Most AI assistants wait for you to talk to them. You wake up, open an app, ask a question. The assistant answers. You close the app and get on with your day.
That’s not how ours works. By the time we pour a coffee in the morning, OpenClaw has already pulled the day’s Bitcoin price movement, checked for new AI tool launches worth covering on computertech.co, scanned our site for broken links, and sent a clean briefing to Telegram — all without us touching a single key.
This is what a properly configured OpenClaw morning briefing looks like. It’s not magic, it’s a cron job, a handful of skills, and about 90 minutes of setup. This guide walks you through exactly how to build one.
What Is an OpenClaw Morning Briefing?
An OpenClaw morning briefing is an automated daily summary delivered to your messaging channel — Telegram, Discord, Slack — on a schedule you define. It pulls together whatever context matters to you: market data, news, your calendar, task lists, site metrics, or anything else your agent can reach.
Think of it like a newspaper that was written specifically for your brain, by a researcher who knows your exact priorities, delivered before you’re even fully awake. The difference from a regular newsletter or news aggregator is that OpenClaw actually reasons about what it finds. It doesn’t just dump headlines — it surfaces what’s relevant to your specific situation and flags what needs your attention.
In our setup, the morning briefing does six things every day at 6:45 AM Mountain Time:
- Checks Bitcoin price and 24-hour movement against key support/resistance levels
- Scans AI news feeds for tool launches and significant model releases worth covering
- Reviews computertech.co uptime and flags any 404s or crawl errors from the night
- Pulls our top-performing posts from the past 7 days
- Checks the content queue and flags anything overdue
- Generates one strategic suggestion based on everything it found
The whole thing takes about 45 seconds to run and lands as a clean Telegram message every morning. Here’s how to build yours from scratch.
Prerequisites: What You Need Before You Start
Before touching a cron job, make sure your OpenClaw environment is actually ready for it. A morning briefing that fails silently is worse than no morning briefing — you’ll miss things and not know it.
OpenClaw Installed and Running
If you haven’t set up OpenClaw yet, start with our complete Windows installation guide or the Mac and Linux setup guide. You need the Gateway daemon running as a persistent service — not just a terminal session that closes when you log off.
Verify it’s running with openclaw gateway status. If it’s not configured as a system service, the cron will fire into the void at 6:45 AM while your computer is sitting there doing nothing.
A Connected Messaging Channel
The briefing needs somewhere to go. We use Telegram — it’s fast, reliable, and the OpenClaw Telegram integration handles formatting cleanly. If you haven’t set that up, our Telegram connection guide covers it in full. Discord works too — see our Discord and Slack setup guide if that’s your preference.
Your Model Configuration
Morning briefings don’t need your most expensive model. We run ours on anthropic/claude-sonnet-4-6 — fast enough to complete in under a minute, capable enough to actually synthesize what it finds rather than just listing it. Save your Opus quota for the tasks that genuinely need deep reasoning.
Any Skills or Integrations You Want Included
If your briefing is going to check cryptocurrency prices, pull RSS feeds, or ping your WordPress site, you’ll want those integrations tested and working before wiring them into a cron. OpenClaw’s MCP integration guide is the right place to start if you’re connecting external tools. Test everything manually first — a cron is a terrible place to debug a broken API key.
Understanding OpenClaw Crons: The Foundation
If you’ve used cron jobs on Linux, OpenClaw’s cron system will feel immediately familiar. If you haven’t, here’s the mental model: you’re scheduling a task to run automatically at a specified time, with a specified payload, targeting a session. The OpenClaw cron system extends this with AI-native concepts — sessions, agent turns, delivery channels, and structured outputs.
There are two key job types you need to understand:
systemEvent Jobs (Main Session)
These inject text as a system event into your main OpenClaw session. They’re simple and fast, but they run in the context of whatever you’re currently doing. If you’re mid-conversation at 6:45 AM, the briefing fires into that same session. This works fine for simple notifications, but it’s not ideal for a complex morning briefing that needs to do real research work.
agentTurn Jobs (Isolated Sessions)
These are what we use for the morning briefing. An isolated session spins up independently, runs the agent turn with your specified prompt, completes the task, and delivers the result to your channel. It doesn’t interrupt anything, it doesn’t depend on your main session being available, and it runs clean every time. The output is announced to your configured delivery target when the job finishes.
For anything more complex than “remind me to check X,” use an isolated agentTurn job. The slight complexity is worth it.
Building Your First Morning Briefing Cron
Here’s how we set up the core morning briefing. Walk through this step-by-step — don’t rush the prompt writing, because that’s where most people end up with a briefing that’s technically running but practically useless.
Step 1: Define What You Actually Want
Before touching the OpenClaw cron tool, write down — in plain language — what you want to know every morning. Be specific. “News” is not specific. “New AI model releases, tool launches, or significant API pricing changes” is specific.
Our list looks like this:
- Bitcoin price (current, 24h change, whether it’s above or below a key level)
- New AI tool launches or major model releases in the last 24 hours
- computertech.co site health (uptime, any new 404s flagged by Search Console)
- Our week’s top 3 performing posts by traffic
- Content queue status (anything scheduled for today, anything overdue)
- One actionable suggestion based on everything above
Write your own list. Keep it under 8 items — more than that and the briefing becomes noise instead of signal. The whole point is focus.
Step 2: Write the Agent Prompt
This is the most important step. Your prompt is what runs inside the isolated session every morning. Write it like you’re briefing a smart assistant who knows nothing about your specific situation — because each isolated session starts fresh.
Here’s the structure we use:
You are running the daily morning briefing for [YOUR NAME/BUSINESS].
Today is {date}. Complete each of the following sections and deliver
a clean, formatted summary to Telegram.
SECTION 1 — BITCOIN:
Check the current BTC price. Note the 24h change percentage. Flag
if price is above or below [YOUR KEY LEVEL]. Keep this to 2-3 sentences.
SECTION 2 — AI NEWS:
Search for AI tool launches, major model releases, or significant
API changes in the last 24 hours. Focus on tools that would be
worth reviewing on computertech.co. List up to 3 items with one
sentence each. Skip marketing announcements with no substance.
SECTION 3 — SITE HEALTH:
Check computertech.co is returning HTTP 200. Note any crawl errors
or 404s if you can access Search Console data.
SECTION 4 — CONTENT QUEUE:
Check what's scheduled for today. Flag anything overdue.
SECTION 5 — ONE SUGGESTION:
Based on everything above, give ONE concrete action I should
prioritize today. Not a list. One thing.
Format the output clearly with emoji section headers. Keep it tight —
this should be readable in under 60 seconds.
The key principles here: each section has a clear scope, clear output format, and a constraint on length. “Keep it tight” isn’t enough on its own — you need to say “2-3 sentences” or “list up to 3 items” or the agent will write you a novel.
Step 3: Create the Cron Job
With your prompt ready, create the job through OpenClaw. Here’s the cron configuration structure for an isolated agentTurn briefing:
Schedule: cron, "45 6 * * *", tz: "America/Edmonton"
Session: isolated (agentTurn)
Model: anthropic/claude-sonnet-4-6
Delivery: announce → your Telegram channel
Payload: your prompt from Step 2
The 45 6 * * * expression means “6:45 AM, every day.” Adjust the time to something that actually makes sense for your schedule — firing a briefing at 3 AM MST because you copied a UTC example is a common mistake.
You can also limit this to weekdays only: 45 6 * * 1-5. Or add a Saturday version with a lighter format for weekends. The cron expression gives you full control.
Step 4: Test It Immediately
Once the job is created, trigger it manually rather than waiting until tomorrow morning to find out it’s broken. OpenClaw’s cron tool has a “run” action for exactly this — it fires the job immediately regardless of schedule.
What you’re checking:
- Does it actually complete? (Watch for timeout issues with complex prompts)
- Does the output arrive in your Telegram channel?
- Is the format what you wanted, or did it ignore your constraints?
- Did each section produce useful output, or is one of them returning nothing?
Iterate on the prompt before declaring it done. We went through four versions of our briefing prompt before landing on something that consistently delivered what we actually wanted. The first version had a section on “competitor activity” that the agent kept interpreting as “write me a competitive analysis essay.” It took one word change — “flag (max 2 bullet points)” — to fix it.
Advanced Configuration: Making the Briefing Actually Smart
A basic morning briefing is useful. A well-configured one is the difference between having an assistant and having a strategic partner who shows up every morning having already done homework.
Use OpenClaw Memory for Continuity
One limitation of isolated sessions is that they start fresh each time. They don’t remember what last Tuesday’s briefing said. This means if something was flagged yesterday, it’ll be flagged again today with no context about history.
The fix: use OpenClaw’s memory system inside your briefing prompt. You can instruct the agent to check memory before running, and update memory with key findings after completing.
Add this to your prompt:
Before starting: Search memory for "morning-briefing-context" to load
any ongoing items that need follow-up.
After completing: Store any new flagged items, ongoing issues, or
decisions that should carry forward to tomorrow's briefing under
"morning-briefing-context".
Now the briefing has continuity. If a keyword was trending yesterday, today’s briefing knows that. If you flagged an article idea two days ago, it’ll show up in the queue until you clear it. Our OpenClaw memory system deep-dive explains exactly how Mem0 and LCM work together — worth reading if you want to build on this pattern.
Wire In External Data Sources
The briefing is only as good as what it can actually access. Out of the box, OpenClaw can search the web — but for reliable structured data, you want MCP connections to specific sources rather than general web searches.
For our setup, the high-value integrations are:
- Google Search Console via MCP — Real crawl data, not guesses about site health
- A Bitcoin price API — CoinGecko’s free tier is enough for daily briefing data
- RSS feeds from key AI news sources — More reliable signal than a general search for “AI news today”
For each of these, you configure the MCP connection once (see our MCP integration guide for the process), then reference it in your briefing prompt by tool name. The agent calls the tool, gets structured data back, and incorporates it into the briefing. Much cleaner than hoping a web search surfaces what you need.
Build a Custom Briefing Skill
Once your briefing format is stable and you’re happy with it, consider packaging it as an OpenClaw skill. This means instead of a long prompt in your cron job, you just invoke the skill: run morning-briefing. The skill handles all the logic internally.
The advantages: you can update the briefing format in one place (the skill file) without touching the cron configuration. You can version it. You can test changes by invoking the skill manually before letting it run on schedule. Our custom OpenClaw skills guide covers the skill structure — the morning briefing is actually a perfect first skill project because it has a clear scope and repeatable structure.
Conditional Logic and Smart Filtering
Here’s what most “automated AI briefing” tutorials skip: the difference between a briefing that tells you everything and a briefing that tells you what matters.
Your agent can apply conditional logic. You don’t need to code this — you just describe the conditions in the prompt. Some examples from our setup:
Bitcoin section: Only escalate the alert tone if price has moved
more than 5 percent in 24 hours OR if it's within 3 percent of a key level.
Otherwise, keep it to one neutral sentence.
AI news section: If there are no significant launches, say so
in one sentence. Don't manufacture content to fill the section.
Content queue: Only flag items as overdue if they're more than
48 hours past their scheduled date. Don't flag items due today
as at-risk — that's noise.
Conditional logic prevents the briefing from becoming the boy who cried wolf. If it’s screaming about everything, you’ll start skimming it. If it only escalates when something actually matters, you’ll read every word.
Real Example: Our Actual Morning Briefing Output
Here’s a representative output from our setup, formatted as it arrives in Telegram (content adjusted for privacy):
☀️ Morning Briefing — March 24, 2026
₿ BITCOIN
$87,420 | +2.3% (24h) | Holding above the $85K range support. No urgency. Watch $90K resistance if momentum continues.🤖 AI NEWS
• Anthropic released Claude 3.7 Haiku — 40% faster than Haiku 3.5, same pricing tier. Review candidate.
• Google launched Imagen 4.0 with native upscaling. Potential computertech.co coverage.
• Nothing else material in the last 24h.🌐 SITE HEALTH
computertech.co: HTTP 200. Search Console shows 2 new crawl errors on /tag/ pages. Not urgent, batch fix this week.📊 CONTENT QUEUE
1 article scheduled today (Claude 3.7 Haiku review). 0 overdue.⚡ TODAY’S PRIORITY
Haiku review is the move — first-mover window is 24-48 hours and you have the infrastructure ready. Start the Codex build session now.
That’s it. Readable in 45 seconds. Actionable. No filler. The “TODAY’S PRIORITY” section is the one we actually pay attention to most — it synthesizes everything into one concrete move, which is the hardest thing to do manually when you’re half-awake and looking at a bunch of inputs.
Troubleshooting Common Issues
The Briefing Fires But Nothing Arrives in Telegram
Check three things in order: (1) Is the Gateway daemon actually running as a system service? (2) Is the Telegram bot token still valid — bot tokens can expire if the bot hasn’t been used. (3) Is the delivery configuration in your cron job pointing to the right channel? Check delivery.channel matches your configured Telegram channel ID.
The Briefing Times Out
Isolated agentTurn jobs have a default timeout. If your briefing is doing too much — too many tool calls, too many web searches, too complex a prompt — it might hit the limit and fail silently. Fix this by either setting timeoutSeconds: 0 (no timeout) in your job configuration, or by breaking the briefing into two separate jobs: a data-gathering job at 6:30 AM and a summarization job at 6:45 AM that reads what the first one stored in memory.
The Output Format Is Inconsistent
LLMs are stochastic — the same prompt won’t produce byte-identical output every time. If your format is drifting (sections in different orders, different emoji, different length), the fix is more explicit constraints in the prompt. Add a format template with exact section headers. Specify the maximum character count per section. The more explicit the format instructions, the more consistent the output.
One Section Always Returns Useless Data
This usually means the data source for that section isn’t reliable. Web searches for “AI news today” will return different quality results depending on what’s actually happening that day. If a section is consistently garbage, switch to a structured data source — an RSS feed, an API, or a pre-configured MCP tool — instead of relying on general search.
The Briefing Runs at the Wrong Time
This is almost always a timezone issue. The cron expression runs in UTC by default unless you explicitly set a tz parameter. If you’re in Mountain Time (UTC-7) and you want 6:45 AM, your UTC cron is 45 13 * * * — or you set tz: "America/Edmonton" and use 45 6 * * *. Don’t mix these up. And remember that Mountain Time shifts between MDT (UTC-6) and MST (UTC-7) seasonally if you’re using a UTC expression without a timezone parameter.
Expanding the Briefing: Ideas Worth Building
Once the core morning briefing is solid, you have infrastructure you can build on. Some patterns we’re either running or planning:
An Evening Wind-Down Briefing
A 5 PM check that reviews what got done, flags anything unfinished, and preps the next day’s priority. Uses the same architecture as the morning briefing but with a different prompt focused on completion and handoff rather than planning.
Alert-Based Briefings
Instead of scheduled, these fire when a condition is met. Bitcoin drops 10% in an hour — immediate Telegram alert. A new Claude model drops — research trigger fires automatically. OpenClaw’s cron system supports time-based schedules, but you can also trigger jobs programmatically from within sessions. Combine this with a monitoring script and you have event-driven AI responses.
Weekly Strategy Briefing
A Sunday morning job that does a deeper review — week’s content performance, revenue trends, what worked and what didn’t, strategic priorities for the coming week. Same architecture but a longer prompt, more data sources, and more reasoning expected. This one we run with Opus rather than Sonnet because the synthesis task is more complex.
Client or Team Briefings
If you’re running OpenClaw for a team or a client, you can configure multiple briefing jobs — one per person, each with their own prompt, delivery channel, and context. The isolation of agentTurn sessions means each briefing runs clean without interfering with others.
The Honest Take on Automated AI Briefings
Here’s what other guides won’t tell you: the briefing is only as useful as the work you put into the prompt and data sources. We’ve seen people set up a morning cron in 10 minutes, get frustrated when the output is generic and unhelpful, and conclude that automated briefings don’t work. They do work — but they require the same careful thinking you’d apply to briefing a human employee.
Think about it this way: if you hired a research assistant and handed them a note that said “summarize AI news and give me a briefing,” you’d get exactly what the briefing prompts without guardrails produce — technically correct, practically useless. If you handed them a 2-page brief that specified exactly what to look for, what to ignore, what thresholds matter, what format you want, and what decision you’re trying to make — you’d get something worth reading.
Your prompt is the brief. Write it like your results depend on it. They do.
Once it’s dialed in, though, the compounding effect is real. We’ve been running ours for several months and the thing that surprises us most isn’t the convenience — it’s how rarely we miss important things now. The briefing has caught tool launches we would have missed by a day, flagged site issues before users hit them, and surfaced keyword opportunities we wouldn’t have looked for manually. That’s the actual value: not automation for its own sake, but reliable coverage of things that matter.
Where to Go Next
The morning briefing is one of the most immediately useful things you can build with OpenClaw, but it’s a starting point, not a ceiling. Once you understand the cron + isolated session + delivery pattern, you can apply it to almost anything that benefits from regular automated attention.
- OpenClaw Cron Jobs: Full Deep-Dive — The complete guide to what OpenClaw’s scheduling system can do
- Our AI Content Pipeline — How we automate research, drafting, and publishing for computertech.co
- OpenClaw for Affiliate Marketing — Revenue operations automation that runs while you’re away from the desk
- Full OpenClaw Review 2026 — Our overall take on the platform after running it daily
- OpenClaw on GitHub — Source code, issues, and the community building on top of it
- OpenClaw Documentation — Reference for cron configuration, session types, and delivery options
Frequently Asked Questions
Do I need to keep my computer on for the OpenClaw morning briefing to work?
Yes — OpenClaw’s Gateway daemon needs to be running. On Windows, configure it as a Windows Service so it starts automatically on boot and doesn’t require you to be logged in. If you need it to run 24/7 without depending on your local machine, consider running OpenClaw on a small VPS or Raspberry Pi instead.
How much does running a daily morning briefing cost in API tokens?
With Claude Sonnet as the model, a typical morning briefing that does 4-6 tool calls and produces 400-600 words of output costs roughly $0.01-0.03 USD per run. At daily frequency that’s under $1/month. Using Opus would increase costs 5-10x — not worth it for a structured briefing task where Sonnet performs comparably.
Can the morning briefing check my email or calendar?
Yes, if you configure the appropriate integrations. Google Calendar can be connected via MCP. Email is more complex — you’d need either an MCP tool that wraps Gmail’s API, or a pre-processing script that extracts relevant items and writes them to a file your briefing agent can read. The integration work is front-loaded, but once it’s done it’s automatic.
What happens if the briefing fails — do I get notified?
By default, failed jobs don’t send a notification — they just fail silently. You can check the cron job’s run history to see failure logs. For mission-critical briefings, add a second simple cron that fires 30 minutes after the main one and checks whether the briefing memory was updated — if not, it sends a fallback alert. Belt-and-suspenders approach but worth it if you rely on the briefing for time-sensitive decisions.
Can I have different briefings for weekdays vs weekends?
Yes. Use separate cron jobs with different schedule expressions: 45 6 * * 1-5 for weekdays and 45 8 * * 6,0 for weekends (with a different start time and lighter prompt). Each job can have a completely different prompt, model, and delivery target.
How do I stop the morning briefing from being repetitive when nothing changes?
Instruct the agent explicitly in the prompt to vary phrasing and only escalate when thresholds are crossed. Also use memory to track what was said previously — if Bitcoin has been flat for three days, the briefing should say “still consolidating at $X, same story as yesterday” rather than re-explaining the entire price situation from scratch.
Is there a limit to how many cron jobs I can run in OpenClaw?
No hard limit enforced by the platform — it depends on your API rate limits and Gateway resource constraints. In practice, running 3-5 scheduled jobs daily is trivial. If you’re running dozens of jobs with heavy tool usage in tight windows, you may hit API rate limits. Stagger your job times to avoid concurrent execution during the same time window.
