go to API Keys in the sidebar
  5. 5. Create and copy your key (starts with sk-ant-)

Anthropic usage typically costs $5-20/month.

Creating Your Telegram Bot

Telegram bots are free and take about 60 seconds to create:

  1. 1. Open Telegram and search for @BotFather (or tap this link). BotFather is Telegram's official tool for creating bots.
  2. 2. Type /newbot and send it.
  3. 3. BotFather asks for a display name. This is what people see (e.g., "Sarah's AI Assistant").
  4. 4. Then it asks for a username. This must end in "bot" (e.g., "SarahAIBot"). This is how people find your bot.
  5. 5. BotFather sends you an access code (long string of letters and numbers). Copy the whole thing and paste it in your OCP dashboard.

Tip: Using a phone? Tap and hold the code to copy it. Prefer desktop? Use web.telegram.org.

Choosing the Right AI Model

Your agent can use different AI models. Here's how to pick the best one.

Our Recommendation

Use Claude Opus 4.6 as your main agent

Opus 4.6 is the smartest model available right now. It reasons deeply, follows complex instructions, remembers context well, and produces the highest quality output. This is the brain of your AI.

Then let your agent delegate simpler tasks to cheaper models automatically (see Cost Savings below).

Model Comparison

Claude Opus 4.6

RECOMMENDED

The most capable model. Excellent reasoning, writing, and coding. Best for your main agent brain.

Provider: Anthropic Speed: Moderate Cost: ~$15/1M input tokens

Claude Sonnet 4

GOOD VALUE

Very capable and faster than Opus. Great balance of quality and cost. Good alternative main model.

Provider: Anthropic Speed: Fast Cost: ~$3/1M input tokens

GPT-5.2

VERSATILE

OpenAI's latest. Strong all-around performance. Great for sub-agent tasks and general use.

Provider: OpenAI Speed: Fast Cost: ~$2/1M input tokens

GPT-4o

BUDGET PICK

Fast and cheap. Good for routine tasks. Best used as a sub-agent model for cost savings.

Provider: OpenAI Speed: Very Fast Cost: ~$2.50/1M input tokens

MiniMax M2.5

CHEAPEST

Extremely affordable. Solid for simple tasks, research, and data processing. Best as a sub-agent.

Provider: MiniMax Speed: Very Fast Cost: ~$1/1M input tokens

Saving Money on AI Costs

Smart strategies to keep your AI costs low without sacrificing quality.

Strategy 1: Smart Agent + Cheap Sub-Agents

This is the #1 cost saving strategy

Set your main agent to Claude Opus 4.6 (the smartest model). Then configure it to delegate sub-tasks to cheaper models:

  • Main agent (Opus 4.6) handles your conversations and makes decisions
  • Sub-agents (GPT-5.2 or MiniMax M2.5) handle research, data processing, and simple tasks
  • You get top-tier intelligence where it matters, budget pricing for the grunt work

This can reduce your AI costs by 50-70% compared to using Opus for everything.

Strategy 2: Use Anthropic Through OpenClaw Gateway

When you connect your Anthropic key through OCP, your agent accesses Claude models through the OpenClaw gateway. This means:

  • Automatic caching of repeated context (saves on token costs)
  • Efficient message batching
  • Smart token management that reduces waste

You pay Anthropic directly for usage, but the gateway optimizes how tokens are used, so you get more value per dollar.

Strategy 3: Set Usage Alerts

Both OpenAI and Anthropic let you set spending limits:

  • OpenAI: Settings > Billing > Usage limits. Set a monthly cap (e.g., $20)
  • Anthropic: Settings > Plans > Usage limits. Set spending alerts

Start with a $20 monthly limit on each provider. You can always increase it later.

Typical Monthly Costs

$5-10

Light Use

A few conversations per day, simple questions

$15-30

Regular Use

Daily conversations, research tasks, some coding

$30-60

Heavy Use

All-day usage, complex tasks, lots of coding

These are AI provider costs (paid to OpenAI/Anthropic), separate from your OCP subscription.

Talking to Your AI

The difference between a good AI experience and a great one is how you communicate. Here's the complete guide.

The Golden Rule: Be Specific Up Front

The single biggest thing that improves your AI results: give clear, detailed instructions BEFORE it starts working. Here are three techniques:

1. Start Big, Then Refine

Instead of this:

"Write me an email"

Try this:

"Write a professional email to my landlord about a broken heater. Firm but polite tone. Mention this is the third time I've reported it. Keep it under 200 words."

2. Ask for a Summary First

Before the AI builds something complex, say:

"Before you build anything, give me a summary of what you think I'm asking for."

This catches misunderstandings before 20 minutes of wasted work.

3. Let the AI Ask Questions

For complex projects, start with:

"I want to [describe idea]. Ask me questions before you start building."

The AI will ask the right questions to fill in gaps you might not have thought of.

Pro Tips for Better Conversations

Be direct, skip the pleasantries

You don't need "please" or "could you." Just say what you need. The AI won't be offended.

Correct mistakes immediately

"No, I meant X" works perfectly. Don't restart the whole conversation.

Use "remember this"

When you want the AI to remember something across conversations, just say "remember this." It will save it to its memory file.

Give multiple tasks at once

You can list several things. The AI handles them in order.

Long tasks work in the background

For big jobs, the AI can work while you do other things. Close Telegram and come back later.

Set reminders naturally

"Remind me to [thing] at [time]" works. The AI will ping you.

What Your AI Can Do

Your agent is more capable than you might think. Here's the full list:

Conversations & Thinking

Answer questions, brainstorm ideas, analyze problems, give advice

Writing & Content

Emails, documents, reports, creative writing, editing, translations

Research & Web

Search the web, read articles, summarize sources, compare options

Code & Technical

Write code, debug, run scripts, automate tasks, build tools

Files & Documents

Create, read, edit, and organize files on your server

Scheduling & Reminders

Set reminders, recurring tasks, time-based notifications

Images & Vision

Analyze photos you send, describe images, read screenshots

Memory & Learning

Remembers your preferences, past conversations, important details

Command Reference

Type these commands directly in your Telegram chat with your AI agent.

Essential Commands (Start Here)

/status
See your current setup, which model is active, and usage stats
/help
Show available commands
/stop
Stop the AI mid-response if it's going in the wrong direction
/reset
Start a fresh conversation (clears context, keeps long-term memory). Also: /new

Model & Thinking

/model
Show available models with a numbered picker
/model list
Detailed model list with providers
/model 3
Switch to model #3 from the picker
/model status
Detailed view of current model, endpoint, and API mode
/think high
Enable deep thinking for complex problems (costs more but much better output). Options: off, minimal, low, medium, high, xhigh
/think off
Turn off deep thinking for simple conversations
/reasoning on
Show the AI's reasoning process in a separate message
/reasoning off
Hide reasoning (default)

Usage & Cost Tracking

/usage tokens
Show token count after each response
/usage cost
Show a cost summary from session logs
/usage full
Detailed token breakdown per response
/usage off
Hide usage info (default)
/context
See how much conversation context the AI is tracking
/context detail
Detailed breakdown: files, tools, skills, system prompt sizes

Voice & Output

/tts always
AI sends voice messages instead of text
/tts off
Back to text responses (default)
/tts status
Check current voice settings
/verbose on
Extra detail in responses (debugging mode)
/verbose off
Normal output (default)

Sub-Agents & Background Tasks

/subagents list
See currently running background tasks
/subagents stop
Stop a running sub-agent
/subagents log
View output from a sub-agent
/compact
Compress conversation history to save context space (useful for long sessions)

Identity & Info

/whoami
Show your sender ID (useful for configuration). Also: /id
/commands
Full list of all available commands

Tip: Inline Commands

You can mix commands with regular messages. For example: /think high Analyze this complex business problem for me...

The command is processed and stripped before the AI sees your message. Works with: /think, /model, /verbose, /reasoning, /elevated

Frequently Asked Questions

Quick answers to common questions.

You get your own dedicated server running 24/7 with a personal AI agent. The agent connects to Telegram so you can chat with it anytime, from anywhere. It can help with writing, research, coding, planning, answering questions, running tasks, browsing the web, and much more. Think of it as having a brilliant assistant that never sleeps.

OCP provides the server and infrastructure. The AI models (ChatGPT, Claude) are made by separate companies (OpenAI, Anthropic) who charge for usage. You pay them directly, which means you have full control over your spending, can set your own limits, and you're never locked in. It also means we never see or touch your AI billing.

It depends on your model choices. During setup, you pick a "brain" (main AI) and a "coder" (for quick tasks). If both use the same provider (e.g., both from OpenAI or both from Anthropic), you only need one key. If they use different providers, you need both. We recommend Claude Opus 4.6 as the brain and GPT-5.2 Codex as the coder, which gives you the best of both worlds.

For most people, $10-30/month total across both providers. Light users (a few chats per day) spend around $5-10. Heavy users (all-day usage, complex tasks) might spend $30-60. You can set hard spending limits on both OpenAI and Anthropic so you never get surprised. Start with a $20 limit on each and adjust from there.

Starter is perfect for personal use. 4 GB of memory is enough for most conversations and tasks. Professional is for power users who want their agent running multiple tasks simultaneously or handling very long conversations. Enterprise is for maximum performance with heavy workloads. When in doubt, start with Starter. You can upgrade anytime.

Yes. Your server is completely isolated. No one else shares it. Your conversations, files, and data stay on your server. Your API keys are encrypted at rest. We don't read your messages or access your AI billing. The only time we access your server is for maintenance or if you request support.

Telegram is the primary and recommended way to chat with your agent. OpenClaw (the software powering your agent) also supports Discord, Signal, and other platforms. If you need a different messenger, contact us and we can help set it up.

A lot! Here are some examples:

  • • Answer questions and have conversations on any topic
  • • Write emails, documents, reports, and creative content
  • • Research topics and summarize information from the web
  • • Help with coding and technical tasks
  • • Set reminders and manage schedules
  • • Analyze documents, images, and data
  • • Run background tasks while you're away
  • • Learn your preferences over time and personalize responses

Email us at support@ocpbot.com and we'll walk you through it. Most setup issues are resolved within a few minutes.

Yes. Your subscription is month-to-month. Cancel anytime and your server will run until the end of your billing period. The setup fee is non-refundable, but the monthly subscription can be canceled with no penalties.

Still have questions?

We're here to help.

Email Support