Daily Morning Report: Fetch News on Schedule and Send It to Your IM
💡 One-line summary: use OpenClaw to build an AI daily news morning report pipeline — scheduled fetching, automatic summarization, and on-time delivery to your IM.
The first lesson on the creator track tackles the creator's input problem first: knowing what happened in your field every day, in a stable and structured way. This lesson combines the cron recurring tasks you learned in Scheduled Tasks Automations with Claw's built-in web_search search tool — Gateway wakes Claw on schedule → Claw searches the whole web → writes the morning report → delivers it into your specified IM. Once configured, generating the morning report becomes something that happens automatically in the background every day; you just glance at it in the morning. All examples are based on OpenClaw v2026.10.9.
What You'll Be Able to Do After This Lesson
- Build a morning report task that runs automatically on time every day: it searches, writes, and delivers into IM on its own, fully unattended
- Set up a search provider for Claw (Kimi / MiniMax with direct connectivity in China, or the key-free DuckDuckGo) so it actually has something to search
- Deliver the morning report to any connected channel: a Telegram group, a Slack channel, or straight to your DMs
- Tune the morning report's style with the prompt: topic scope, time window, output structure, word-count limits — set once with
editand it sticks long-term
Why OpenClaw Instead of a Web-Based AI?
| Capability | Web-Based AI | OpenClaw Morning Report Task |
|---|---|---|
| How the report is produced | ❌ Open a webpage yourself every day, type the prompt, copy the result | ✅ Runs automatically on a cron schedule; configure once and it reports every day |
| Delivery | ❌ The result sits in a chat box and you have to remember to go back for it | ✅ Delivered straight into a Feishu / WeCom / Telegram group or your DMs |
| Style memory | ❌ Re-describe "what kind of daily report I want" every single day | ✅ Requirements live in the task prompt — edit set once, sticks long-term |
| Finding material | ❌ You have to find the links yourself and feed them in | ✅ web_search searches the whole web on its own, web_fetch fetches the article text itself |
| Cost of sticking with it | ❌ Burns half an hour of your manual effort every day; you'll quit within three days | ✅ Burns a few search calls; as long as Gateway is awake, it shows up every day |
Your Current Struggles
- First thing after waking up, you scroll news sources for half an hour piecing together "what happened in AI overnight," and still worry you missed something
- The AI content you scroll is scattered and unorganized, your bookmarks are full of "read later," and by the time you actually pick a topic the hot takes have already gone cold
- You tried having a web-based AI do the summarizing, but opening a page, pasting the prompt, and copying the result every day didn't last a week
When to Use This Move
When you need:
- A news digest for a specific field at a fixed time every day (AI updates, competitor news, industry policy)
- Information that comes from public webpages and search, with no login-required private content
And you don't want to:
- Repeat the "open AI → type prompt → copy result" routine every day
- Build a crawler + scheduled script + push service just for one daily report
💡 Two ways to "watch" — don't pick the wrong one
- Watch news across the whole web (what's new each day) → this lesson's
web_searchmorning report, OpenClaw's lightest form of news monitoring - Watch a specific page (has that page changed) → the conditional trigger in the next lesson Web Monitoring: Get Notified the Moment a Page Changes
Content that requires login can't be fetched by search; you'll need the logged-in session from Browser Control.
🎒 Prep Before You Start
- You've worked through Scheduled Tasks Automations: you can create cron recurring tasks and know
--announce --channel --to(this lesson assembles directly, no re-teaching cron basics) - Gateway is running and
/model statuscan report the current model - A search provider: recommended are Kimi or MiniMax with direct connectivity in China (API key required); if you have no key at all there's a key-free fallback, covered in step 1
- The IM channel where you want the morning report is connected (connect one in Scenario Playbooks: Feishu, WeCom, or Telegram all work; you can still learn without one — the morning report will just land in the run history for now)
- A cup of water ☕ — in about 12 minutes, your IM will gain an editor who clocks in on time every day
Follow Along
Step 1: Set Up a Search Provider for Claw (web_search)
Why: the morning report's facts all come back via web_search. OpenClaw's rule is straightforward: when not a single managed search provider is configured, web_search disappears from the tool list — Claw couldn't search even if it wanted to.
How (pick one of three, same effect):
- Control UI: In Settings → Search, choose a provider, fill in credentials, and click Test search to try a test search (configuring and testing providers requires Gateway admin permissions)
- CLI:
openclaw configure --section web - Environment variables: Put the provider's key into the Gateway process environment (for a gateway install, that's
~/.openclaw/.env); OpenClaw auto-detects in a fixed order: Brave → MiniMax → Gemini → Grok → Kimi → Perplexity → …; the first one that's ready wins
Recommended for users in mainland China:
| Provider | What it needs | Notes |
|---|---|---|
| Kimi (Moonshot AI) | KIMI_API_KEY or MOONSHOT_API_KEY | Uses the Moonshot search endpoint; when configuring you can pick a region (e.g. api.moonshot.cn) |
| MiniMax Search | MINIMAX_CODE_PLAN_KEY etc. | Results can be filtered by region (global / cn) |
| DuckDuckGo | No key, no account needed, but you must install the @openclaw/duckduckgo-plugin plugin first | Must be explicitly selected to take effect; it is never picked automatically |
You should see: Test search returns results and latency; or, in a chat, just say "search for big AI news from the past day" — Claw calls web_search and replies with source links.
- Quick check: have Claw search for a piece of today's news, with source links in the reply
💡 The model and the search provider are two separate tracks
The model that runs the morning brief (Qwen, DeepSeek, GLM, …) decides "who writes the copy," while the search provider (Kimi, Brave, MiniMax, …) decides "where to find material" — the two are not bound to each other and are configured separately.
Step 2: Create an AI daily news morning report task (cron + delivery to IM)
Why: A single command settles all three things — "when to run, what to run, where to send it" — and that's the entire infrastructure of the morning brief.
Based on the official docs' Morning brief example, switch to a mainland-China timezone and Chinese prompts (using a Telegram group as an example):
openclaw automations create "0 7 * * *" \
"搜索过去 24 小时 AI 领域的重要新闻,汇总成一份中文晨报。用中文回复;链接、代码和产品名保持原样。" \
--name "Morning brief" \
--tz "Asia/Shanghai" \
--session isolated \
--announce \
--channel telegram \
--to "-1001234567890"Map each parameter:
| Parameter | Purpose |
|---|---|
"0 7 * * *" | cron expression: triggers every day at 07:00 |
| Prompt (positional argument two) | What the task should do. The language is not inferred automatically; Chinese-language rules must be written into the prompt |
--tz "Asia/Shanghai" | Timezone used to evaluate the cron expression; if omitted, the Gateway host's timezone is used |
--session isolated | Each run starts a brand-new session, with no history baggage |
--announce --channel telegram --to "…" | Result delivery: for a Telegram group, use the -100… numeric id |
Match delivery targets to channels:
| Where the morning brief arrives | How to write it |
|---|---|
| Telegram group | --to "-1001234567890" (forum topic -100…:topic:123) |
| Slack / Discord / Mattermost | --to "channel:<id>" or user:<id> |
| Feishu / WeCom | For channel setup and target syntax, see Feishu setup, WeCom setup |
| Your own private chat | Don't write --to; use --delivery-target owner instead |
You should see: Successful creation. add/create, list, and show all preview the resolved delivery route — an unresolvable destination fails on the spot, so you don't have to wait until the next morning.
- Quick check: the delivery route resolves correctly in
openclaw automations show <job-id>
💡 This is OpenClaw's unique value
A web-based AI can't "run on schedule by itself and deliver itself into the group afterward." No crawler, no crontab, no push scripts — the single command above is the morning brief's entire ops cost.
Step 3: Do a test run right now, don't wait until tomorrow morning
Why: If you wait until 7 o'clock to discover the delivery config is wrong, a whole day's morning brief is gone. A forced run validates the full chain immediately.
openclaw automations run <job-id>
openclaw automations runs <job-id> --limit 5You should see: run returns output like { ok: true, enqueued: true, runId: "…" }; in runs this run's status is ok and completionStatus is succeeded; a few minutes later, the first morning brief appears in your IM.
Didn't receive it? Check in this order:
openclaw automations show <job-id>: Is the delivery route resolved correctlyAny delivery errors in the
runsrecords — successful execution and successful delivery are two different things, check them separatelyFor hosts with multiple channels configured, isolated tasks must be given
--channelexplicitlyA task that fails 10 times in a row is automatically disabled; find it with
openclaw automations list --all, fix it, thenenableQuick check: the first morning brief actually arrived in your IM
Step 4: Tune the morning brief's flavor (the prompt is the brief's persona)
Why: The first version of the brief will almost certainly not suit your taste — too long, too scattered, headlines about things you don't care about. The brief's "persona" is the task prompt; edit swaps it out without rebuilding the task.
openclaw automations edit <job-id> --message "<把下面魔法 Prompt 里的完整版贴进来>"Four tuning essentials:
- Pin down the time window: "the past 24 hours" must appear in the prompt, otherwise the search may return old news
- Pin down the Chinese-language rules: The task won't infer the reply language from the channel, region, or message history; when Chinese and English are mixed, state explicitly "narrative in Chinese, technical terms kept as-is"
- Pin down the output structure: how many headlines, a link for each item, a one-sentence summary at the end — without spelling this out, the daily digest format will drift every day
- Pin down the length: The morning brief is for bleary-eyed morning reading, so give it a word-count cap
You should see: after edit succeeds, the message in show is updated; then manually run once more, and the new-format morning brief arrives immediately.
- Quick check: run it twice in a row—the brief's format is stable and its length stays under control
📋 Magic Prompt
The “Prompt” on this page is not something you send to a chat window—it is the task prompt, the second positional argument in the step 2 command, or the full version swapped in during step 4 via edit --message. Write the role, task, output spec, and constraints clearly, and the brief's style stays fixed for the long term.
📝 Daily AI Morning Brief
Expected result: every day at 7 o'clock, you receive a Chinese AI brief in IM with a fixed structure and controllable length—top stories, worth a look, and a one-sentence summary, each item with a source link.
## Role
You are a tech news editor responsible for producing an AI morning brief every morning.
## Task
Use web_search to find important AI news from the past 24 hours and compile it into a Chinese morning brief.
## Input
### Required
- Topic scope: [AI model releases, AI product updates, AI open-source projects]
### Optional
- Special focus: [companies, products, or technical directions you want covered first?]
- Excluded content: [topics you don't want to see, such as funding gossip or stock prices?]
## Output spec
Output in the following structure, no more than 500 characters in total:
- 📋 AI Morning Brief · [today's date]
- Top stories (at most 3 items): one line per item—headline — one-sentence explanation + source link
- Worth a look (at most 5 items): one sentence per item + source link
- One-sentence summary: the single most notable thing today
## Constraints
- ✅ Every news item must include a source link
- ✅ Summaries in Chinese; technical terms, product names, and code stay in the original
- ❌ Do not fabricate news that can't be found; if fewer results come up, write fewer items
- ❌ Do not include news older than 24 hours in the brief📋 Multi-Topic Custom Morning Brief
Expected result: not just AI—list every topic you care about, each topic gets its own section, and topics with no activity are honestly marked “no updates today.”
## Role
You are a private intelligence officer responsible for compiling new developments across multiple topics every day.
## Task
Use web_search to search each topic in the topic list for news from the past 24 hours, and compile the results into a Chinese morning brief organized by topic sections.
## Input
### Required
- Topic list: [one topic per line, e.g., AI model news / frontend framework updates / product changes from competitor X]
### Optional
- Per-topic item cap: [default 3 items?]
- Special-focus keywords: [keywords you want prioritized under a certain topic?]
## Output spec
- First line: 📋 Morning Brief · [today's date]
- Organize by topic sections, at most N items per section, one sentence per item + source link
- If a topic has no updates worth reporting in the past 24 hours, write “no updates today”
## Constraints
- ✅ Every item includes a source link; summaries in Chinese
- ✅ For topics with no news, honestly write “no updates today”
- ❌ Do not pass off week-old news as today's news
- ❌ Do not pad the count with low-value filler itemsCheckpoint ✅
- Search provider is configured: Test search returns results, and asking Claw to search news in chat yields source links
openclaw automations listlistsMorning brief, and the delivery routing ofshowresolves correctly- A forced
runtest run succeeds, and the first morning brief arrives in IM - The prompt hard-codes “past 24 hours” and the Chinese-reply rule
- You can explain why an hourly on-the-hour recurring task may run a few minutes past the hour (it's not a malfunction)
Pitfall Reminders
❌ You configured a key, but Claw still says “I can't search the web”
The three most common causes, ordered by hit rate:
- No hosted provider configured:
web_searchdisappears from the tool list—Claw isn't unwilling to search, it simply doesn't have the tool - A key-free provider (such as DuckDuckGo) is selected: it is never auto-detected and selected; you must choose it explicitly in the config
- The key is in the wrong place: it must go into an environment the Gateway process can read (for a gateway install, put it in
~/.openclaw/.env), not your own shell config
Debugging entry point: the Test search in Settings → Search—it shows search results directly, or errors and latency.
❌ The brief's language drifts, mixing Chinese and English
Cause: scheduled tasks do not infer the reply language from the channel, region, or message history—this is by design, not a bug.
Correct approach: write the language rule into the task prompt. If it still drifts, make the rule more specific: “narrative text in Chinese; technical terms and product names stay in English.”
❌ Stale news slips into the brief
Cause: the prompt didn't constrain the time window, so old news from search results got written in as well.
Correct approach: hard-code “past 24 hours” in the prompt, and add a constraint “do not include news older than 24 hours in the brief.” Also worth clarifying: web_search caches per query for 15 minutes (cacheTtlMinutes defaults to 15)—a brief that runs once a day will never hit the cache, so don't worry about the cache feeding stale results.
❌ An hourly on-the-hour task runs a few minutes past the “hour”
Cause: recurring expressions with minute set to 0 and a wildcard hour (such as 0 * * * *) are automatically staggered by up to 5 minutes to smooth the on-the-hour load spike—official design, not a malfunction.
Correct approach: for exact timing, add --exact when creating the task; to customize the stagger window use --stagger 30s (cron scheduling only). Note: expressions with the hour hard-coded, like 0 7 * * *, are not subject to staggering—7 o'clock means 7 o'clock.
Advanced Tips (Optional)
Deliver to your private chat instead of a group. Omit --to and use --delivery-target owner instead—at the scheduled time it resolves the explicitly identified owner and sends a direct message, without guessing “who I chatted with last”:
openclaw automations edit <job-id> --announce --channel telegram --delivery-target ownerNote that it cannot be combined with --to, --thread-id, or --webhook.
Add a layer of X (Twitter) sentiment. Add a line to the prompt: “also use x_search to check X discussions about [topic] from the past day.” Note that x_search is billed via xAI: every 1000 calls costs $5, plus model token fees—a once-a-day brief won't cost much, but keep it in mind.
Morning brief + weekly brief on alternating shifts. Same production line—just stretch the time window to 7 days and it becomes a weekly brief:
openclaw automations create "0 9 * * 1" \
"搜索过去 7 天 AI 领域的重要进展,汇总成一份中文周报。用中文回复;链接、代码和产品名保持原样。" \
--name "Weekly digest" \
--tz "Asia/Shanghai" \
--session isolated \
--announce \
--channel telegram \
--to "-1001234567890"Search results not good enough? Switch providers. A typo in tools.web.search.provider (e.g., typing brave as brvae) fails outright at config validation with no silent fallback—after fixing it, run openclaw config validate.
Want to watch specific pages rather than the whole web? That's the territory of the next lesson: conditional triggers wake Claw only when the thing that should change has changed.
Lesson Recap
✅ You can now:
- Build an AI daily news brief production line: cron sets the rhythm,
web_searchfinds the material, the prompt sets the output, and--announcedelivers into IM—configure once, get a brief every day - Configure a search provider for Claw: Kimi / MiniMax recommended for mainland China, or DuckDuckGo for a key-free option (must be selected explicitly)
- Use
runfor an immediate test run andrunsto check execution and delivery status, distinguishing "execution succeeded" from "delivery succeeded" - Use
edit --messageto tune the morning report's taste: time window, Chinese-language rules, output structure, length limit
Key configuration recap:
| Configuration | Recommended value | Description |
|---|---|---|
--tz | "Asia/Shanghai" | Timezone for cron evaluation; mandatory in mainland China |
--session | isolated | First choice for report tasks like the morning report; a brand-new session every time |
--announce --channel --to | Match by channel | Telegram group -100…; Slack/Discord channel:<id>; to send to yourself use --delivery-target owner |
| Search provider | Kimi / MiniMax (mainland China), DuckDuckGo (no key needed) | When no managed provider is configured web_search does not appear |
| Three elements of the prompt | Time window + Chinese-language rules + output structure | Language is not inferred; the rules must be written into the prompt |
Next lesson preview
The morning report watches "what's new across the whole web each day"; the next lesson Web monitoring: get notified the moment a page changes watches "whether that page has changed" — a condition trigger, where the script must say fire before Claw wakes up.
You will learn:
- Condition triggers
--trigger-script: scheduled checks + change detection, so no model turns are burned when nothing changed - How the trigger script's state memory (limit 16 KB) is used to "remember what it looked like last time"
- Active hours and yield strategies, so monitoring doesn't wake you up in the middle of the night
- The same set of delivery and troubleshooting commands shared with the morning report