A small team that serves clients, whether it is an agency, a trade shop or an appointment business, builds knowledge everywhere: client requests arrive by email, decisions happen in chat threads, project files scatter across a shared drive, and the owner's head holds the rest. When someone needs an answer six months later, the search starts over.
Quick answer
The best shared AI knowledge base for a small team is one vault of plain-markdown documents the whole team works in, with an assistant that answers from it and shows which files it read. Mycel is built this way. It is the business's memory and the owner's desk, made for trade shops and appointment businesses, and the apps they run for customers, reviews and posts read from it.
At a glance
- What it is: Mycel is the business's memory and the owner's desk: one shared vault of plain-markdown folders the team works in, with an assistant that answers from it and shows which documents it read.
- Who it is for: trade shops of about 3 to 25 people (HVAC, plumbing, electrical, roofing, landscaping and lawn care, cleaning, pest control) and appointment businesses (salons, barbers, lash and brow studios, med spas, tattoo studios, groomers).
- How it connects: Mycel is the brain of one connected stack. Relay (appointment businesses) or TradesBackbone (trade shops) holds customers, jobs and conversations; ReputeLift asks for reviews and replies; SynthPrism drafts and publishes posts. The business connects the apps it uses and chooses which folders each one may read.
- What it is not: Mycel is not a CRM. Customer records stay in Relay or TradesBackbone, one click away.
- Team features: team chat with @assistant in any thread, four roles (Owner, Admin, Member, Viewer), private folders, document history with restore, and Today, one list of what is waiting on a person across the connected apps.
- Price: Free $0 for one person; Personal $69 a month; Personal Plus $119 a month; Business $199 a month for 3 people, plus $59 a month per extra seat up to 25. Annual billing is 12 months for the price of 11. See pricing.
- Your data: the whole vault exports as markdown at any time.
Key Takeaways
- A shared AI knowledge base for an agency team should center on one vault of plain-text documents the whole business works in, with an assistant that answers from it and cites which documents it read.
- The vault needs role-based folder permissions, document history with restore, and the ability to accept what the team learns from email, voice input and connected business apps.
- The assistant should be callable from team chat channels, and any change it offers to make should wait for a person's approval first.
- Export the entire vault as markdown at any time so your business memory stays portable, with each file carrying its created and updated dates in a frontmatter header.
- On Mycel, team roles, private folders, document history and team chat come with the Business plan: $199 a month for 3 people, plus $59 a month per extra seat up to 25.
Why Agency Teams Need a Shared Knowledge Base
Client work compounds knowledge. Each job or client generates requests, notes, quotes, scope changes, preferences and lessons learned. The team makes dozens of small decisions every week—brand voice guidelines, how to handle a recurring request, which vendor to use for a specific service—and those decisions drift into email threads, chat history and individual notebooks.
When the person who handled that client last year is out, or a new team member joins, the hunt begins. Someone asks in the #general channel, checks three shared folders, scrolls old chat threads, and eventually pings the owner. The answer exists, but it's not reachable.
A shared AI knowledge base solves this by putting the business memory in one place the assistant can read. Instead of searching by filename or keyword and hoping, the team asks a question in plain language and gets an answer drawn from the vault, with a citation showing which documents the assistant consulted. If the assistant read nothing from the vault to form the reply, that fact is marked clearly.
What to Look for in a Shared AI Knowledge Base
A Single Vault the Whole Team Works In
The foundation is a shared vault where the team files and organizes what the business knows. This should not be a siloed knowledge base the assistant manages and the team occasionally consults—it should be the working set of folders and documents the team opens, edits and references every day.
Look for plain markdown storage. Markdown is human-readable, works in any text editor, and exports without conversion. Proprietary formats lock your memory inside one vendor's platform. With markdown, each document is a .md file you can grep, version-control or migrate if the tool changes.
The vault should accept input from multiple channels: manual writing, email forwarding, voice notes transcribed and filed, and connected business tools that drop structured notes directly into folders. If every piece of knowledge requires someone to copy-paste it into a web form, adoption stalls.
An Assistant That Reads the Vault and Cites Its Sources
The assistant's job is to surface what the team already documented. When someone asks a question, the assistant should search the vault, read the relevant documents, and answer from them—then show which files it consulted.
A reply that read nothing from the vault should be marked that way. This distinction matters: an answer synthesized from the assistant's training data is a guess; an answer citing three specific documents from your vault is institutional memory.
The assistant should be callable from team chat. Typing @assistant in a channel thread and getting an answer in-line, visible to everyone in that conversation, turns the knowledge base into a teammate rather than a separate app someone has to remember to check.
Role-Based Folder Permissions
Not every document belongs to the whole team. A small agency typically has client work that everyone touches, financials and contracts that only owners and admins should see, and sensitive client information that should not be visible to part-time contractors or viewers.
The vault should support at least three permission levels: everyone, team members but not viewers, and owners and admins only. Each folder can be set independently, so the team can file openly by default and lock down the exceptions.
Document History and Restore
People make mistakes. Someone edits a process doc and accidentally deletes a paragraph. The assistant files a decision note with a detail wrong, or a team member approves a change that shouldn't have gone through.
Every document should keep a history of who changed what and when, with the ability to restore a previous version. If the assistant filed something, undoing it should be straightforward.
Approval Gates for Assistant Actions
An assistant that can directly change records in connected tools, send messages on behalf of the business, or edit documents without review introduces risk. A better design is one where the assistant may offer to file a decision note, add a task, or update a record—and the action waits in an approval queue until a person confirms it.
This keeps the human in the loop without forcing manual data entry. The assistant does the drafting and structuring work; the owner or admin checks it and approves or discards.
Export the Entire Vault as Markdown
Your business memory should not be held hostage. The vault should export in full, at any time, as a folder of markdown files. Each exported file should carry its created and updated dates and any metadata in a frontmatter header, so the export is not just a text dump—it's a portable, dated archive of the current version of every document.
The export should be the current version of each document, not the full version history. The history belongs in the live system where it can be browsed and restored interactively; the export is a snapshot for backup, migration or offline reference.
How a Shared AI Knowledge Base Fits Into the Agency's Daily Work
Filing What the Team Learns
Knowledge enters the business from multiple directions. A client sends project requirements by email. The team settles a decision in a chat thread. The owner captures a voice note walking between meetings. A connected business app logs a client interaction or closes a project milestone.
A working knowledge base should accept all of these. Email forwarding turns an inbox into an intake channel. Voice capture with transcription turns verbal notes into searchable text. Connected apps can file what the business learns there into the vault, so less has to be copied by hand.
The vault becomes the place where what the business learns accumulates, regardless of where it started.
Answering Questions Inside Team Chat
Team chat is where quick questions happen. "What was the deliverable timeline we agreed to for this client?" "How do we handle rush requests?" "What's our standard hourly rate for design revisions?"
If the knowledge base is a separate app, the team has to context-switch, search, return and paste the answer. That friction means questions get answered from memory—which might be outdated—or left hanging.
A better design integrates the assistant into team chat. In a business workspace with channels and direct messages, typing @assistant in a message thread calls the assistant into that conversation. It answers from the vault, cites the documents it read, and the reply appears in-line where the whole channel can see it.
The assistant may offer to file a decision note or add a task based on the conversation, but the action does nothing until someone approves it. The chat remains the source of truth for the discussion, and the vault remains the source of truth for decisions that should persist.
Surfacing What's Waiting Across Connected Apps
Small teams often use separate tools for customer conversations, jobs, reviews and social posts. Each app has its own inbox or dashboard, and work that needs attention scatters across tabs.
A unified Today view—what is waiting on you across all connected apps, with the source app, the reason it surfaced, how long it has waited, and a link straight to it—turns multiple dashboards into one starting point each morning. This belongs on the Home screen of the knowledge base, so checking what needs action and asking the assistant a question happen in the same place.
Maintaining the Vault Without a Librarian
A common objection to knowledge bases is that they require a dedicated person to organize, tag and curate. Small agencies don't have that person.
The design that works is one where the team files into folders as they go, the assistant helps surface what's already there, and permissions keep sensitive material contained without requiring elaborate taxonomy. Folder structure can stay simple—Clients, Projects, Processes, Decisions, Brand—and search combined with AI question-answering handles the rest.
Document history with restore means a mistake is fixable, not permanent. The assistant filing notes and offering tasks reduces manual data entry. The result is a knowledge base that the team maintains as a side effect of doing the work, not as a separate job.
Comparing Shared Knowledge Base Approaches for Small Teams
| Approach | What It's Built For | Access Model | Assistant Integration | |----------|---------------------|--------------|----------------------| | Shared markdown vault with AI | Central business memory the team works in daily, with an assistant that answers from it and cites sources | Team vault with role-based folder permissions; connected apps can file into shared folders | Assistant callable in team chat, answers cite vault documents, offers to file or add tasks pending approval | | Helpdesk knowledge base | Customer-facing support articles and agent answer scripts for support teams | Public or agent-only articles; typically not a working document set for the internal team | Built for support ticket deflection, not internal business memory or team question-answering | | Collaborative wiki | General team documentation and process pages editable by many people | Open editing for the team; permissions vary by platform | Assistant features vary widely; many rely on keyword search rather than document-reading AI | | Cloud storage with search | File storage and sharing with folders, versions and keyword search | Folder and file-level sharing; typically no role-based vault permissions | AI search features emerging; most do not cite which specific documents were consulted for an answer |
A shared markdown vault with an integrated assistant fits small client-serving teams that need one place to file, organize and retrieve business memory, with the assistant helping the team find answers rather than replacing the team's judgment.
What Mycel Brings to a Small Team's Knowledge Base
Mycel is the memory a small business runs on and the owner's desk. It provides a shared vault of plain-markdown folders the team works in, with an assistant that answers questions from those documents. Each answer shows which vault documents it read, and a reply that read nothing from the vault is marked that way.
The Martello apps a business connects—Relay, TradesBackbone, ReputeLift, SynthPrism—can file what the business learns into the vault. Customer records themselves stay in Relay and TradesBackbone, one click away. Relay's and TradesBackbone's AI read only the folders the owner shares with them; ReputeLift's AI reads the business vault; SynthPrism's assistant reads it for brand voice and business facts.
The owner's main sections are Home, Assistant, Team, Capture, Approvals and Vault. Home opens on Today, a list of what is waiting on the person across the connected apps, with which app it came from, why it surfaced, how long it has waited and a link straight to it.
Team chat in business workspaces includes channels (every business has #general) and direct messages between the people in the business. Typing @assistant in a message gets an answer in the thread from the business vault, and the assistant may offer to file a decision note or add a task, which does nothing until a person approves it.
A business has four roles: Owner, Admin, Member and Viewer. A folder can be kept to everyone, to team members but not viewers, or to owners and admins only. Every document keeps a history of who changed what and when, with restore. Something the assistant filed can be undone.
The whole vault exports as markdown at any time. Each exported file carries its created and updated dates and metadata in a frontmatter header; the export is the current version of each document.
Connecting the apps requires the Business plan: $199 a month for 3 people ($2,189 a year), plus $59 a month for each extra seat up to 25. It includes the shared vault and team chat, connecting Relay or TradesBackbone, ReputeLift and SynthPrism, Today, team roles, private folders, document history with restore, automations, and every add-on except Advanced AI models ($20 a month). Crew added from TradesBackbone or Relay are free and do not take a seat. Personal ($69 a month or $759 a year) and Personal Plus ($119 a month or $1,309 a year) are for one person and do not include team roles, private folders or team chat. Free is $0 for one person, with the vault, filing and search and 50 AI credits a month. Enterprise past 25 seats is custom.
Learn more about how Mycel works and see the full pricing page.
Setting Up a Shared Knowledge Base for Your Agency Team
Start With Core Folders and One Channel
Begin with a simple folder structure: Clients, Projects, Processes, Decisions, Brand. Resist the urge to build an elaborate taxonomy up front—folders can be added and reorganized as the team learns what it actually files.
Open one team chat channel, typically #general, and introduce the @assistant command. The team will ask questions as they come up, and the assistant will either answer from the vault or note that it found nothing filed yet. The gaps reveal what the team needs to document next.
File Decisions and Process Notes as They Happen
The goal is not to pause work and write a knowledge base. The goal is to file decisions and outcomes as a byproduct of making them.
When the team settles a question in chat—how to handle a client request type, which vendor to use, the standard turnaround for revisions—file a short decision note into the Decisions or Processes folder. When a project wraps, file a brief outcome or lessons-learned note into the Projects folder. These can be one paragraph each; the assistant reads full sentences, not detailed documentation.
Over weeks, the vault fills with the actual decisions and outcomes of the business, not aspirational documentation someone drafted and nobody updated.
Set Folder Permissions for Sensitive Material
Most folders can default to team access—everyone in the business sees Clients, Projects, Processes, Brand. Move financials, contracts, payroll and confidential client information into folders restricted to owners and admins.
Connect the Apps the Business Already Runs
If the business runs Relay (appointment businesses) or TradesBackbone (trade shops), ReputeLift and SynthPrism, connect them to Mycel on the Business plan. They can file what the business learns there into the vault, and their AI reads back what the owner allows: Relay's and TradesBackbone's AI read only the folders the owner shares with them, ReputeLift's AI reads the business vault, and SynthPrism's assistant reads it for brand voice and business facts. Customer records stay in Relay or TradesBackbone.
Review and Approve Assistant Suggestions
The assistant may offer to file a decision note or add a task from a team chat thread, and anything it would change inside another Martello app appears in Approvals.
Review them before approving. The assistant drafts based on what it read, but it can misinterpret, combine unrelated facts, or file something with a detail wrong. The approval step keeps the human in the loop without requiring manual note-taking.
Anything the assistant filed can be undone using document history and restore, so a mistake is fixable.
Export the Vault Periodically for Backup
The vault is the business memory. Export it as markdown monthly or quarterly and store the snapshot in a separate location—cloud backup, local drive, version control.
Each exported file includes frontmatter with created and updated dates, so the backup is not just a pile of text—it's a dated archive of the business's knowledge at that point in time. If the knowledge base service changes, the business migrates, or something breaks, the memory survives.
Frequently Asked Questions
What makes a shared AI knowledge base different from a file storage system with search?
A shared AI knowledge base is built so an assistant can read the documents and answer questions from them, citing which files it consulted, rather than returning a list of filenames for the person to open and read. The vault is a working set of documents the team edits and organizes, not just archived files. The assistant surfaces the answer, not just the location of the answer.
Can the AI assistant in a knowledge base make changes without approval?
In well-designed systems, the assistant may offer to file a note, add a task, or update a record, but the action waits in an approval queue until a person confirms it. Anything the assistant would change inside another connected app should wait for approval first. This keeps the assistant helpful without granting it unsupervised write access to business records.
How do role-based permissions work in a shared vault?
A business typically has roles like Owner, Admin, Member and Viewer. A folder can be set to everyone, to team members but not viewers, or to owners and admins only. Each folder's permission level is set independently, so the team can file client work openly and restrict financials or contracts to owners and admins.
Why use markdown for a knowledge base instead of a rich text editor or database?
Markdown is plain text with lightweight formatting syntax, so the files are human-readable in any text editor and export cleanly without conversion. A vault stored in markdown can be version-controlled, searched with standard tools, and migrated to another platform if needed. Proprietary formats lock the business memory inside one vendor's system and require that vendor's software to read or export.
What happens if the assistant answers incorrectly or cites the wrong document?
The assistant shows which documents it read to form the answer, so the person asking can open those files and verify. A reply that read nothing from the vault is marked that way, signaling the answer is not drawn from filed knowledge. If the assistant files something wrong, document history with restore lets you undo it. The assistant helps surface and organize information; the team remains responsible for checking it.
How much does Mycel cost for a team?
Mycel's Business plan is $199 a month for 3 people, plus $59 a month for each extra seat up to 25, and it adds team chat, roles, private folders, document history and connecting the other Martello apps. Personal is $69 a month and Personal Plus is $119 a month, each for one person, and Free is $0 for one person. Annual billing is 12 months for the price of 11.
Who is Mycel built for?
Mycel is built for trade shops of about 3 to 25 people (HVAC, plumbing, electrical, roofing, landscaping, cleaning, pest control) and appointment businesses (salons, barbers, lash and brow studios, med spas, tattoo studios, groomers). It is the brain of one connected stack with Relay or TradesBackbone, ReputeLift and SynthPrism, and the business sets it up and runs it itself.
A shared AI knowledge base for a small agency team should do one thing well: help the business remember what it learned, and surface that memory when the team needs it. The vault holds the decisions, processes, client details and project outcomes the team files as it works. The assistant answers questions from those documents and cites its sources. Team chat integration, role-based permissions, document history and markdown export keep the system practical for a small team without a librarian.
The result is a business that builds institutional memory instead of losing it, where the answer to a recurring question lives in the vault rather than scattered across inboxes and the owner's head. That memory compounds, and over time the knowledge base becomes the place the team checks first—not because they were told to, but because it works.