The point at which a small agency stops being able to run on heroics is somewhere around fifteen active clients. Below that, the principal can hold most of the context in their head. Above it, things start dropping. The brief from client seven hasn't been re-read in a month. The status update for client twelve goes out generic because nobody had time to pull the actual work into it. The new strategist who joined in March still doesn't know which clients prefer email over Slack.
This guide is about a working setup for an agency at that size. The CRM and the project management tool still do their jobs. The vault is where the thinking, the soft context, and the working playbook live — readable by the whole team and queryable by an agent.
The shape of a small agency that holds at fifteen-plus clients
The failure mode at scale is consistent. Each client has their own set of artifacts spread across email, Slack, Drive, Figma, Notion, and someone's notebook. When the principal isn't on the call, the context evaporates. New hires take three months to ramp because the institutional knowledge isn't documented anywhere. By the time the agency is at twenty clients, the principal is still the only person who can answer most operational questions, which means the agency hasn't actually scaled — it's just expanded.
The shape that holds:
- One top-level page per client, with sub-pages for everything that pertains to them — discovery, contracts, calls, deliverables, internal notes.
- A practice library — the templates, frameworks, past deliverables, and the agency's accumulated playbook.
- A few small tracker databases — active engagements, deliverables, invoices, promises — embedded in pages people actually visit.
- An agent that reads across all of it so the daily questions get answered from the record, not from the principal's memory.
Note the important constraint: today, the Docapybara vault is single-user by design. That means the operational reality for agencies is that each person on the team has their own vault. The shared playbook lives wherever the agency keeps shared docs (Drive, Notion, etc.), and the per-person vault is where each strategist or account lead organizes the clients they handle. We're describing the personal-vault shape; the team-coordination layer still happens in your existing tools. For the broader version of the same setup that's not agency-specific, see AI notes for client work.
One client per page, organized for fast retrieval
The structure that scales is dead simple. Each client gets a top-level page in the strategist's vault. Underneath, sub-pages for every kind of artifact. Page nesting has no depth cap.
A typical client branch:
- Acme Corp
- Brief and discovery
- Signed contract (PDF)
- Calls
- 2026-02-14 kickoff
- 2026-03-01 mid-project sync
- 2026-03-22 review
- Deliverables
- Phase 1 audit
- Phase 2 strategy doc
- Phase 3 implementation
- Pipeline (database)
- Invoices and payments
- Internal notes (your private observations about the relationship)
The benefit isn't visual neatness — it's that the agent can scope its work to one client by reading from one branch. "Read everything under Acme. Draft a status update I can send tomorrow covering progress, open questions, and what I need from them." The agent doesn't have to guess what's relevant; the structure tells it.
Plain markdown matters here. The pages aren't database rows with attachments hanging off — they're text files the agent can read end-to-end. That's why "summarize where this engagement stands" actually works as a prompt, even when the engagement has six months of history.
Discovery and the first proposal — turning intake into something workable
The intake on every new client is chaotic. A discovery call. Three follow-up emails. A shared folder of "everything we have." A loose Loom recording. By the time someone sits down to write the proposal, they've reviewed all of it twice and probably forgotten one important constraint.
The agent helps with the synthesis pass. Drop the discovery call recording — it transcribes with speaker labels so you know which stakeholder said what. Save the email thread as a page. Drop in any PDFs the client sent — they auto-convert to markdown via docstrange, so attachments become searchable text instead of opaque files.
Then ask: "Pull the five things the client said are non-negotiable. Pull the constraints they mentioned around budget, timeline, or team capacity. List the three areas where they contradicted themselves between the call and the email thread. Draft a proposal that mirrors the structure of my last three accepted proposals from the practice library."
You get a proposal draft that's grounded in what they actually said and the shape of what's worked before for the agency. You spend half an hour editing instead of three hours assembling. The client gets a proposal that quotes them back to themselves, which tends to land better than one written from scratch.
The practice library — the agency's accumulated playbook
The thing that distinguishes a five-year agency from a five-month one is the playbook. The frameworks the team reaches for. The diagnostic questions that get to the truth fast. The proposal structure that closes. The internal templates for status updates, kickoff agendas, end-of-engagement memos.
Most small agencies carry the playbook in the principal's head and slowly forget the parts that don't get used weekly. A "Practice library" section in each strategist's vault makes the personal version of it explicit. Pages for templates. Pages for frameworks. A "lessons learned" log where you write down what you'd do differently next time after each engagement wraps.
When a strategist asks the agent to draft a proposal or a status note, it reads the relevant template page first. The output respects the strategist's specific approach, not a generic one. The drafting workflow itself is covered in more depth in drafting emails, proposals, and newsletters inside your notes app.
For the agency-shared version of the playbook — the canonical templates, the frameworks the whole team uses, the brand standards — that still lives in the team's shared docs (Drive, Notion, etc.). Each strategist's personal vault holds the working version they actually use day-to-day, which references the canonical playbook but extends it with their own observations.
Calls and the speaker-label dividend
The single biggest improvement to agency work with many meetings is letting something else handle verbatim capture.
The pattern: record the call (with the client's knowledge), drop the audio onto the relevant call page in the client's branch, and the transcription runs with speaker labels. The strategist is free during the call to actually be present — listening, asking the second-order question, watching for what's not being said. Afterward, they skim the transcript instead of straining to remember.
The transcript is searchable, which matters more than people expect. Three months later when the client asks "what did we say about the rollout?", the strategist asks the agent: "Find every time we discussed the rollout across calls with this client. Pull the relevant exchanges with who said what." The moments come back, with speaker labels, in plain text. No more "I think we said early Q3?"
For agency work specifically, the speaker labels matter because client meetings have multiple stakeholders. You want to know what the CEO committed to versus what the marketing director committed to versus what the agency committed to. A wall of dialogue without attribution is just dialogue. The same commitment-capture move applies during all meetings — see how to capture action items so they actually get done for the meeting-side workflow.
Deliverables, drafts, and the version question
Agency deliverables go through revisions. The first draft. The version after the client's notes. The version after a second pass. The "final" that became "final-v2" two days later. Most agencies manage this in a folder of files with increasingly desperate naming conventions.
In a markdown vault, every revision is its own page under a "Deliverables" sub-page, with the date in the title. The agent can compare them on demand: "Show me what changed between Draft 2 and Draft 3 of the strategy doc, in plain English, focused on the recommendations section." That question is hard to answer fast in a folder of Word files.
For deliverables with structured information — recommendation matrices, prioritized backlogs, audit findings — an inline database inside the deliverable page works better than a static table. Embedded directly via the :::database::: directive. The client's edits become row updates instead of "track changes" hell. The whole deliverable stays in one document.
Cross-client trackers and the Friday review
Across many clients, each strategist needs a few small databases the agent can read alongside everything else. Not a separate project management tool — a few embedded tables in pages they actually visit.
A project tracker with rows per active engagement: client, current phase, next deliverable, due date, status, hours logged this month. An invoices database (if the strategist handles billing): client, amount, sent date, paid date, status. A "promises" database for the things committed to that aren't formal deliverables — small follow-ups, intro emails, decks said-they'd-share, the article said-they'd-send.
Once a week, scan the running databases. To shorten the review, ask the agent first: "What did I commit to this week across all clients? What's overdue? What's coming due in the next seven days? What invoices are unpaid past 30 days?" You get a focused short list in seconds. The hour-long Friday reckoning becomes ten minutes.
For storage and upload limits if the vault is holding many large client recordings, see pricing.
Onboarding new strategists and the institutional memory problem
A real practice note for agencies: when a new strategist joins, the personal vault structure helps with the operational ramp. The new hire gets the agency's shared playbook from the team's central docs, then sets up their own vault for the clients they're going to handle.
Each client they take over starts with a handoff page from the strategist who handled them before. The handoff page captures the running context — the relationship dynamics, the open commitments, the things to know. The new strategist reads the handoff, sets up their client branch, and asks the agent for a synthesis of the past six months: "From the handoff page and any historical artifacts I've been given, summarize where the Acme engagement stands. List the three things I should know before my first call." Walking into the first call with two pages of grounded context, instead of a vague "they're a brand client" briefing, makes the early weeks calmer.
A boundary on what AI should and shouldn't do
A practical note: the agent helps with assembly, structural drafting, and surfacing material from the vault. It doesn't replace strategic judgment, client-relationship work, or the actual taste that makes an agency's work good. Drafts that are AI-generated end-to-end have a flat tone you can hear in three sentences and they don't develop the trust that long-term client relationships run on.
The shape that works: AI does the chores around the work — pulling references, suggesting structures, drafting first passes from the strategist's notes. The strategy is the strategist's. The relationship is the agency's. The vault is the workshop, not the substitute for either.
A calmer way to run an agency at scale
Running a small agency at fifteen-plus clients doesn't have to feel like spinning plates. The plates are real, but most of what makes them feel precarious is the search-and-reassembly cost between them. Each strategist's personal vault holds the working context for the clients they handle. The team's shared playbook lives where it always has. The agent does the boring searches and the structural drafts. Saturday morning becomes ten minutes of housekeeping instead of a guilty hour.
Try Docapybara free — sign up, pick one active client you handle, drop in their brief, the contract, and the last two call recordings, then ask the agent for a one-page status summary you could send tomorrow.