Solution · Intelligence

Second Brain — One Week, Three Human Gates

  • Telegram
  • Whisper
  • Ollama
  • Neo4j
  • Qdrant
  • Honcho
  • Frappe CRM
  • Listmonk
  • Windmill
  • DeerFlow
  • Claude Agent SDK
  • Postiz

The problem

Useful details are lost faster than anyone can type them. A name from a site visit, a promise made in a corridor, a battery that will be past its end-of-life date by winter: each is lost between the moment it happens and the next time someone sits at a desk. The tools sold for this assume a structured capture session that rarely happens. Automated outreach products fill a prospect's inbox with mail that nobody read before it was sent. I wanted neither, so I built the opposite.

What I built

This is a sales-intelligence system that I control from one phone. It runs on a VPS and a home lab, and it sends nothing to a prospect until a person taps a button. At seven each morning one Telegram message reports what the lead scorer and the equipment monitor did overnight. A separate cron job, outside the workflow engine, confirms that the engine itself is still running. During the day I send a thirty-second voice note. The home GPU transcribes it, and a local model pulls out the names, sites, equipment and standards. The result is written three times, to the graph, the vector store and the CRM, and each copy points at the other two. A follow-up draft arrives on the phone with two buttons and waits up to a day. If I do nothing, it is skipped. Whatever I tap, a second, independent lock must also read true before any mail leaves. Two crawlers look for prospects. One searches the public web with Anthropic's model, and the other is sandboxed behind a metered proxy. Neither can write to the CRM. Their finds wait in a quarantine queue until I press Approve in the browser, which commits a find to all three stores at once or rolls the write back. A sequencer wakes every ten minutes, finds leads that have changed stage and picks one of eighteen templates along four axes. On Monday morning the phone offers three episode topics, ranked from what prospects have said. After one tap the script is written from sources it can cite, narrated in a cloned voice, animated scene by scene and staged for a Wednesday contact sheet.

The logic flow

Click any node to see what it does, what it passes on and where a person must approve.

Loading diagram...

What it returns

A voice note reaches the graph, the vector store and the CRM as a lead in about 65 seconds, measured. There are eighteen nurture templates across stage, engagement, language and brand, and none goes to a lead before I have approved that lead once. The whole human workload is three deliberate taps a week: to send an email, admit a prospect and publish an episode. Current status, from the system's own notes: every video stage up to staging has run from start to finish. The public publishing step is deployed and gated, and has not yet run against a public channel.

Why it fits

I did not build this system to make my old process faster. Adding AI to the strategy and workflow a firm already has only makes the firm depend on both more, and a sales assistant that writes my emails faster would do exactly that. The real change is in the knowledge gap that paid for the last twenty years of this work. I knew the standards, the sizing method and the failure modes, and the customer did not. That gap will not reopen.

A prospect's own assistant now arrives with the load, the string and a defensible first estimate, before anyone picks up a phone. The firm is left selling either a number the customer already holds, or the judgement that goes with it. The assistant cannot produce a record of what the assets did: the discharge test that fell short in the ninth year, or the two sites where the ventilation calculation was right and the installation was not. Second Brain holds that record. The system writes it three times, to the graph, the vector store and the CRM, and each copy points at the other two. I answer from the record, not from memory.

A five-year version, as an illustration and not a plan: make your systems reachable by your customers' own AI tools. That means an API their tools can query, under the access rules and the audit trail you set. It means a record of what your assets did that is worth querying, and a price on the answer, not on the drawing. As their tools improve, those tools come back to me more often, which is the opposite of what most firms fear. Three taps a week are the whole human workload here, and each one is a decision about a relationship, not about a document.

See it run

Want one of these?

Every part of it is open source and you can host it yourself. I bring the failure record and the judgement about where your gates should go.