When you type @assistant in a team message thread and an AI answers, everyone in that conversation should know the reply came from AI—not a person. And when that assistant offers to file a decision or create a task, someone should approve it before anything changes.
A messaging tool that labels AI composed messages is team chat software where AI-generated replies carry visible attribution showing they came from the assistant, not a human team member, and where every answer cites which documents it read or explicitly marks that it read nothing. Mycel's Team chat does exactly this: type @assistant in any channel or direct message, and the response appears in the thread with full transparency about what the assistant read from your business vault. If the assistant offers to file a decision note or add a task, that action waits in Approvals until a person accepts or rejects it—nothing happens automatically.
Key Takeaways
- A messaging tool that labels AI composed messages shows every AI reply with clear attribution and cites the vault documents the assistant read, or marks when it read nothing.
- Mycel's Team chat lets anyone type
@assistantin a channel or direct message, and the answer arrives in-thread with transparent sourcing from the business vault. - When the assistant offers to file a note or add a task, the proposal waits in the Approvals queue until a person reviews and accepts it—no automatic changes.
- Team chat with channels and direct messages is available only in Business workspaces; Personal and Personal Plus plans do not include team roles or messaging.
- Every business workspace starts with a #general channel, and team members control which folders the assistant can read based on role-based permissions set at the folder level.
Why AI Message Attribution Matters in Business Team Chat
Business messaging tools add AI assistants to speed up answers, draft replies, and suggest next steps. But when an AI response lands in a thread without a label, the person reading it may not realize a machine wrote it—and may act on information the assistant misread, hallucinated, or pulled from the wrong document.
Transparent labeling solves three problems at once. First, it sets expectations: the reader knows the reply is AI-generated and can verify the information before relying on it. Second, it surfaces the assistant's sources, so the team sees which vault documents informed the answer. Third, it prevents silent errors—when the assistant offers to change something (file a note, add a task, update a record), a human reviews that proposal before it takes effect.
Teams that run on shared knowledge—field service companies tracking service history, contractors managing job specs, property managers coordinating maintenance—need every answer to point back to the source document. A messaging tool that labels AI composed messages makes that chain of custody visible in the conversation itself.
How Mycel's Team Chat Labels AI Responses
Mycel's Team chat operates inside a business workspace. Every business workspace includes channels (every business starts with #general) and direct messages between the people on the team. When anyone types @assistant in a message, the assistant answers from the business vault and the reply appears in the thread.
Each answer shows which vault documents the assistant read to compose the response. If the assistant read nothing from the vault—because no relevant document exists, or because the question falls outside the vault's scope—the reply carries a label marking that fact. The assistant does not guess, and it does not claim certainty when it has no source to cite.
The assistant may also offer to file a decision note or add a task based on the conversation. That offer does nothing until a person opens Approvals, reviews the proposal, and accepts or rejects it. If a team member asks the assistant to update a customer record in Relay or add a line item in TradesBackbone, the proposed change waits for approval before Mycel sends it to the connected app.
This approval gate applies to anything the assistant would change inside another Martello app. Changes the assistant makes directly inside the vault—filing a decision note the owner accepts, for example—appear in the document history with a timestamp and the assistant's name, and any team member with vault access can restore a previous version.
What Business Workspaces Add to Team Messaging
Team chat, team roles, and private folders all belong to a business workspace. Personal and Personal Plus plans give one person a vault and an assistant; they do not include team roles or messaging.
A Business workspace supports up to 20 seats at $119 per seat per month ($1,190 per seat per year). Each seat includes everything in Personal—vault, assistant, 1,500 AI credits per month, Email in and Omi—plus Today (a cross-app list of what is waiting on each person), team roles, private folders, document history with restore, automations, and every add-on. Business workspaces also connect TradesBackbone, Relay, ReputeLift, and SynthPrism; those apps can file what the business learns into the vault, and their AI reads only the folders the owner shares with them.
Enterprise pricing past 20 seats is custom. Full details appear on the pricing page.
The Four Team Roles and What Each Can See
Every business workspace assigns one of four roles to each person: Owner, Admin, Member, or Viewer.
- Owner and Admin can read every folder, create and delete folders, set folder permissions, invite and remove team members, and approve assistant proposals.
- Member can read folders shared with team members, edit documents in those folders, and chat in any channel or direct message they are invited to.
- Viewer can read folders shared with everyone, but cannot edit documents or change folder contents.
A folder can be kept to everyone, to team members but not viewers, or to owners and admins only. When the assistant answers a question in Team chat, it reads only the folders the person who asked the question has permission to see—so a Viewer who types @assistant will receive an answer drawn from the folders shared with everyone, not from the owners-and-admins-only folder containing financial records.
Approvals: The Queue Where AI Proposals Wait
Mycel's Approvals screen lists every action the assistant has proposed that requires human review before it takes effect. When the assistant offers to file a decision note, add a task, or update a record in a connected Martello app, that proposal appears in Approvals with the conversation context, the proposed change, and a timestamp showing how long it has waited.
An Owner or Admin opens Approvals, reads the proposal, and either accepts or rejects it. Accepting the proposal executes the action—the note files in the vault, the task appears in the connected app, the update writes to the customer record. Rejecting it cancels the action and logs the rejection.
Nothing the assistant proposes happens automatically. This gate exists because AI can misinterpret a conversational request, apply an action to the wrong record, or file a note in the wrong folder. The approval step lets a human catch those errors before they propagate into the vault or into a connected app's data.
Every document in the vault keeps a history of who changed what and when. If the assistant files a note that the owner later decides was incorrect or unnecessary, the owner can open the document history and restore the previous version. The assistant's entries in the history carry a label showing they were filed by the assistant, not by a person.
How the Vault Powers Transparent AI Answers
Mycel is the memory a small business runs on: one shared vault of plain-markdown folders the team works in, with an assistant that answers from it. The vault stores the business's operating knowledge—service protocols, pricing sheets, vendor contacts, project specs, maintenance logs—organized into folders the owner creates and names.
When the assistant answers a question in Team chat, it searches the vault for relevant documents, reads the passages that match the question, and composes an answer that cites those documents by name. The reply shows the person who asked which files the assistant read, so they can open those documents and verify the information themselves.
If the vault contains no relevant document, the assistant says so and marks the reply as reading nothing from the vault. It does not fabricate an answer, pull from the internet, or guess based on the question's phrasing. This constraint—answer from the vault or mark the answer as unsourced—is what makes the assistant useful for a team that relies on shared records.
The Martello apps a business connects can file what the business learns into the vault. When a customer calls Relay with a question the team answers, Relay can file that answer as a note in the vault's customer-knowledge folder. When a job in TradesBackbone finishes and the team documents a process improvement, TradesBackbone can file that improvement in the vault's operating-procedures folder. The records themselves—customer profiles, job histories, review data—stay in Relay, TradesBackbone, ReputeLift, and SynthPrism, one click away from the vault.
Relay's and TradesBackbone's AI read only the folders the owner shares with them. ReputeLift's AI reads the business vault to understand how the business talks about its services when drafting review responses. SynthPrism's assistant reads the vault for brand voice and business facts when generating marketing copy. Learn more about how Mycel works.
Practical Uses for AI-Labeled Messaging in Small Business Teams
Field Service and Maintenance Teams
A field service company with six technicians stores service histories, equipment manuals, and troubleshooting guides in the vault. A technician on-site types @assistant in the #service channel and asks which valve replacement procedure applies to a specific boiler model. The assistant answers with a citation to the vault document, and the technician opens that document to confirm the steps before starting work.
If the technician discovers a step in the procedure that no longer applies and asks the assistant to update the document, the assistant offers to file a revision note. That proposal waits in Approvals until the service manager reviews the suggested change and either accepts it (updating the procedure document) or rejects it (leaving the original procedure intact and asking the technician to clarify).
Property Management and Contractor Coordination
A property management company manages 40 rental units and stores lease terms, maintenance schedules, and vendor contact lists in the vault. The property manager types @assistant in a direct message and asks when the HVAC contract for a specific building renews. The assistant answers with the renewal date and cites the vendor-contract document in the vault.
A tenant reports a leak, and the property manager asks the assistant to create a maintenance task in TradesBackbone for the plumber the company uses for that building. The assistant offers to add the task with the plumber's contact information pulled from the vault. The proposal appears in Approvals, the property manager verifies the plumber and the unit address, and accepts the task. TradesBackbone receives the task, and the plumber sees it in their queue.
Retail and Wholesale Businesses
A wholesale distributor with a small sales team stores product specifications, pricing sheets, and inventory policies in the vault. A salesperson types @assistant in the #sales channel and asks what the current lead time is for a product line. The assistant answers with the lead time and cites the inventory-policy document that contains it.
A customer asks for a custom quote, and the salesperson asks the assistant to file a decision note documenting the discount the sales manager approved. The assistant offers to file the note in the pricing-decisions folder. The sales manager sees the proposal in Approvals, checks that the discount matches what they approved verbally, and accepts it. The note files in the vault, and the next time anyone asks about that customer's pricing, the assistant cites the decision note.
Comparing Team Messaging Tools With and Without AI Labeling
| Feature | Team Chat Without AI Labels | Mycel Team Chat | |---------|------------------------------|-----------------| | AI response attribution | AI replies may appear as regular messages with no indication they were machine-generated | Every @assistant reply shows it came from AI and cites vault documents or marks that it read nothing | | Source transparency | Answers typically do not link back to specific documents the AI read | Each answer lists the vault files the assistant read to compose the response | | Action approval | AI may execute tasks, file notes, or update records immediately when asked | Assistant proposals wait in Approvals queue until a person reviews and accepts them | | Message history | Chat history shows who sent each message, typically without distinguishing AI edits from human edits | Vault document history shows which changes the assistant filed, with restore for anything it added | | Folder-level permissions | Access control varies by tool; check vendor | Folders can be kept to everyone, team members only, or owners and admins only; assistant reads only folders the person asking has permission to see | | Business focus | General team collaboration or consumer messaging | Built for small businesses that store operating knowledge in a shared vault and need the assistant to answer from that vault |
What the Assistant Can and Cannot Do in Team Chat
The assistant in Mycel's Team chat answers questions by reading the business vault. It does not search the internet, pull from a general knowledge base, or train on the team's chat history. Every answer either cites vault documents or explicitly states it read nothing from the vault.
The assistant may offer to file a decision note, add a task, or update a record in a connected Martello app when the conversation calls for it. These offers appear as proposals in the thread, and the action waits in Approvals until a person reviews it.
The assistant does not automatically send messages, schedule reminders, or notify anyone outside the chat thread where it was invoked. It does not read private direct messages unless someone in that direct message types @assistant and asks it a question. It does not monitor channels for keywords or surface information without being asked.
Mycel does not build workflows, agents, or prompt templates. The assistant operates from a fixed interaction model: you ask, it answers from the vault, and it offers to file what the conversation decides. That model keeps the assistant predictable and the vault clean.
How Vault Export Works When the Team Needs a Backup
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 contains the current version of each document, not its version history.
An Owner or Admin opens the Vault screen, selects export, and downloads a .zip archive containing every folder and document in the vault as plain .md files. The folder structure in the export mirrors the folder structure in the vault, so the exported files open in any text editor, markdown app, or version-control system without conversion.
This export model serves two purposes. First, it gives the business a portable backup of its operating knowledge that does not depend on Mycel remaining available. Second, it lets a team migrate to a different tool or archive the vault when the business winds down, without losing the records or their timestamps.
The export does not include the assistant's internal search index, the document version history, or the team chat message threads. It contains the knowledge the business filed in the vault, structured as markdown.
Frequently Asked Questions
What does it mean when a messaging tool labels AI composed messages?
A messaging tool that labels AI composed messages shows every AI-generated reply with clear attribution indicating it came from an assistant, not a human team member, and cites the documents or sources the assistant read to compose the answer. Mycel's Team chat displays which vault documents the assistant read for each response, or marks when the assistant read nothing from the vault, so everyone in the conversation knows where the information came from.
Can the assistant in Mycel change things in the vault without approval?
The assistant offers to file decision notes or add tasks based on conversations, but those proposals wait in the Approvals queue until an Owner or Admin reviews and accepts them. Anything the assistant would change inside a connected Martello app—such as updating a customer record in Relay or adding a job note in TradesBackbone—also waits for approval before Mycel sends the change to that app. No action the assistant proposes executes automatically.
Do Personal and Personal Plus plans include team messaging?
No. Team chat, team roles, and private folders belong to a business workspace. Personal plans ($50/month or $500/year) and Personal Plus plans ($89/month or $890/year) are for one person and include a vault and assistant but not team messaging. Business workspaces start at $119 per seat per month for up to 20 seats and include Team chat with channels and direct messages.
How does folder permission control what the assistant can read?
A folder can be kept to everyone, to team members but not viewers, or to owners and admins only. When someone types @assistant in Team chat, the assistant reads only the folders that person has permission to see based on their role. A Viewer asking the assistant a question will receive an answer drawn only from folders shared with everyone, while an Owner or Admin asking the same question will receive an answer that may include information from owners-and-admins-only folders.
What happens to the document history when the assistant files a note?
Every document in the vault keeps a history of who changed what and when, including changes the assistant made. Entries the assistant filed appear in the document history with a timestamp and a label showing they came from the assistant. An Owner or Admin can open the document history and restore a previous version if the assistant filed something incorrect or no longer needed, undoing the assistant's addition.
Can I export the vault with all the version history included?
No. The vault export contains the current version of each document as a plain-markdown file with created and updated dates and metadata in the frontmatter. The export does not include the version history for each document, the team chat message threads, or the assistant's search index. It gives you a portable backup of the knowledge the business filed in the vault, structured as markdown files you can open in any text editor.
Team Messaging That Shows Where AI Answers Come From
Transparent AI attribution in team chat matters when the business runs on shared knowledge and every answer needs to point back to a document someone can verify. A messaging tool that labels AI composed messages shows the team which replies came from the assistant, which vault documents it read, and gives someone the chance to review any action the assistant proposes before it takes effect.
Mycel's Team chat puts that transparency into the conversation thread itself—@assistant brings an answer with citations, and every proposal waits for approval. The vault holds the operating knowledge, the assistant reads from it, and the team sees the chain from question to document to answer. That visibility keeps the assistant useful and the vault trustworthy, even as the business grows and more people ask it questions.