Small business owners face a new coordination problem: their team includes AI assistants now, and those assistants need to work from the same information people do, without inventing answers or making changes no one approved. Coordination software for people and their AI assistants solves this by providing a shared memory both can read, transparent attribution for every AI answer showing which documents it consulted, and approval gates that prevent an assistant from filing decisions or changing records until a person confirms it. The software typically includes team chat where typing @assistant surfaces an answer from the business vault in the thread, role-based folder privacy so sensitive information stays with owners and admins, and a unified inbox that shows what is waiting across connected apps.
The core challenge is not whether an AI assistant is useful—it often is—but whether the business can trust what it says and control what it does. A coordination layer addresses both by grounding the assistant in the company's own written record and requiring human approval before the assistant changes anything outside that record.
Key Takeaways
- Coordination software for people and AI assistants provides a shared vault both teams and AI read from, with every AI answer citing which vault documents it consulted or marking that it read nothing.
- An approval workflow prevents AI assistants from filing notes, adding tasks, or changing records in connected apps until a person explicitly confirms each action.
- Team chat with an @assistant mention brings AI answers into the conversation thread, grounded in the business vault and visible to everyone in the channel.
- Role-based folder access keeps sensitive information—financial records, HR notes, strategy documents—visible only to owners and admins while the rest of the vault remains shared.
- Document history with restore and full markdown export ensure the business retains control and portability of its institutional memory.
What Coordination Software for AI Assistants Actually Does
Coordination software creates a single source of truth that both people and AI assistants work from. This typically takes the form of a shared vault organized as plain-text folders and markdown documents, with an AI assistant that answers questions by reading from those folders. Each answer displays which vault documents the assistant consulted, and if the assistant replies without reading anything from the vault, that reply is marked accordingly. This attribution discipline addresses the most common problem with general-purpose AI tools: the business cannot tell whether an answer came from its own records or from the assistant's training data.
The software also manages what the AI assistant can change. When an assistant offers to file a decision note, add a task, or update a record in a connected app, that action waits for a person to approve it before taking effect. This approval gate ensures that AI assistance remains advisory rather than autonomous. The business benefits from speed and suggestions without ceding control.
Coordination software typically includes team messaging with AI integration. Typing @assistant in a team channel or direct message prompts an answer from the business vault, delivered as a reply in the thread. The assistant may offer to file what was decided or add a follow-up task, which again does nothing until someone approves it. This keeps AI assistance available in the flow of conversation without requiring people to switch contexts or copy answers between tools.
Why Businesses Need a Coordination Layer Between People and AI
General-purpose AI assistants answer from their training data, which does not include a business's pricing, processes, past decisions, or customer context. When a team member asks such an assistant a question about the business, the answer is often plausible-sounding but wrong. Worse, the team member may not realize the answer was invented rather than retrieved.
A coordination layer grounds the AI assistant in the business's own institutional memory. The assistant reads the vault the team maintains, and every answer shows its sources. If the vault does not contain the information needed to answer a question, the assistant says so rather than guessing. This makes the assistant useful for onboarding new team members, refreshing someone's memory before a client call, or checking what was decided last quarter.
The coordination layer also prevents AI assistants from acting without oversight. If a business connects apps for customer communication, project tracking, or reputation management, those apps can file what they learn into the vault—but anything an AI assistant would change inside an app waits for approval first. The assistant might suggest updating a customer record or adding a task based on a conversation, but the suggestion surfaces in an Approvals queue for a person to confirm or reject. This approval workflow keeps the assistant helpful without making the business chase down changes it did not authorize.
How Shared Vaults and Transparent Attribution Work
A shared vault is a collection of folders and markdown documents the entire team can access. Typical folder structures include Clients, Projects, Processes, Pricing, Brand Voice, and Decisions. Each folder can be kept to everyone, restricted to team members but not viewers, or limited to owners and admins only. This role-based access ensures sensitive information stays private while the bulk of the vault remains shared.
Every document in the vault carries a history of who changed what and when, with the ability to restore any previous version. If an AI assistant files a note automatically, that filing appears in the history and can be undone. This version control gives the business confidence that nothing disappears and no change is irreversible.
When the AI assistant answers a question, it displays the vault documents it read to construct the reply. If someone asks "What is our cancellation policy?" the assistant answers and shows it read /policies/cancellation.md and /pricing/terms.md. If the assistant replies without consulting any vault document, that reply is explicitly marked. This transparency lets the team judge whether an answer is grounded in company knowledge or should be verified.
The vault exports as markdown at any time. Each exported file carries its created and updated dates and metadata in a frontmatter header. The export includes the current version of each document, not its version history, making it straightforward to migrate the business's knowledge base to another system if needed.
What Does AI Assistant Approval Workflow Look Like?
An approval workflow intercepts any action an AI assistant proposes that would change something outside the vault. These actions typically include filing a decision note after a team discussion, adding a task to a project tracker, or updating a customer record based on a conversation.
When the assistant offers to take such an action, it does nothing until a person explicitly approves it. The proposed action appears in an Approvals queue, showing what the assistant wants to change, why, and which vault context or conversation prompted the suggestion. The person can approve the action, edit it before approving, or reject it outright.
For example, during a team chat in the #general channel, someone might type @assistant and ask about lead qualification criteria. The assistant answers from the vault, and at the end of the discussion it offers to file a decision note summarizing what the team agreed. That offer appears in the thread, and if someone clicks approve, the assistant files the note into the vault. If no one approves it, nothing happens. The conversation remains in the chat log, but the vault is unchanged.
This workflow extends to connected apps. If a business links customer communication software or project management tools to the coordination platform, those apps can file notes into the vault automatically, but any change the AI assistant would make inside the app itself waits for approval. The assistant might suggest updating a contact's status or adding a follow-up task based on an email thread, but those changes surface for review before taking effect.
Team Chat and the @assistant Pattern
Coordination software for AI assistants typically includes team messaging with channels and direct messages between people in the business. Every business workspace includes a #general channel by default, and team members can create additional channels for projects, departments, or topics.
Typing @assistant in any message prompts an answer from the business vault, delivered as a threaded reply visible to everyone in the channel. The assistant shows which vault documents it read, maintaining the same attribution standard as standalone queries. This integration keeps AI assistance available during team discussions without requiring people to leave the conversation, open a separate tool, ask a question, and paste the answer back.
If the discussion resolves a question or reaches a decision, the assistant may offer to file a decision note or add a task. That offer appears in the thread, and any team member with sufficient permissions can approve or reject it. This keeps institutional memory up to date without interrupting the flow of conversation.
The @assistant pattern also works in direct messages. If two people are discussing a client issue and need to check what was decided last time, typing @assistant in the DM retrieves the relevant vault context. The answer is visible only to the participants in that direct message, but it draws from the same shared vault the rest of the team uses.
Role-Based Access and Folder Privacy
A business workspace typically includes four roles: Owner, Admin, Member, and Viewer. Owners and admins have full access to the vault, including private folders. Members can read and edit shared folders but cannot access folders marked owners-and-admins-only. Viewers can read shared folders but cannot edit them.
Each folder can be configured with one of three privacy levels: open to everyone in the workspace, restricted to team members excluding viewers, or limited to owners and admins. This granular control lets a business share operating procedures, brand guidelines, and project notes widely while keeping financial records, personnel files, and strategic plans private.
When an AI assistant answers a question, it reads only the folders the person asking has permission to see. If a Member asks a question that would require consulting an owners-only folder, the assistant replies based on the folders the Member can access and notes that some vault areas were not consulted. This prevents the AI from leaking restricted information through its answers.
Document history applies across all privacy levels. Even in a private folder, every change is logged with who made it and when, and any previous version can be restored. This audit trail is particularly important for sensitive information, where knowing who changed a policy or pricing document and when can matter for compliance or dispute resolution.
Connecting Customer Records and Filing What the Business Learns
Coordination software for AI assistants often integrates with customer communication and project management apps. In Mycel's case, customer records themselves stay in Relay and TradesBackbone, one click away from the vault but not duplicated inside it. This separation keeps the vault focused on the business's institutional knowledge—processes, decisions, brand voice, pricing—while customer-specific interaction history remains in the tools built for that purpose.
Connected apps can file what the business learns into the vault automatically. For example, a conversation with a customer might surface a new objection or a request the business has not seen before. The app handling that conversation can file a note about the pattern into the vault, making it available for the AI assistant to reference when someone asks about common customer questions. The customer's name and contact details stay in the customer record; the vault receives the anonymized insight.
When AI assistants in connected apps read the vault, they read only the folders the business owner shares with them. Relay's and TradesBackbone's AI might read a folder containing response templates and qualification criteria but not the folder with financial projections. ReputeLift's AI reads the full business vault to maintain consistent brand voice across public responses. SynthPrism's assistant reads the vault for brand voice and business facts when generating marketing content. This selective sharing gives the business control over which AI has access to which context.
Unified Inbox: Today Across Connected Apps
One practical coordination challenge for small business owners is knowing what needs attention across multiple tools. Email, project tracker, customer communication app, and reputation monitoring each have their own notifications, and it is easy for something to sit unaddressed simply because the owner did not check that particular app.
Coordination software addresses this with a unified inbox—in Mycel's case, called Today. Today lists everything waiting on a person across connected apps, showing which app it came from, why it surfaced, how long it has waited, and a link straight to it. This aggregated view lets the owner see at a glance that a customer reply has been waiting three hours, a project approval has been pending two days, and a review response draft needs final approval, all without opening each app individually.
Today does not replace the connected apps or duplicate their functionality. It is a coordination layer that surfaces what needs attention and links directly to the source. The owner clicks through to handle the item in its native app, but the decision about what to tackle next happens in one place.
What to Look For in Coordination Software
When evaluating coordination software for people and AI assistants, consider these factors:
Transparent attribution for AI answers. Every answer from the AI assistant should show which documents it consulted, and replies that drew from no source documents should be marked accordingly. This transparency is the difference between an assistant that helps and one that invents.
Approval workflow for AI actions. The assistant should not file notes, add tasks, or change records in connected apps without a person confirming the action first. Look for software where proposed changes surface in an approval queue before taking effect.
Shared vault with role-based access. The business needs one place where institutional knowledge lives, with granular control over who can see and edit which folders. Owners and admins should be able to keep sensitive information private while sharing operating knowledge widely.
Team messaging with AI integration. Chat channels and direct messages should allow typing @assistant to get an answer from the vault directly in the conversation thread, maintaining the same attribution and approval standards as standalone queries.
Document history and export. Every document should log who changed what and when, with the ability to restore previous versions. The entire vault should export as plain markdown, with metadata intact, so the business is never locked in.
Unified view of what is waiting. If the software connects multiple apps, it should provide a single list showing what needs attention across all of them, with links straight to each item.
How Coordination Differs From General-Purpose AI Tools
General-purpose AI assistants are trained on broad datasets and answer from that training. They are useful for drafting text, summarizing articles, or explaining concepts, but they do not know anything specific about a particular business. When a team member asks a general-purpose assistant a business-specific question, the assistant often produces a plausible-sounding answer that is factually wrong.
Coordination software grounds the AI assistant in the business's own documents. The assistant reads the vault the team maintains and cites its sources. If the vault does not contain the information needed to answer a question, the assistant says so. This grounding makes the assistant useful for internal questions—"What is our cancellation policy?" "How do we qualify leads?" "What did we decide about pricing last quarter?"—where a general-purpose tool would guess.
Coordination software also manages what the AI can change. General-purpose assistants produce text the user then copies into other tools. Coordination software lets the assistant interact with connected apps but gates every proposed action behind an approval. The assistant suggests, a person confirms, and the system logs who approved what.
Coordination Software Pricing Models
Coordination software typically offers tiered pricing based on the number of people using it and the features included. A basic tier might provide a personal vault and AI assistant for one person, while a business tier adds team roles, private folders, team chat, and the ability to connect external apps.
Mycel, for example, offers a Free plan at $0 for one person, including vault, filing, and search with 50 AI credits a month and hard monthly ceilings. Personal is $50/month or $500/year for one person, adding an assistant, 1,500 AI credits a month, and Email in and Omi. Personal Plus is $89/month or $890/year for one person, including everything in Personal plus every available add-on and more AI credits. The Business plan is $119 per seat per month for up to 20 seats, or $1,190 per seat per year, and includes everything in Personal per seat plus connecting TradesBackbone, Relay, ReputeLift, and SynthPrism, Today, team roles, private folders, document history with restore, automations, and every add-on. Enterprise pricing past 20 seats is custom. Team roles, private folders, and Team chat require a business workspace; Personal and Personal Plus plans do not include those features. For current details, see pricing.
When evaluating pricing, consider how many people need access, whether the business requires team roles and private folders, and whether connecting external apps matters. A solo business owner may find a personal plan sufficient, while a team of five with customer communication software and project tracking tools will need a business plan.
A Worked Example: Coordinating a Client Inquiry
Consider a hypothetical scenario. A landscaping company with seven employees uses coordination software to manage its institutional memory. The vault includes folders for Services, Pricing, Processes, and Decisions. Customer records live in Relay, one click away, but the vault holds the company's service descriptions, pricing structure, and notes on past decisions about scope.
A potential client emails asking whether the company offers seasonal contracts and what the cancellation terms are. The email arrives in Relay, which files a note into the vault summarizing the inquiry. Today surfaces the waiting email for the owner, who clicks through to Relay to read the full message.
Before replying, the owner opens team chat and types "@assistant Do we offer seasonal contracts and what are the cancellation terms?" The assistant reads /services/contracts.md and /pricing/terms.md from the vault and replies in the thread: "Yes, seasonal contracts are available for spring-through-fall or year-round. Cancellation requires 30 days' notice with no penalty if given before the start of the next billing cycle. After the billing cycle starts, the client is responsible for that cycle's fee." The answer cites both source documents.
The owner uses the vault information to draft a reply in Relay. After sending it, Relay's AI offers to update the Services folder with a note that seasonal contracts are a common inquiry. That suggestion appears in the owner's Approvals queue. The owner reviews it, edits the wording slightly, and approves. The note is filed into /services/common-questions.md, timestamped and attributed to the owner who approved it.
Two weeks later, a new team member handling inquiries encounters a similar question. She types "@assistant What are our seasonal contract terms?" in the #general channel. The assistant reads the same vault documents and provides the answer, now updated with the note about common inquiries. The new team member replies to the client accurately without needing to ask the owner or search through past emails.
What Approval Gates Prevent
Approval gates prevent AI assistants from taking actions that would be difficult to undo or that require business judgment. Typical actions that wait for approval include:
Filing decision notes into the vault. After a team discussion, the assistant might offer to summarize what was decided and file it into the Decisions folder. The summary is useful, but a person should confirm it accurately reflects the conversation before it becomes part of the institutional record.
Adding tasks or updating records in connected apps. If a customer conversation reveals a follow-up action, the assistant might suggest adding a task to the project tracker or updating the customer's status. The suggestion is often correct, but a person should verify it fits the broader context before the change takes effect.
Updating vault documents with new information. As the business learns, the assistant might offer to update a process document or pricing sheet. This is convenient, but changes to shared documents should be reviewed to ensure accuracy and consistency with other parts of the vault.
The approval workflow surfaces these proposed actions in a queue showing what the assistant wants to change and why. The person can approve, edit before approving, or reject. Approved actions proceed immediately, and the vault or connected app reflects the change. Rejected actions are discarded. Every approval is logged with who made it and when.
Approval gates add a small amount of friction—one click to confirm—but they prevent the much larger friction of discovering later that an AI assistant filed incorrect information, added tasks no one wanted, or changed records based on incomplete context.
How Document History and Restore Work
Every document in a coordination software vault typically keeps a history of changes. Each edit is timestamped and attributed to the person or AI assistant that made it. This history serves several purposes: it provides an audit trail for sensitive documents, it lets the business see how a policy or process evolved, and it makes it possible to undo mistakes.
If an AI assistant files a note automatically and the note turns out to be inaccurate, the team can open the document history and restore the previous version. If someone edits a pricing document and later realizes the old version was correct, they can restore it. The restore function typically allows choosing any previous version by date and author, not just the most recent.
Document history does not export with the vault. When the business exports its knowledge base as markdown, each file includes its current created and updated dates and metadata in a frontmatter header, but the version history remains in the coordination software. This keeps exports clean and portable while retaining the audit trail in the live system.
The history also shows when the AI assistant filed something versus when a person edited it. This distinction helps the team decide how much to trust a particular document. A pricing sheet last edited by the owner three days ago is authoritative. A decision note filed by the assistant after a team chat and never reviewed by a person might need a second look.
Plain Markdown and Export Portability
Coordination software for AI assistants often stores documents as plain markdown. Markdown is a lightweight, human-readable format that works in any text editor and converts easily to other formats. Storing the vault as markdown ensures the business is not locked into proprietary file formats or dependent on a specific vendor's export tools.
The vault typically exports as a zip archive containing one markdown file per document, organized into folders that mirror the vault structure. Each file includes a frontmatter header with metadata: title, created date, updated date, author, and any tags or custom fields. This structured export makes it straightforward to import the knowledge base into another system, archive it for compliance, or process it with scripts.
Plain markdown also makes the vault easy to search and version-control outside the coordination software. A business could clone its vault into a Git repository for additional backup and versioning, or pipe the markdown files through a static site generator to create an internal wiki. The coordination software provides the AI assistant, approval workflow, and team collaboration features, but the underlying data remains accessible and portable.
Common Mistakes When Implementing Coordination Software
Businesses adopting coordination software for AI assistants often make a few predictable mistakes:
Treating the assistant as infallible. The AI assistant is useful, but it reads from the vault the team maintains. If the vault is incomplete or out of date, the assistant's answers will be incomplete or out of date. Teams should expect to spend time filling the vault with current information before the assistant becomes reliably helpful.
Skipping folder privacy setup. By default, many systems make all folders visible to everyone. Businesses that handle sensitive information—financial records, HR notes, strategic plans—need to configure folder privacy so those documents remain restricted to owners and admins. Failing to do this can leak information through AI answers or direct vault access.
Approving assistant suggestions without review. The approval workflow exists because the assistant's suggestions are not always correct. Teams that approve every suggestion reflexively to "save time" end up with inaccurate vault notes or unintended changes in connected apps. The review should be quick, not skipped.
Letting the vault grow unorganized. A vault with hundreds of documents and no folder structure becomes difficult for both people and AI to navigate. Businesses should establish a folder structure early—Clients, Projects, Processes, Decisions, Brand—and file documents consistently. The AI assistant can help by offering to file notes into the appropriate folder, but the structure itself requires human judgment.
Not exporting regularly. Plain markdown export makes it easy to back up the vault outside the coordination software. Businesses that skip regular exports risk losing their institutional memory if something goes wrong with their account or the software itself. A monthly export to local storage or cloud backup takes minutes and provides peace of mind.
How Mycel Coordinates People and AI Assistants
Mycel provides the memory a small business runs on and the owner's desk: one shared vault of plain-markdown folders the team works in, with an assistant that answers from it. 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, but customer records themselves stay in Relay and TradesBackbone, one click away.
Anything Mycel's assistant would change inside another Martello app waits for approval first. The assistant might offer to file a decision note or add a task based on a conversation, but that offer does nothing until a person approves it. Team chat includes channels and direct messages, and typing @assistant in a message gets an answer in the thread from the business vault. The assistant may offer to file what was decided or add a task, which again waits for approval.
A business workspace includes 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. The whole vault exports as markdown at any time, with each file carrying its created and updated dates and metadata in a frontmatter header.
Home opens on Today, a list of what is waiting on a 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. For details about how Mycel works or to see pricing, visit https://usemycel.com.
When to Choose Coordination Software Over General-Purpose AI
General-purpose AI tools are useful for drafting, summarizing, and explaining, but they do not coordinate work between people and AI assistants in a business context. Choose coordination software when:
The business has institutional knowledge that changes. Processes, policies, pricing, and past decisions change over time. Coordination software lets the team maintain that knowledge in a shared vault, and the AI assistant answers from the current version rather than guessing or citing outdated training data.
Multiple people need access to the same information. A solo business owner can keep everything in their head or in personal notes. A team needs a shared source of truth, and coordination software provides that with role-based access so sensitive information stays private.
AI assistants interact with business systems. If the business uses apps for customer communication, project tracking, or reputation management, coordination software lets those apps file insights into the vault and ensures any change the AI wants to make waits for approval.
Accountability for AI actions matters. Every AI answer in coordination software cites its sources, and every change the assistant makes is logged with who approved it and when. This audit trail matters for regulated industries, sensitive client work, or any situation where the business needs to trace how a decision was made.
The business wants to avoid lock-in. Plain markdown export and version history give the business control over its institutional memory. If the coordination software stops meeting the business's needs, the vault exports cleanly and can be imported elsewhere.
General-purpose AI tools remain useful for tasks that do not require business-specific knowledge or oversight. Coordination software addresses the narrower but critical problem of ensuring AI assistants work from the same information the team does and cannot act without approval.
Frequently Asked Questions
What is coordination software for people and their AI assistants?
Coordination software provides a shared memory that both team members and AI assistants work from, with transparent attribution showing which documents the AI consulted for each answer. It includes approval gates that prevent the AI from filing notes, adding tasks, or changing records until a person confirms the action, and team chat where typing @assistant brings AI answers into the conversation thread. The software typically integrates with business apps to surface what needs attention across all of them in one unified inbox.
How do approval gates for AI assistants work?
When an AI assistant offers to take an action that would change something outside the shared vault—filing a decision note, adding a task, updating a customer record—that action does nothing until a person explicitly approves it. The proposed action appears in an approval queue showing what the assistant wants to change and why. The person can approve it as-is, edit it before approving, or reject it outright. Every approval is logged with who made it and when, creating an audit trail for AI actions.
Can AI assistants in coordination software access sensitive business information?
Coordination software typically includes role-based folder access, where each folder can be kept open to everyone, restricted to team members excluding viewers, or limited to owners and admins only. The AI assistant reads only the folders the person asking has permission to see, preventing it from leaking restricted information through its answers. Connected apps read only the folders the business owner shares with them, giving granular control over which AI has access to which context.
What happens if an AI assistant files incorrect information into the vault?
Every document in the vault keeps a history of who changed what and when, with the ability to restore any previous version. If an AI assistant files a note that turns out to be inaccurate, a team member can open the document history and restore the version before the assistant's edit. The filing can also be undone entirely. This version control ensures that no change is irreversible and provides an audit trail for how documents evolved.
How does team chat with @assistant differ from using a separate AI tool?
Typing @assistant in a team chat channel or direct message brings an AI answer directly into the conversation thread, grounded in the business vault and visible to everyone in that channel. The assistant shows which vault documents it consulted, maintaining the same attribution standard as standalone queries. If the discussion resolves a question or reaches a decision, the assistant may offer to file a note or add a task, which appears in the thread for anyone to approve. This keeps AI assistance available during team conversations without requiring people to leave the chat, open another tool, copy the answer, and paste it back.
What should a business export from coordination software for backup?
The business should export its entire vault as markdown at regular intervals—monthly is typical—to ensure it has a backup outside the coordination software. Each exported markdown file carries its created and updated dates and metadata in a frontmatter header, making it straightforward to import into another system if needed. The export includes the current version of each document but not its version history, so the backup captures the institutional knowledge base as it exists at that moment. Storing these exports in local or cloud backup ensures the business retains its memory even if something goes wrong with the coordination software account.
Small businesses today employ both people and AI assistants, and the two need to work from the same information. Coordination software makes that possible by grounding AI answers in the team's shared vault, requiring approval before the AI changes anything outside that vault, and surfacing what needs attention across all connected apps in one place. The result is institutional memory that stays current, AI assistance that shows its work, and a business that retains control over what its assistants say and do. For teams moving past general-purpose AI tools and toward coordinated knowledge work, this layer of transparency and oversight turns assistants from a risk into a reliable resource.