When you ask your AI assistant a question and your coworker asks the same AI the same question an hour later, you expect the same answer. When your AI tells a customer one thing and you tell them another because you didn't know what the AI said, coordination has failed. The best tool for coordinating your AI with other people is one that puts the AI's memory in a shared vault the whole team works in, shows which documents the AI read to answer each question, and lets everyone see what the AI knows and what it told someone else.
Most AI tools give each person their own conversation history but no shared memory. Your assistant learns from your questions, your coworker's assistant learns from theirs, and nobody knows what anyone else's AI said until something goes wrong. A coordination tool puts one business vault at the center, gives the team's AI assistant access to that vault, and marks every answer with the documents it read—or marks it clearly when the answer came from the AI's training rather than your business knowledge.
Key Takeaways
- The best tool for coordinating AI with other people stores the AI's knowledge in one shared vault the team reads and edits together, not in separate conversation histories.
- Every AI answer should show which vault documents it referenced, so the team knows whether the answer came from business knowledge or the AI's general training.
- Team chat with an assistant you can call into a thread by typing @assistant lets the team ask questions together, see the same answers, and approve what the AI wants to file back into the vault.
- Business workspaces add team roles so owners and admins can keep certain folders private while members and viewers see the rest, and every document tracks who changed what and when with restore.
- Connecting the AI to other business apps means those apps can file what the business learns into the vault, and the AI in those apps reads only the folders you share with them.
Why Coordinating AI Across a Team Is Different from Personal AI Use
When you use an AI assistant alone, you build up context across your conversation history. The AI remembers what you told it three messages ago, and you remember what it told you. You know which answers sounded confident and which ones you should double-check. Your mental model and the AI's context window stay loosely in sync.
Add a second person and that breaks. Your coworker doesn't know what you asked or what the AI answered. If they ask a related question, the AI starts fresh unless you shared a link to your chat, and even then they're reading a transcript, not working in shared memory. If the AI gave you an answer based on outdated information and you corrected it in your thread, that correction lives in your history, not in the AI's knowledge base for the next person.
The problem gets worse when the AI is answering customer questions or drafting messages on behalf of the business. One team member asks the AI to draft a response about your return policy, and the AI invents a 60-day window because that's common in its training data. Your actual policy is 30 days, but the AI doesn't know that unless someone wrote it down in a place the AI reads every time. Another team member asks the same question the next day and gets a different answer. The customer gets two conflicting messages, and nobody on the team knows what the AI said to whom until the customer forwards both emails back asking which one is correct.
Coordinating AI across a team means the AI reads from the same source of truth every time, no matter who asks. It means the team can see what the AI knows, correct it when it's wrong, and track what it told someone else. It means the AI's answers are reproducible: ask the same question with the same vault contents, get the same answer.
What Makes a Tool Good at Coordinating AI for Multiple People
A tool built to coordinate AI across a team needs a few things most personal AI apps don't provide.
A Shared Vault the AI Reads Every Time
The AI's memory has to live somewhere the whole team can read and edit. Not a shared conversation history—those get long, contradictory, and impossible to search. A structured vault of folders and documents, plain text or markdown, organized by topic the way the business thinks about it. Customer onboarding goes in one folder, product specs in another, pricing and policies in a third.
When anyone asks the AI a question, the AI searches that vault, reads the relevant documents, and answers from them. The answer is the same whether you ask or your coworker asks, because the AI is reading the same source material.
The vault needs to be editable by the team, not just the AI. When the return policy changes, someone opens the policy document, updates the text, and saves it. The next time anyone asks the AI about returns, the AI reads the updated document and gives the updated answer. No retraining, no prompt engineering, no telling the AI to forget what it said before—just edit the document.
Answers That Show Their Sources
The single most important feature for coordinating AI across people is that every answer shows which documents the AI read. Not a footnote or a citation style reference—an explicit list of filenames or document titles the AI opened to generate the answer, visible in the same interface as the answer itself.
This does two things. First, it lets the person reading the answer decide whether to trust it. If the AI says your return window is 30 days and cites the "Return Policy Updated March 2026" document, you know where it got that. If the AI answers a question about a product feature and the source list is empty—meaning the AI read nothing from the vault and answered from its general training—you know to double-check it.
Second, it helps the team maintain the vault. When the AI gives a wrong answer and cites a document, you open that document, see what it says, and fix it. When the AI gives a wrong answer and cites nothing, you know the vault is missing something it should contain, so you write it and file it where the AI will find it next time.
A reply that read nothing from the vault should be marked that way. The user needs to know the difference between "I read your business documents and found this" and "I'm answering from my general training, which may not match your business."
Team Chat with the Assistant Built In
If the AI lives in a separate app from the team's conversations, coordination fails in practice. Someone asks the team a question in chat, someone else answers, and a third person says "I think the AI had a different answer yesterday" but nobody wants to leave the chat app, open the AI app, re-type the question, copy the answer, and paste it back.
Team chat needs the assistant available inside it. When someone types @assistant in a message thread, the assistant reads the question, searches the vault, and posts its answer in the thread where everyone in that conversation can see it. The answer still shows which documents the assistant read, and everyone in the channel sees the same answer at the same time.
The assistant may offer to file a decision note or add a task based on the conversation—"It sounds like you decided to extend the return window to 45 days, would you like me to file that?"—but it does nothing until a person approves it. This keeps the assistant from changing the shared vault based on one person's question without the rest of the team knowing.
Mycel is built exactly this way. Business workspaces include team chat with channels (every business has #general) and direct messages. Type @assistant in any message and the assistant answers in the thread from the business vault, showing which vault documents it read. If the assistant offers to file something, it waits for someone to approve it in the Approvals nav, so the whole team knows what the AI added.
Permissions and Roles for Different Levels of Access
A coordinated AI needs everyone reading from the same vault, but not every person on the team should see everything in it. The business needs folders for employee compensation, acquisition discussions, or sensitive customer issues that only owners and admins should read.
Team roles let you set a folder to everyone, to team members but not viewers, or to owners and admins only. The AI reads only what the person asking the question has permission to see. When a member asks about pricing strategy and that folder is restricted to owners, the AI doesn't read it, doesn't answer from it, and doesn't leak it.
Every document should keep a history of who changed what and when, with restore. When the AI files something into the vault, that edit shows up in the document history with the assistant's name and a timestamp. If the assistant filed something wrong, anyone with edit access can open the history and restore the version from before. Something the assistant filed can be undone the same way any human edit can.
How Connecting AI to Other Business Apps Changes Coordination
Most teams don't do all their work in a vault and a chat app. Customer records live in a customer app, service requests come through a different system, reputation monitoring happens in another tool. If the AI only knows what's in the vault, it misses half the context.
Connecting the business apps the team already uses means those apps can file what the business learns into the vault automatically. A customer calls with a question your team hasn't seen before, someone researches it and answers, and the app files a note summarizing the question and answer into a vault folder the AI reads. The next time anyone asks—in the vault, in team chat, or inside the connected app—the AI has that context.
The AI running inside those connected apps should read from the same business vault, but only the folders the owner shares with that app. A customer communication app doesn't need access to the HR folder, and a reputation monitoring app doesn't need the product development roadmap. Sharing specific folders with each app keeps the AI coordinated—it reads the business's voice, policies, and relevant knowledge—without giving every app access to everything.
When Mycel connects to the Martello apps a business uses—Relay for customer communication, TradesBackbone for trades and service business operations, ReputeLift for reputation monitoring, and SynthPrism for content and brand—those apps file what the business learns into the Mycel vault. Relay's and TradesBackbone's AI read only the folders the owner shares with them. ReputeLift's AI reads the business vault for context about the business when interpreting reviews and mentions. SynthPrism's assistant reads the vault for brand voice and business facts when drafting content. Customer records themselves stay in Relay and TradesBackbone, one click away, not duplicated into the vault.
Connecting the apps requires a business workspace. The Personal and Personal Plus plans are for one person working in their own vault with their own assistant; team roles, private folders, and team chat belong to the Business plan.
The Home View: What Is Waiting Across All the Connected Tools
When AI and business apps are coordinated through a shared vault, a second problem appears: the team now has things waiting for them in multiple places. An approval waiting in the vault, a customer reply waiting in the communication app, a review that needs a response in the reputation tool, a content draft waiting for edits in the brand assistant.
The best coordination tools surface everything waiting across the connected apps in one list. Not a dashboard with widgets—a plain list of items that need a person's attention, with which app it came from, why it surfaced, how long it has waited, and a link straight to it in the app where the work happens.
Mycel's Home view opens on Today, a list of what is waiting on you across the connected Martello apps. Each item shows the app it came from, the reason it surfaced, how long it has been waiting, and a direct link to it. You see the approvals waiting in Mycel, the customer conversations waiting in Relay, the service requests waiting in TradesBackbone, and the review responses waiting in ReputeLift, all in one place. Click an item and you're in the app where you handle it.
Step-by-Step: Setting Up a Coordinated AI for a Small Team
Here's how a small business would typically set up AI coordination for a team of three to ten people.
1. Start with the vault structure. Create folders for the topics the business gets questions about most often: policies (returns, shipping, service guarantees), product or service details, pricing, brand voice and messaging, customer onboarding steps, common troubleshooting. Use plain language folder and file names—the AI searches them.
2. Write the foundational documents. A one-page return policy, a pricing sheet, a service area map, a brand voice guide with example phrases. Keep each document short and specific. The AI reads faster and answers better from "Return Policy" and "Shipping Policy" as separate documents than from one long "Customer Policies" file that covers six topics.
3. Invite the team and set roles. Add the people who will ask the AI questions and edit the vault. Assign roles: Owner and Admin for people who should see everything and approve vault changes the AI proposes, Member for people who should see most folders but not the restricted ones, Viewer if someone needs read-only access. Set permissions on the folders that should stay private.
4. Connect the business apps the team uses. If customer communication happens in Relay, service scheduling in TradesBackbone, or reputation monitoring in ReputeLift, connect those apps so they can file learnings into the vault and their AI can read the folders you share with them. Each app will ask which vault folders it can read—share the relevant ones, not everything.
5. Teach the team to check the sources. Every time the assistant answers a question, look at which documents it read. If the answer is right and cited the correct document, you're done. If the answer is right but cited nothing—meaning it came from the AI's training—decide whether that topic belongs in the vault. If the answer is wrong, open the document it cited, see what it says, and fix it.
6. Use @assistant in team chat. When someone asks a question in #general or a direct message and the answer should come from the vault, type @assistant and let it answer in the thread. Everyone sees the answer and the sources at the same time, and if the answer is wrong, someone can correct the vault document right then.
7. Review what the assistant wants to file. When the assistant offers to file a decision note or add something to the vault after a conversation, check the Approvals nav before approving it. Read what it wants to write, make sure it's accurate, and approve it if it's helpful. If it's wrong, reject it and file the correct version yourself.
8. Export the vault periodically. 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, not its version history, so it's a snapshot of what the AI knows right now. Keep a quarterly export as a backup and a record of how the business knowledge evolved.
Comparing Tools for AI Coordination Across Teams
| Tool type | Shared memory | Shows sources | Team chat | Built for | |---------------|-------------------|-------------------|---------------|---------------| | Personal AI assistants | Each user has their own conversation history | Rarely; some add footnotes | No | Individual productivity | | Shared wikis with AI add-ons | Yes, one wiki | Sometimes, depends on the add-on | Typically separate app | Documentation, not real-time questions | | AI built into team chat apps | Conversation history in each channel | Rarely | Yes, it's the platform | General conversation, not business memory | | Business AI with shared vault | Yes, one vault the whole team edits | Every answer shows which documents were read | Yes, with assistant callable in threads | Coordinating business knowledge and AI answers |
The tools most teams already use—personal AI assistants, shared wikis, chat apps with AI features—weren't built to coordinate AI across people. Personal assistants keep separate memory per user. Wikis organize knowledge but the AI feature is often an afterthought, and answers don't reliably cite the pages they read. Chat apps with AI let everyone see the bot's reply in the channel, but the bot typically doesn't read a shared business vault; it answers from its training or from the conversation history in that one thread.
A tool purpose-built for coordinating business AI puts the shared vault first, makes the assistant read from it every time, shows the sources in every answer, and embeds the assistant in team chat so asking a question and seeing the answer happens where the team already works.
What to Look for in Pricing and Plans
AI coordination tools typically offer a free or low-cost plan for one person to try the vault and assistant, a personal plan with higher AI usage limits, and a business plan that adds the team features.
Check whether team roles, private folders, and team chat require the business plan—they usually do, because those are the features that turn a personal vault into a coordinated team tool. If you need multiple people working in the vault with an assistant they can all call, plan for the business tier.
Look at how AI usage is metered. Most tools count questions or credits per month, with a published limit. Make sure the limit fits the team's typical usage—if four people each ask the assistant ten questions a day, you're looking at roughly 800-1,200 questions a month depending on working days.
For Mycel, pricing is published and straightforward, and the pricing page always has the current numbers. Free is $0 for one person, with vault, filing and search, and 50 AI credits a month. Personal is for one person, adding your own assistant and vault, with Email in and Omi included. Personal Plus is for one person, with everything in Personal plus every available add-on and more AI credits. Business is for a team: it connects TradesBackbone, Relay, ReputeLift and SynthPrism, and adds Today, team roles, private folders, document history with restore, automations, and every add-on except Frontier models. Enterprise is custom. Team roles, private folders, and team chat belong to a business workspace, so if you need those, the Business plan is where they appear.
Common Mistakes When Rolling Out AI to a Team
Not writing anything down before turning on the AI. Teams sometimes connect an AI assistant to an empty vault and expect it to answer business questions. The AI answers from its training, gives generic replies that don't match the business, and the team decides AI doesn't work. Write the foundational documents first—policies, processes, product details—then turn on the assistant.
Forgetting to check the sources. Someone asks the AI a question, reads the answer, copies it into an email to a customer, and sends it without looking at which documents the AI cited. The answer was wrong, the AI cited nothing because the vault didn't contain that information, and the customer got bad information. Always check which documents the AI read, and if it read nothing, verify the answer before using it.
Letting the AI file things without approval. The assistant offers to file a note after a conversation, someone clicks approve without reading it, and the assistant writes something incorrect into the vault. The next person asks a related question, the AI cites the wrong note the previous person approved, and now the wrong answer is spreading. Approvals exist because the AI should not change the shared vault without a human checking its work.
Not setting folder permissions. The team puts everything in the vault with no restricted folders, then realizes the assistant is reading employee salary information to answer a question from a viewer, or a member sees acquisition discussions they shouldn't. Set permissions on sensitive folders from the start, and review them when you add a new person.
Duplicating customer data into the vault. Some teams try to copy customer records, email threads, or service tickets from the app where they live into the vault so the AI can read them. This creates two places to update, two places to search, and often violates the business's own data policy. Customer records should stay in the customer app; the AI in that app should read the vault for policies and knowledge, but the customer data stays in its system of record, one click away.
Using the Vault for Things Other Than AI Answers
A shared vault the AI reads becomes the center of the business's memory, and teams start using it for more than just answering questions.
Onboarding new team members. The new person reads the vault to learn policies, processes, and the way the business talks to customers. They ask the assistant questions as they go, and the assistant answers from the same documents everyone else uses, so they're learning the current version, not something that went out of date six months ago.
Recording decisions. After a team discussion in chat, someone asks the assistant to file a decision note, approves it, and the decision lives in the vault where the whole team can find it later. When someone asks "why did we decide to do it this way?" the assistant cites the decision note.
Drafting templates. The team keeps email templates, proposal language, and service agreement clauses in the vault. When someone needs to send a proposal, they ask the assistant to draft one, and it pulls language from the templates in the vault, adjusted for the context of the question.
Tracking what changed and when. Every document keeps a history of who changed what and when. When a policy changes, the team can see the old version, who updated it, and when. If a customer says "you told me X last month and now you're saying Y," the team can open the document history, see whether it changed, and explain what happened.
The vault exports as markdown at any time, so the business always has a copy of its own knowledge outside the tool. Each exported file carries its created and updated dates and metadata in a frontmatter header. The export is the current version of each document, not the version history, so it's a clean snapshot.
Frequently Asked Questions
What does it mean when the AI answer shows no sources?
When the AI answer shows no sources or is marked as reading nothing from the vault, it means the AI did not find relevant documents in the business vault and answered from its general training instead. The answer may be correct in a general sense, but it is not based on your business's specific policies, products, or knowledge. Treat it as a suggestion to verify, not as the business's answer. If the topic comes up often, write a document covering it and file it in the vault so the AI reads it next time.
Can team members see each other's conversation history with the assistant?
Conversation history is typically private to the person who had the conversation, but the vault itself is shared. If someone asks the assistant a question in their own Ask view, their coworkers don't see that thread. If someone asks the assistant a question in a team chat channel by typing @assistant, everyone in that channel sees the question, the answer, and the sources. The coordination happens through the shared vault, not through reading each other's private threads.
What happens if two people edit the same vault document at the same time?
Most coordination tools handle simultaneous edits the same way collaborative document editors do: the second person to save sees that the document changed and decides whether to overwrite, merge, or discard their edits. Because every document keeps a history of who changed what and when, you can restore the version from before the conflict if something goes wrong. In practice, vault documents in a small business don't change frequently enough for edit conflicts to happen often—policies and process documents are more stable than customer records or conversation threads.
Do I need a business workspace to use the AI assistant?
No, the AI assistant works on the Free, Personal, and Personal Plus plans for one person working in their own vault. You need a business workspace—the Business plan—when you want team roles, private folders, team chat with the assistant callable in threads, and the ability to connect the Martello apps (TradesBackbone, Relay, ReputeLift, SynthPrism) so they file into the vault and read from it. If you're the only person using the vault and you don't need those features, a personal plan is enough.
How do I stop the AI from giving wrong answers?
The AI gives wrong answers for two reasons: it read a vault document that contained incorrect information, or it read nothing from the vault and answered from its general training. Fix the first by editing the document the AI cited and correcting it. Fix the second by writing a document that covers the topic and filing it in the vault where the AI will find it next time. Check the sources on every answer—if the AI cited a document, open it and make sure it says what the AI claims; if it cited nothing, decide whether the answer is trustworthy or whether you need to write that knowledge into the vault.
Can the AI change the vault on its own?
No. The AI can offer to file a decision note, add a task, or write something into the vault, but it does nothing until a person approves it. When the assistant offers to file something, the request appears in the Approvals nav, someone reviews what the AI wants to write, and they approve or reject it. If approved, the edit shows up in the document history with the assistant's name and a timestamp, and it can be undone with restore the same way any human edit can. This is true for the assistant in Mycel's vault and for anything the assistant in a connected Martello app would file into the vault—all of it waits for approval first.
Making AI Coordination Work in Practice
The best tool for coordinating your AI with other people is the one that makes the AI read from the same source of truth every time, shows which documents it read to answer each question, and lets the team see, edit, and approve what the AI knows. It has to embed the assistant in team chat so asking a question and seeing the answer happens where the work happens, and it has to connect to the other business apps the team uses so the AI has context from across the business, not just the vault.
Coordination is less about the AI's capabilities and more about shared memory, visible sources, and team access. A powerful AI that reads from separate memory per user will give inconsistent answers. A simple AI that reads from one shared vault every time will give the same answer to everyone who asks, and when that answer is wrong, the whole team can see the document that caused it and fix it together.
Start with the vault, write the knowledge the business uses most often, invite the team, and teach everyone to check the sources before trusting the answer. The AI becomes useful when it reads what the business knows, not when it tries to guess.