A Second Brain for the Solo Consultant

A solo consultant is a business with one employee and a dozen open relationships. Every client, every prospect, every dormant contact who might wake up next quarter: all of it runs through one head. And that head is good, which is exactly the trap. You can hold most of it, most of the time, so you never build the systems a bigger firm would, and the leaks stay invisible. The follow-up you meant to send after the conference. The prospect who said “ask me again in the spring.” The client who mentioned, in passing, in October, the budget cycle that matters enormously right now. None of these failures announce themselves. They just quietly become revenue that went somewhere else.

There is a newer leak too. You already think out loud with AI: a proposal drafted in ChatGPT, a workshop plan worked through in Claude, a client’s org chart explained to both. Every one of those chats starts from nothing, so you paste the same background in again, and whatever you worked out stays in a chat history you will never search.

The big-company answer, hire ops, deploy a CRM, run account reviews, does not fit a business of one. Most consultants have a CRM they abandoned within a month, because it is an empty database demanding manual tribute from the same person doing the billable work. What fits is a memory that the tools you already use fill for you: a context engine.

The setup, concretely

Context Engine is one memory for all your AI tools. Claude and ChatGPT connect to it over MCP and read and save memories as you work. It does not do the thinking; your AI tools do. It remembers.

The memory is organised into engines, and for a solo practice the shape that works is simple:

• One engine per client. Everything about that engagement: the people, what was scoped, what was decided and why, what is open.
• One engine for your practice. Your methods, your pricing history, your positioning, your voice, prospects who are not clients yet.

Each engine has a purpose: a few sentences in your words saying what it is for, what belongs in it and what does not. AI tools read the purpose to decide where to save. “Everything about my engagement with Northpoint: their team, the scope, decisions and open work. Not my rates or other clients” is enough for a tool to put a Northpoint decision in Northpoint’s engine and a pricing lesson in your practice’s.

Then the connections. A connection is one signed-in app, reaching the engines you tick, each at Can read or Can read and save. Give each client its own. In Claude Desktop and claude.ai every connector is switched on in every chat by default, so the habit is to switch on only that client’s connector in each chat. It keeps a Northpoint conversation from quoting another client’s numbers, which for a consultant is the whole job.

From there nothing about how you work changes, which for a solo practice is not a convenience but the adoption criterion. You keep working in Claude or ChatGPT. When something is worth keeping, the tool saves it: a decision with its reason, a note, a task with a due date, a person, a meeting. Proposals, contracts and transcripts you already have go into Files: drop a file, a whole folder or a .zip, and text and PDFs become searchable.

Four habits that do the work of an ops hire

Prepared for every call. When you open a chat with a client’s connector on, your AI tool gets the brief: what that engine is for, what happened recently, and what changed since it was last there. Ask it what you need before the call and it searches by meaning across what the connection reaches. For a consultant this is not a convenience; it is the product. You are selling the feeling of being your client’s only client, and “picking up from the pilot scope we discussed in March” delivers it in one sentence.

The Monday look across clients. Keep one more connection for yourself that reaches every engine at Can read, and use it for a weekly review. Ask your AI tool who has gone quiet and what is open. An illustrative answer, with invented names:

Monday review, from your engines

Elif (Northpoint): last meeting on record 7 weeks ago. The
  project closed well; she mentioned a phase two "after
  summer". [Northpoint]
Daniel (prospect): said "ask me again in the spring". It is
  spring. [Practice]
Marta (Ferrow, former client): no recent contact on record.
  [Ferrow]

Open tasks due this week:
  Send revised scope to Northpoint, due Wed. [Northpoint]
  Invoice Ferrow for the final workshop, due Fri. [Ferrow]

Every result names its engine. Ask your tool to use only what is in the engines and to say “no recent contact on record” rather than guess a date; a list you trust is a list you open every Monday. Nothing is sent to anyone. It is a list for you to act on.

Proposals from real history. Your next proposal can draw on every similar engagement: what you scoped, what it cost, where it overran, what the deliverables actually were. Ask your AI tool to find the closest past engagements before it drafts, and keep your voice as identity memories in your practice engine so the draft sounds like you. The overrun history is the quietly valuable part: your own past is the only honest calibration data your pricing will ever get.

The dropped-thread net. Commitments in both directions become tasks: memories that are open or done, with an assignee and a due date. Context Engine does not send reminders; there is no nudge and no task board. The net is that the tasks exist at all, in the right engine, and that you look. The Monday review above is where you look.

Ask and Summaries, in the console

Two parts of the console earn their place in a solo practice.

Ask answers questions about the engine you have open, and every answer names the memories it came from. It does not browse the web, so when something is not on record it says so. “What did Northpoint agree to in the August call, and what did we promise in return?” gets an answer you can check line by line.

Summaries, a tab of Memories, is a written account of each day, rolled into weeks and months, from what was saved. At the end of a month a client’s monthly summary is most of a status report, and your practice engine’s is an honest picture of where your time went.

What changes, at three time scales

This week: preparation time shrinks and less slips. The follow-ups go out, the promises get kept, and clients start using the word “organised,” which for a solo consultant is code for “safe to recommend.”

This year: the practice gets calmer. The background hum of “what am I forgetting?” is a real tax on solo work, paid in evenings and in the reluctance to take on one more client. The threshold to watch for: the day you stop keeping a parallel system of sticky notes, because you trust the engines more than your notes.

This decade: the compounding asset. A consultancy of one produces a great deal of knowledge, methods, pricing history, what worked for which kind of client, and normally it evaporates, because there is no institution to hold it. Your practice engine becomes that institution: a searchable record of every engagement you have run, each decision with its reason. That makes raising your rates defensible and makes a first hire onboardable in days if you ever grow: invite them to the engines they need. And if a client relationship ends with the client wanting the record, you can hand that client’s engine over, and it moves onto their plan.

It is also yours in the practical sense. Export any engine as one ZIP of Markdown and JSON that opens without us. Your content is sent to models only to answer you and is never used to train them.

Tuning it

Knob	Default	Alternatives
Engines	one per active client, one for the practice	one per client tier, if your plan’s engine count is tight
Connections	one per client, plus one read-only review connection	one per client only, reviewing each engine in Ask
Access	Can read and save for client work	Can read for the review connection
Review	weekly, Monday	fortnightly for smaller rosters
”Gone quiet” bar	about six weeks for clients	longer for prospects and conference contacts

The per-tier quiet bar is worth stating in your review question: a client silent for six weeks is a signal, a conference contact silent for six weeks is a Tuesday. How many engines and connections you get depends on the plan; see pricing.

Failure modes, and the fixes

You stop looking. The system survives on one habit: the Monday review. If it lapses, shrink it rather than abandon it; three lines you read beat a thorough list you skip.

Things land in the wrong engine. A tool saves a pricing note into a client’s engine. Sharpen that engine’s purpose, then open the memory and Move to another engine: it goes with its history, and the client’s engine keeps only the fact that something moved.

Every connector is on at once. In Claude Desktop and claude.ai that is the default. Switch on only the client you are working on, every chat. It takes a second and it is the difference between a separated practice and a mixed one.

Personal and professional blur. Your purpose decides what belongs. Say in your practice engine’s purpose that personal matters do not belong there, and the tools have a rule to follow.

Build this yourself

The pieces, deliberately few, because a solo stack must be maintainable by its one user:

• An engine per active client and one for your practice, each with a purpose that says what does not belong.
• One connection per client, switched on one at a time, and one Can read connection for the weekly review.
• A starting pile: last quarter’s proposals, call notes and transcripts dropped into each client’s Files, so the memory is useful in week one.
• Ten minutes on Monday with the review question and the open tasks.

Start on the free plan at app.contextengine.com, and see pricing when you need more engines than that.