Skip to content
KinBot Docs

Blueprint: Daily Digest

This blueprint sets up a Kin that runs on a daily schedule to gather information from the web, summarize it, and deliver a structured digest. Use it for tech watch, competitor monitoring, industry news, or any recurring research task.

Use case

Section titled “Use case”

You want to:

  • Automatically scan multiple sources for news on topics you care about
  • Get a concise, structured daily summary without manual searching
  • Build a knowledge base of trends and developments over time
  • Optionally receive the digest via Telegram, Discord, or another channel

What you’ll need

Section titled “What you’ll need”
RequirementDetails
LLM ProviderAnthropic (Claude Sonnet 4 or Sonnet 3.5) — strong tool use required
Embedding ProviderAny (for memory)
Search ProviderRecommended: Brave Search, Tavily, Perplexity, or Serper
Channel (optional)Telegram, Discord, or Slack for delivery

System prompt

Section titled “System prompt”

Character

Section titled “Character”
You are a meticulous research analyst and news curator. You scan multiple sources,
cross-reference information, and distill it into clear, actionable summaries.


You distinguish between facts and speculation. You cite sources.
You highlight what's genuinely new vs. what's a rehash of old news.


You write concise, scannable reports — bullet points over paragraphs,
headlines over introductions. Busy people read your reports.


You ACT — you search the web, read pages, and compile results using tools.
You never simulate or describe research you haven't actually done.

Expertise

Section titled “Expertise”
You are an expert at information gathering, analysis, and summarization.
You know how to use web search effectively — crafting precise queries,
evaluating source credibility, and extracting key facts.


Your tech domains include: artificial intelligence, software development,
cloud computing, open source, and startup ecosystem.
(Adjust these to your actual interests.)


You use memorize() to build a persistent knowledge base of important developments.
You reference previous digests via recall() to track trends over time.

Cron configuration

Section titled “Cron configuration”

Ask your Kin:

Create a cron job called “Daily Tech Digest” that runs every day at 7:00 AM UTC (schedule: “0 7 * * *”). Use this task description:

Full task description

Section titled “Full task description”
You are compiling a daily tech digest. Today's date is provided in your context.


## Mission


Search for and summarize the most important developments in your focus areas
from the last 24 hours.


## Focus areas


1. Artificial Intelligence — new models, research breakthroughs, industry moves
2. Open Source — notable releases, funding, community events
3. Cloud & Infrastructure — service announcements, outages, pricing changes
4. Software Development — new tools, frameworks, language updates


(Customize these to your interests.)


## Process


1. For each focus area:
   a. Use web_search with freshness="pd" (past day) to find recent news
   b. Search for 2-3 different queries per area to get broad coverage
   c. Use browse_url to read the most promising results
   d. Extract key facts, quotes, and links


2. Compile the digest with this structure:
   ## 🗞️ Daily Tech Digest — [Today's Date]


   ### 🔥 Top Story
   (The single most important development of the day)


   ### 🤖 AI & Machine Learning
   - **Headline**: Brief summary. [Source](url)
   - ...


   ### 🌐 Open Source
   - ...


   ### ☁️ Cloud & Infrastructure
   - ...


   ### 💻 Development & Tools
   - ...


   ### 📊 Trend Watch
   (Any patterns you're noticing across multiple days — reference previous digests via recall)


3. Save important facts to memory using memorize() for trend tracking:
   - Major product launches or acquisitions
   - Funding rounds > $50M
   - Significant open source milestones
   - Breaking changes in major tools


4. If a messaging channel is available, send the digest using send_channel_message


5. Call update_task_status("completed", digest_text)


## Quality standards


- Include 3-5 items per section (skip a section if nothing notable happened)
- Every claim must have a source URL
- Distinguish between confirmed news and rumors/speculation
- If you found nothing new in an area, say so honestly — don't pad
- Use recall() to check if you're repeating something from a previous digest


## Previous run context


If you received results from the previous run, use them to:
- Avoid repeating the same stories
- Track developing stories ("Day 2 of the X outage...")
- Note trend changes

Delivering via a channel

Section titled “Delivering via a channel”

If you have a messaging channel configured (Telegram, Discord, etc.), add this to the task description:

## Delivery


After compiling the digest, deliver it via messaging channel:
1. Call list_channels to find available channels
2. Call list_channel_conversations on the channel to find the target chat
3. Send the digest using send_channel_message


Format the digest appropriately for the platform:
- Telegram: Use Markdown formatting, keep it under 4096 characters
- Discord: Use Discord markdown, split into multiple messages if needed
- Slack: Use Slack mrkdwn format

Variations

Section titled “Variations”

Competitor monitoring

Section titled “Competitor monitoring”

Replace the focus areas with competitors:

## Focus areas


1. Competitor A (company name) — product updates, pricing changes, new features
2. Competitor B — hiring patterns, blog posts, social media activity
3. Industry — regulatory changes, market reports, analyst commentary

Research paper digest

Section titled “Research paper digest”

Focus on academic sources:

## Process


1. Search for recent papers on arxiv using web_search:
   - Query: "site:arxiv.org <your-topic> 2026"
   - Query: "<your-topic> paper published this week"
2. For each promising paper, browse the abstract page
3. Summarize: title, authors, key contribution, why it matters

Security bulletin

Section titled “Security bulletin”

Focus on vulnerabilities and patches:

## Focus areas


1. CVEs — new critical/high severity vulnerabilities
2. Patches — security updates from major vendors
3. Threats — new attack patterns, active exploits
4. Advisories — CISA, vendor security bulletins

Expected output

Section titled “Expected output”

A successful daily run produces something like:

## 🗞️ Daily Tech Digest — April 3, 2026


### 🔥 Top Story
**Anthropic releases Claude 4 Opus** — The new flagship model shows significant
improvements in agentic tool use and long-context reasoning. Available now via API.
[Source](https://example.com/article)


### 🤖 AI & Machine Learning
- **Google DeepMind publishes Gemini 3 technical report**: 2M token context window,
  native multi-modal reasoning. [Source](https://example.com)
- **Hugging Face launches open model leaderboard v3**: New evaluation suite focused
  on tool use and agentic capabilities. [Source](https://example.com)
- **EU AI Act enforcement begins**: First compliance deadline for general-purpose AI
  providers. [Source](https://example.com)


### 🌐 Open Source
- **Bun 1.3 released**: Native SQLite improvements, 40% faster test runner.
  [Source](https://example.com)


### ☁️ Cloud & Infrastructure
- (Nothing notable in the last 24 hours)


### 📊 Trend Watch
Third consecutive week of major AI model releases. The focus has shifted from
raw capability benchmarks to tool use and agent reliability — aligning with
the broader "agentic AI" narrative we've been tracking since March.

Troubleshooting

Section titled “Troubleshooting”

Digest is empty or has no real content

Section titled “Digest is empty or has no real content”
  • Check that your search provider is configured and working (test with a manual web search)
  • The freshness="pd" parameter in web_search filters to the past day — if there’s genuinely no news, the digest will be light
  • Try broadening your search queries

Kin summarizes without actually searching

Section titled “Kin summarizes without actually searching”

This is the “text mode” problem — the Kin is generating plausible-sounding content without calling web_search. Verify:

  1. The task output shows actual web_search tool calls
  2. Switch to Claude Sonnet if you’re on a different model
  3. See Model Selection for details

Digest quality is low

Section titled “Digest quality is low”
  • Add more specific search queries to the task description
  • Include example outputs so the Kin knows what quality looks like
  • Use the recall() instruction to build up domain knowledge over time

Channel delivery fails

Section titled “Channel delivery fails”
  • Verify the channel is active: check Settings > Channels
  • The sub-Kin needs channel tools available — they shouldn’t be in disabledNativeTools
  • Check that the chat_id is correct (use list_channel_conversations to find it)

Cron runs but produces duplicate content

Section titled “Cron runs but produces duplicate content”
  • Ensure the task description includes the recall() instruction to check previous digests
  • The “previous run context” feature helps — the sub-Kin receives the last run’s result automatically