Skip to main content

Second Brain

The context arrives before the question does.

Second Brain is an agentic workflow system I built and have run daily for many months. It connects email, calendar, meetings, tasks, and documents, then does the synthesis work on its own: morning briefings, meeting notes filed and turned into tasks, drafts written in the right voice, and a knowledge base I can question in plain language.

Book a 30-minute callSee a morning briefing

Free introductory conversation.

At a glance

In daily use
Built for
A consultant working across several clients and a fragmented toolset
Connected to
Outlook, calendar, ClickUp, Fathom meeting recordings, Google Drive, Gamma
Runs on
n8n for orchestration, Weaviate for memory, OpenAI and Anthropic models for reasoning
Final call
Me. Every output is labeled Verified or Inference so I know what to check.

The problem

Information gatherer, or insight architect?

Consulting work runs across a fragmented ecosystem: communication in Outlook and Slack, projects in ClickUp and client systems, documents in Drive, meeting intelligence in Fathom and handwritten notes. The cost is time spent connecting dots by hand: finding the governance protocol for one client, recalling the minutes from three weeks ago on another.

General-purpose AI tools don't solve this. They lack the context, and they lack the security boundaries that client work requires.

Second Brain is not a chatbot. It's a set of automated workflows that proactively connect these tools and hand me the context before I ask for it. The chatbot is one capability among several, and it sits on top of everything the workflows have already gathered.

What it does

Six workflows, mostly running without me.

Each runs on a schedule or a trigger. I read the results; I rarely start them.

  1. Daily briefing

    Aggregates calendar context, relevant emails, and outstanding ClickUp tasks into one morning digest. The day starts with a synthesized plan, not inbox triage.

  2. Weekly briefing

    Scans all client and internal inputs to recap progress and flag blockers and opportunities, so risks surface before they hit a timeline.

  3. Email triage

    AI-assisted classification keeps attention on the messages that matter and files the rest.

  4. Meeting notes and action items

    Categorizes each meeting by client, sends the summary to a ClickUp doc, and creates a ClickUp task for every action item, assigned to the person responsible.

  5. Context-aware drafting

    Drafts replies to tasks and emails using historical client tone and verified documentation. Produces deliverables in different formats on request.

  6. Industry and competitor intel

    Scans public sources for industry trends, including AI, and competitor publications, then summarizes what changed.

And the chatbot on top

Natural-language access to the whole knowledge base: meetings, emails, documents, and everything the workflows above have filed. Equipped with MCP tools, it can also create outputs directly in the tools and formats the work requires.

Sample

What a morning briefing looks like.

Illustrative sample. No client material appears on this site.
Daily briefing · TuesdayGenerated 6:30 am · 3 sourcesIllustrative sample, generated from synthetic context. No client material.
Today's meetings

Verified10:00 Client A steering call. Last meeting's two open actions are still unassigned in ClickUp.

InferenceTheir Friday email about timeline likely comes up. Draft response is in your Drafts folder.

Email needing a reply

Verified2 of 41 overnight messages classified as action-required. Both from Client B, both about the data retention clause.

Tasks due

Verified3 ClickUp tasks due today, one overdue from yesterday’s meeting: “Send revised scope to Client A.”

Two labels

Verified means the statement is sourced from a document, message, or record in the system.

Inference means the model generated it. I know exactly which lines to check.

Also logged

Every AI-generated output is written to an audit trail with its inputs and timestamp. If a briefing was wrong, I can see why.

Architecture

Orchestration, memory, and reasoning as separate layers.

The same retrieval-augmented, agentic architecture I recommend to clients, running on my own work first.
  1. Sources

    Outlook, ClickUp, Fathom, Google Drive

    Connected through their APIs. New mail, meetings, tasks, and files trigger workflows as they arrive.

  2. Orchestration

    n8n

    Runs every workflow: the schedules, the triggers, the routing between tools, and the logging.

  3. Memory

    Weaviate

    A semantic knowledge graph. Everything is stored with its source, date, and client segment attached.

  4. Reasoning

    OpenAI and Anthropic models

    Summarize, classify, draft, and answer, working from retrieved context. Outputs return to ClickUp, email, or the chat.

Governance

Designed governance-first, not bolted on.

Public AI tools can't hold client boundaries. This one was built around them.

Client isolation

Data is segmented by client ID. Material from one client never informs an output for another.

Hybrid boundaries

Sensitive personal data stays within secure boundaries. Only anonymized or necessary context is sent to a model for reasoning.

Audit trails

Every AI-generated output is logged with what went in and when.

Strict labeling

Every statement is tagged Verified (sourced from records) or Inference (model-generated), so the reader knows when to check.

What it costs

Running costs for one person.

Usage-based items vary with volume. The fixed floor is under $50 a month; the working total for my own use has stayed under $100.

  • n8n private cloud

    Monthly$15
    BasisFixed

  • OpenRouter (model access)

    Monthly$34+
    BasisUsage

  • Datalab (file conversion)

    Monthly$25+
    BasisTiered

  • Weaviate Cloud

    Monthly$0+
    BasisUsage

  • Google Cloud API

    Monthly<$2
    BasisUsage

  • OpenAI embeddings

    Monthly<$1
    BasisUsage

  • Fathom, ClickUp, Microsoft Graph APIs

    Monthly$0
    BasisFree tiers

Optional: ElevenLabs API for voice, $20+ a month, usage-based. Figures as of December 2025.

Limits

Where it is, and what's still open.

Stated so you can judge whether the pattern fits your team.

Proven for one user.

The workflows are validated on my own work. Team deployment is the next phase: pilot by feature set, then with teams that have high-value use cases.

Inference still needs a reader.

The labels tell me what the model generated. They don't make it right. Anything tagged Inference gets checked before it leaves my desk.

Vendor terms are ongoing work.

API exposure, data processing agreements, retention policies, and encryption with each vendor are reviewed as the system grows past one user.

The same problem, at team scale

Scattered context is a workflow problem.

Which tools connect, who sees what, and where outputs land are design questions. The Sprint maps them before anything gets built.

Before you automateWorkflow Design SprintOne workflow mapped as it is and as it should work, plus a specification someone can build.2 weeksFixed feeAnother buildContent Resonance EngineStructured content review for marketing teams, with an editor keeping the final decision.Working systemFeatured build

Bring the tools your team switches between all day. We'll talk through what could connect, and what should stay apart.

Free introductory conversation. No preparation required.

Book a 30-minute call