Automate Research, Deliver Intel to Telegram Groups, and Never Miss a Signal Again

By James Tipton & Chief Wizard 

What We're Building

A scheduled AI agent that:

• Wakes up at a set time every day
• Searches the web for live signals across your chosen topics
• Formats a structured report
• Delivers it automatically to a Telegram group chat
• Updates its own memory for continuity across sessions

No manual triggering. No copy-pasting. Just intelligence on tap.

Prerequisites

Before starting this guide you should have:

• OpenClaw installed and running (see our full setup guide)
• A Telegram bot created via @BotFather and connected to OpenClaw
• Your bot added to the Telegram group you want to post to
• A web search API key (we'll cover which one to use below)

1 The Web Search Problem

Out of the box, OpenClaw ships with DuckDuckGo as its search provider. Sounds great in theory. In practice, DuckDuckGo aggressively bot-detects automated requests and will return challenges instead of results roughly 80% of the time during cron runs. Your bot will sit there timing out while your audience waits for intel that never arrives.

You'll see this in your logs:

[tools] web_search failed: DuckDuckGo returned a bot-detection challenge.

The Fix: Switch to Tavily

Tavily is purpose-built for AI agent workflows. It handles rate limiting gracefully, returns clean structured results, and doesn't bot-detect your automated runs. It's the right tool for this job.

Get your Tavily API key:

1. Go to app.tavily.com
2. Sign up for a free account (1,000 searches free)
3. Copy your API key

Configure OpenClaw to use Tavily:

openclaw config set plugins.entries.tavily.enabled true

Then set your API key (the config path moved in recent versions):

openclaw doctor --fix
openclaw config set plugins.entries.tavily.config.webSearch.apiKey YOUR_TAVILY_KEY

Set Tavily as your search provider:

openclaw config set tools.web.search.provider tavily

Restart the gateway:

openclaw gateway restart

Why not Brave Search? Brave was the community standard but removed their free tier in February 2026. New users get $5 in monthly credits before billing starts — roughly 1,000 searches. Tavily's free tier is a cleaner starting point. Once you exhaust Tavily's free credits, Brave's $5 monthly credit is a solid fallback. You can alternate between them.

2 Finding Your Telegram Group ID

Before you can direct a cron job to post in a group, you need the group's chat ID. Here are three ways to find it:

Method 1 — From Telegram Web

Open your group in Telegram Web. Look at the URL:

https://web.telegram.org/k/#-1001234567890

The number including the minus sign is your group ID.

Method 2 — Use @userinfobot

1. Add @userinfobot to your group temporarily
2. Send any message
3. It replies with the group's chat ID
4. Remove @userinfobot

Method 3 — Watch the OpenClaw Logs

Send a message to your group while tailing the logs:

tail -f /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log | grep -i "chatId"

Your group ID will appear in the output.

3 Creating Your First Cron Job

In the OpenClaw dashboard, navigate to Cron Jobs → New Job. Here's how to fill out each field:

Basics

• Name: Something descriptive — e.g. Tech Temple Alpha Report
• Description: Brief note on what this job does and when
• Agent ID: main
• Enabled: On

Schedule

• Schedule type: Cron
• Expression: Standard cron syntax — minute hour day month weekday
• Timezone: Set your local timezone (e.g. America/Los_Angeles) — OpenClaw will convert to UTC automatically

Common cron expressions:

55 17 * * *    →  5:55 PM daily
0  9  * * *    →  9:00 AM daily
0  9  * * 1-5  →  9:00 AM weekdays only
0  9  * * 1    →  9:00 AM Mondays only

Execution

• Session: Isolated — runs in its own dedicated agent turn
• Wake mode: Now — triggers immediately when the cron fires
• What runs: Run assistant task (isolated)
• Timeout: Set to 90 seconds — prevents hanging on slow searches

Delivery

• Result delivery: Announce summary (default)
• Channel: Telegram
• To: Your group ID including the minus sign (e.g. -1001234567890)

Advanced

• Exact timing: On — fires at the precise cron boundary
• Best effort delivery: On — job doesn't fail if Telegram has a momentary hiccup

4 Prompt Engineering for Cron Jobs

This is where most people go wrong. Cron job prompts are different from conversational prompts. The agent runs in isolation with no back-and-forth — it needs complete, unambiguous instructions in a single prompt.

The Core Principles

Run 4 web searches max.

Use search results only — do not fetch full pages.

Total report under 400 words.

For crypto data prefer coingecko.com or coinmarketcap.com
directly — never cite social media posts as price data.

If no live source found, say so — never fabricate
data, prices, odds, or people.

Save top 3 insights to memory after delivery.

What Breaks Cron Jobs

5 The Full Working Prompt

Here is the complete prompt that produces reliable, useful daily reports after multiple iterations of testing. Copy it directly and adapt the search topics and sections for your use case.

The structural constraints — search limit, word cap, no page fetching, honesty instruction, memory save — should stay in every cron prompt regardless of topic.

6 Testing and Debugging

Manual Test Run

After creating your job, test it before leaving it on schedule. Find your job ID first:

openclaw cron list

Then run it manually:

openclaw cron run YOUR_JOB_ID

Watching the Logs

Always tail logs during a test run to see exactly what's happening:

tail -f /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log | grep -i "web_search\|error\|cron"

Common Errors and Fixes

Context overflow:

Context overflow: estimated context size exceeds safe threshold

Your prompt is doing too many searches. Reduce to 4 searches max and add a 400 word limit.

ChatGPT timeout:

openai-codex timed out. failoverReason: timeout

The job took too long. Set timeout to 90 seconds in job settings and reduce prompt complexity.

Error backoff loop:

consecutiveErrors: 2, backoffMs: 60000

The job failed twice and is pausing itself. Reset it by disabling and re-enabling in the dashboard, or via CLI:

openclaw cron disable YOUR_JOB_ID
openclaw cron enable YOUR_JOB_ID

DuckDuckGo bot detection:

web_search failed: DuckDuckGo returned a bot-detection challenge

Switch to Tavily as described in Part 1.

Silent delivery failure: Job runs but nothing arrives in Telegram. Check your group ID includes the minus sign. Verify the bot is a member of the group and has Group Privacy disabled in BotFather.

7 Scaling Up

Multiple reports for different audiences. Create separate cron jobs for different groups with different prompts. A general audience channel gets a different report than a technical channel. Each job has its own schedule, prompt, and delivery target.

Chained intelligence. Set earlier jobs to save memory, and later jobs to read it. Your morning brief informs your evening report. The agent builds a picture of the day across multiple runs.

Heartbeat jobs. Lighter weight than cron — a heartbeat runs on a regular pulse and checks for anything that needs attention rather than generating a full report. Good for maintenance tasks and memory upkeep between scheduled runs.

Multiple channels. The same cron infrastructure works for WhatsApp, Discord, Slack, and iMessage — just change the channel and delivery target in the job settings.

Key Lessons from the Field

Shorter prompts outperform longer ones. The instinct is to give the agent more instructions. The reality is more instructions means more context consumed before it even starts working. Keep prompts lean and specific.

Honesty instructions matter more than you think. Without explicit "never fabricate" instructions, agents will invent plausible-sounding sources, people, and prices. This destroys audience trust fast. Make the honesty constraint non-negotiable.

DuckDuckGo is not viable for automated runs. It works fine for conversational use but fails consistently under scheduled automation. Switch to Tavily or Brave for any cron-based web research.

The minus sign on group IDs is the most common silent failure. Include it. Every time.

Test with Run Now before trusting the schedule. Don't set up a cron job and assume it works. Run it manually, watch the logs, verify delivery, then let it loose.

What a Working Report Looks Like

Here's a real output from our Tech Temple Alpha Report after the prompt was dialed in:

That's a real, useful daily briefing delivered automatically to a Telegram group at a scheduled time. No human in the loop. Chief Wizard handles the research, formatting, and delivery while you focus on building.

Resources

• OpenClaw docs: docs.openclaw.ai
• OpenClaw GitHub: github.com/openclaw/openclaw
• Tavily API: app.tavily.com
• Brave Search API: api-dashboard.search.brave.com
• BotFather: Search @BotFather in Telegram
• OpenClaw Discord: discord.com/invite/clawd

Keywords: OpenClaw cron jobs, OpenClaw Telegram group, AI agent automation, OpenClaw web search, Tavily OpenClaw, OpenClaw prompt engineering, automated Telegram bot, AI daily report bot, OpenClaw tutorial 2026, agentic AI automation