Your Fellow Agent Documentation
Everything you need to answer website customers while they are still interested. From the first website scan to site launch, customer conversations, and owner handoff.
Overview
Your Fellow Agent turns your website into a real coworker: an AI agent that learns your business, answers everyday questions, uses connected business tools, and brings your team into the conversation when customers need your help.
Key capabilities
Why This Is Different
A customer who cannot find an answer may leave instead of filling out a form. Your Fellow Agent keeps the conversation moving while the need is still fresh.
Why a promising website visit ends
The visitor wants to know whether you offer a service, have availability, or can handle a particular request. If the answer is not obvious, the next business is one click away.
A contact form does not answer the question or make the next step feel certain. The customer may leave before the business ever sees the message.
How Your Fellow Agent is different
When your agent needs help, it doesn't abandon the customer. It asks clarifying questions when useful, collects only the contact fields you configured, and loops in a human with the full background.
Quick Start
Build your agent, test it, and install it with one line of code.
Create the agent from your website
Enter a URL and the system starts crawling pages, extracting business info, and drafting a starter FAQ. Your first agent is typically ready to test in about 90 seconds.
Review setup and training
Pick an avatar, set the agent name, choose tone, review website coverage, and add docs or manual URLs.
Set up connections
Choose contact fields, open the live dashboard, and add email or Slack alerts when your team needs to jump in.
Install once
Claim your business, then copy the script and paste one line of code before </body>. Manage launcher changes from Your Fellow Agent after that.
Your dashboard stays available even if a setup email is delayed. We do not send a setup-ready email after its notice has become stale.
Start: Website Scan
Start by giving the agent your website so it can learn the basics before you customize it.
What happens during the scan
Give Your Agent a Presence
Choose the face, name, and personality customers will interact with.
Configuration options
Style the Customer Experience
Customize the conversation, launcher, and site appearance to match your brand.
Styling options
Train It on Your Business
Review website knowledge, add FAQs, upload documents, and test how the agent responds.
Knowledge sources
- Edit auto-generated FAQ responses from the website scan
- Add your own Q&A pairs manually
- Import/export Q&A as plain text (Q: and A: format)
- Supported formats: PDF, DOC, DOCX, TXT, and CSV
- Maximum file size: 10MB per file
- Up to 5 files per upload batch
Choose Contact Fields
Choose what the agent should collect when your team needs to follow up.
Available fields
The agent should only ask for the fields you enable. If you collect name and email but not phone, it should not keep asking for a phone number.
Connect Your Team
Decide how your team sees customer threads and how the agent should ask for help.
Trigger phrases
Your agent answers from business knowledge and considers connected tools when a customer needs something done. When staff help is needed, your existing reply settings determine whether staff contacts the customer directly or responds through the agent. Sending a notification does not mean a person has read it or started work. Your agent can continue answering questions and using applicable tools while a staff request is pending.
Custom triggers
Add industry-specific phrases that are strong signals for your business. These guide the agent, but the model still uses the conversation context.
Owner instructions
Customize what the agent says when looping you in and how it sets expectations.
Slack
Connect Slack when your team wants alerts and customer threads in a shared workspace.
Setup process
Connect your workspace
Click 'Add to Slack' to authorize the Your Fellow Agent app in your workspace.
Open the alerts channel
The app creates a business-named alerts channel with links to customer conversations. Reconnecting reuses it when available.
Review customer requests
Each conversation has its own channel. Requests marked 'Needs your team' include a summary and a link to reply.
Advanced Settings
Fine-tune launcher behavior and enable power features.
Launcher placement
Additional options
Images and documents (PNG, JPG, WebP, GIF, HEIC, PDF, TXT, CSV, DOCX). Up to 3 per message, 5MB each. Drag/drop or paste supported.
Collect messages without agent replies
How Connecting Your Team Works
The complete flow from customer request to resolution.
Asking your team is a feature, not a failure
When the agent needs help, it acts like a coworker. It keeps the conversation natural, asks clarifying questions when they help, gathers the details you configured, and sends you a simple email or Slack note with the background. No separate acknowledgement email is sent by default; the customer stays in the same thread.
You reply however you want - "tuesday 4 works" becomes a polished customer response when agent replies are enabled. In escalation-only mode, the email makes it clear that you should reply to the customer directly.
How a customer thread moves forward
Asks a question, describes a problem, or requests human help
Answers when it can, asks useful follow-ups, notifies your team, or decides a team member should be involved
Gathers the configured fields, such as name, email, phone, address, or a custom field
Tells the customer it is getting the team involved and keeps the thread open
Via email, Slack, or both, with context and customer details included
You reply from email, Slack, or dashboard, or contact the customer directly
Notify vs Ask the Team
The agent does not need to treat every unusual customer message as a hard handoff. It can choose the lightest useful level of team involvement. An email address supplied for an order or booking lookup stays part of that task; it does not automatically trigger a request to your team.
Routine questions stay with the agent. It answers from your website, FAQs, documents, and training without bothering your team.
For high-value, sensitive, or unusual conversations, the agent can keep helping while your team gets a heads-up in the dashboard, email, or Slack.
When the agent should not guess, it asks useful follow-ups, collects the configured contact fields, and creates a thread for your team to answer.
Signals That Your Team May Be Needed
The agent can loop you in when:
When customers use specific words like "human", "manager", "refund", or any custom triggers you've configured. These guide the agent rather than replacing its judgment.
The agent recognizes when customers are getting frustrated or repeating themselves.
Questions outside the agent's knowledge base loop you in rather than guessing.
Customers can always ask to speak with a human directly.
The Team Flow
When the agent loops you in, it follows a natural process:
The agent identifies what the customer is trying to do and whether it can help directly.
For service requests, troubleshooting, quotes, or appointments, it may ask one or two practical follow-ups before collecting contact info.
Asks for the fields you enabled, and does not keep asking for fields you turned off.
Sends a plain coworker-style email or Slack note with what the customer needs, what has already been said, and how to answer.
Team Channels
Dashboard
- Customer conversation inbox with automatic refresh
- View every message and current status
- Reply from the portal when agent replies are enabled
- Useful for teams handling many conversations at once
- Plain-language note from Your Agent [name]
- Explains what the customer needs and includes the relevant context
- Customer contact summary
- One-click dashboard link (magic link auth)
- Reply directly when you want the agent to relay your answer
Slack
- Posts to your configured coordination channel
- Includes customer info and request summary
- Links directly to the customer's conversation channel
- Mention the agent to give instructions
- Choose Take over to send your own replies as written
Responding to Customers
Three ways to respond
If agent replies are enabled, reply to the alert with what you want to say. The agent turns it into a clear customer response and sends it back in the same conversation.
Open the conversation channel. Mention your agent to give instructions, or choose Take over to send your own replies as written.
Click the magic link in your email to go directly to the conversation in your portal inbox.
Dashboard: Customer Threads
An inbox for the customer questions that need your judgment.
Inbox features
- Opens on Needs you so waiting customers are never buried
- Move between Needs you, Open, and Handled. Open means unresolved, not that a customer is waiting.
- Search by customer name or content
- Sort by last activity
- Contact requests collected when the installed agent's full answers are paused
- Sender info (name, email) if provided
- Mark as read functionality
Customer Thread Detail
What you see in each thread
- Full message thread between the customer, agent, and owner
- Customer contact information
- Timestamps for all messages
- Team involvement status and category
- Send instructions for your agent to write the customer's reply, without taking over the conversation
- No response needed closes the request without sending a customer message
Agents
Manage multiple agents
- Grid view of all your agents with avatars
- Per-agent stats (conversation count, last active)
- Quick actions: Configure, View Inbox
- Create new agent button
Widgets
Primary floating agent vs inline widgets
Your Fellow Agent supports two deployment formats:
The core site-wide coworker that sits in the corner of your website. It features the floating Chat Toast card (up to 120 characters), adaptive hesitation intelligence, and the Klaviyo Email Form Optimizer with chat focus shielding.
Stand-alone widgets embedded directly inside specific page containers (such as FAQ hubs, landing page contact forms, or consultation request sections).
Widget types
Answer questions directly from your knowledge base. Great for help centers.
Collect contact info conversationally. Perfect for landing pages.
Collect preferred dates, times, and contact details for your team to confirm. It does not reserve calendar availability.
Custom agent interactions with your own instructions.
Team Management
Team roles
| Permission | Member | Admin | Owner |
|---|---|---|---|
| View conversations | |||
| Respond to customers | |||
| Train agent | |||
| Configure agents | |||
| Invite team members | |||
| Promote/demote or remove members |
Team management actions
- Invite new members by email with role assignment (Owner/Admin)
- View pending invitations with expiration dates
- Resend or cancel invitations that haven't been accepted (Owner/Admin)
- Owners can promote/demote members between roles
- Owners can remove team members
Agency & Client Setup
Build an agent for a client
Agencies can use the same website-first setup as a business owner. Enter the client website, review and test the agent, set the client's contact and handoff details, then install the generated script when the client is ready.
- Build and test an agent before the client installs anything
- Keep each agent's website knowledge, appearance, and contact settings separate
- Invite client team members as admins or members when they need access
- Let the client respond through email, Slack, or the dashboard
Settings
Settings
Update these from each agent's configuration
Install widgets and manage connected channels
Choose who can answer customer questions
Owner Replies
Reply from Slack or email without opening the dashboard.
Owner replies via Slack
Open the customer's conversation channel to give your agent instructions or answer directly:
- Choose Take over, then send your replies without mentioning the agent. Your messages are sent as written.
- Mention your agent to give instructions instead of writing the customer's reply yourself. This does not pause the agent. Only Take over pauses automatic replies.
- Choose Return to the agent when finished. Control also expires after 10 minutes without a team reply.
@Julie [instruction]Ask the agent to turn your instruction into a customer reply@Julie doneEnd remote control and return to the agent@Julie extend 15Extend the timer (minutes)@Julie statusView remaining time and controller
Remote control via email
Every alert email includes a special reply-to address:
If agent replies are enabled, reply to the email and your message is routed to the right conversation, polished, and delivered to the customer.
Customer requests, contact details, and files are delivered to the business and its authorized team to handle the request. Necessary service providers and separately authorized AI clients process information as described in the Privacy Policy. Keep your own website notice consistent with your configuration; enabling an integration does not grant blanket customer consent or permission for unrelated marketing.
Training Mode
Correct agent responses in real time to improve future answers.
For help using Your Fellow Agent, open the support chat in your portal. It uses your verified login, current workspace, and page area to explain settings and setup steps. Public pages, private account screens, and the verified admin area receive different guidance; your role alone does not determine the kind of help. You do not need Training Mode for product help. This support integration does not change account settings for you. Other SaaS products can supply signed role and surface context through an authenticated integration; installing the widget alone does not verify a visitor's access.
The merchant menu is Inbox, Agents, Widgets & Channels, Team Members, Billing, and Settings. For knowledge, handoff, or installation, open Agents, select your agent, and choose Configure. The Email Form Optimizer (Klaviyo) is in the installation section, not a separate Extras menu. Administrator-only menus are not part of merchant navigation. Support guidance explains these paths but cannot confirm your account settings without an authorized lookup.
How to use training mode
Generate a training PIN
Go to your agent's configuration page and generate a time-limited PIN.
Enter training mode
In the chat widget, type 'enter training mode' and provide the PIN when prompted.
Correct responses
If the agent gives a wrong answer, simply respond with the correct one. Say something like 'No, the correct answer is...' or 'Actually, it should be...'. The agent will save your correction automatically.
Reset and test
Click 'Reset Chat' in the training panel to start fresh, then ask the same question to verify your correction was learned. Repeat this process to dial in your agent's responses.
Exit training mode
Type 'exit training mode' when finished. All corrections are saved automatically.
Trial & Offline Mode
Collect leads and customer messages without agent responses during setup, downtime, or an expired unpaid trial.
Full trial, then contact-form mode if unpaid
New businesses get 14 days of full agent functionality beginning when the agent is installed on the business website and has its first real customer conversation. Creating, claiming, configuring, or privately testing it in the dashboard does not use trial time. If the trial expires without active billing, the installed widget stays on the site but switches to contact-form mode for new and existing conversations. It stops generating agent answers and owner-reply workflows, but it continues collecting visitor messages and contact details so you never miss a customer request.
- Full agent behavior during the 14-day trial
- Contact-form mode after an unpaid trial expires
- Manual pause mode when the owner wants to collect leads without agent replies
- Private dashboard testing before installation and after expiration
Offline message
Customize what visitors see when the agent is offline:
Keeping Knowledge Current
Ways to keep an agent aligned with a changing business site.
Current Your Fellow Agent update paths
The first website scan happens during setup. Fast previews organize the first pages into short, source-linked topics so the agent can start answering quickly. On an installed site, the agent also receives the visible page section when it changes during a conversation. Owners can run a fresh scan at any time. When scheduled refresh is configured, it updates changed pages for fast previews; other agents continue using the full knowledge-rebuild path.
From Configure, owners can trigger a fresh crawl when services, pricing, hours, policies, menus, or support pages change.
Add specific URLs so important pages, like pricing, FAQs, menus, service-area pages, or documentation, are included in the next crawl.
Manual Q&A and training corrections take priority over scanned website content. If a knowledge recompile fails, your saved training corrections remain in place.
Upload PDFs, Word documents, text files, or CSV files when the source of truth is not on the public site.
Advanced refresh options
These automated refresh and integration options are not all enabled by default. Confirm availability with support before relying on them. Use the rescan, document, Q&A, and correction controls above for updates you need now.
The scheduled refresh checks active sites for changed pages. Fast-preview agents update the affected website topics; other agents queue a full knowledge rebuild.
Watch sitemap.xml and page last-modified values to discover changed or newly published pages.
Trigger refreshes from Shopify, WordPress, Webflow, Squarespace, or a custom CMS whenever products, pages, posts, or policies are published.
Hash page text, compare diffs, and re-embed only the changed sections instead of rebuilding everything.
Sync product catalogs, inventory, hours, locations, pricing, or booking availability from APIs instead of scraping rendered pages.
Use unanswered questions, corrections, and owner replies to suggest knowledge updates for approval.
Stage changed answers before publishing when accuracy matters, especially pricing, policies, legal, medical, or financial content.
Notify owners when high-traffic answers are stale, pages disappear, crawl fails, or important site sections have not been refreshed recently.
Billing & Plans
Start with the complete web agent free for 14 days. Launch plans begin at $29 per month, with two months free on annual billing.
Launch plans
Starter is $29 per month for 100 customer conversations, one agent/site, and two seats. Business is $79 per month for 500 conversations, one agent/site, and five seats. Growth is $199 per month for 2,000 conversations, up to three agents/sites, and ten seats. Annual billing costs ten months ($290, $790, or $1,990), while included conversation allowances reset monthly. Every plan includes the complete web agent.
Conversations and overages
A customer conversation is one customer thread or session, not every individual message. Owner replies are part of that thread and are not metered separately. Additional-conversation rates are $0.40 each on Starter, $0.25 on Business, and $0.12 on Growth. Check your billing section for current usage and your invoice for billed charges. Contact support if the figures do not match what you expect.
Included on every paid plan
Add-ons
Higher conversation allowances, extra agents/sites, conversation packs, and white-glove onboarding can be arranged with the team. SMS follow-up is available in a limited beta for selected agents; contact our team about access. Setup and provider approval are required, and texting is not automatically included in every plan. Dedicated client numbers and voice agents are not generally available.
Slack Integration
Full integration with Slack for notifications and conversation management.
Setup process
Install the app
Click 'Add to Slack' in your agent's configuration to authorize the Your Fellow Agent app.
Open the alerts channel
The app creates a business-named alerts channel. Existing channel settings are preserved, and reconnecting reuses an available saved channel.
Start receiving notifications
Conversation links appear in the alerts channel. Customer messages and reply controls appear in each conversation's own channel.
Slack capabilities
The conversation opening includes a How to reply guide and Take over / Return controls once. Later messages show the conversation, with your agent's name and picture, not repeated instructions. Reconnect Slack if it asks for permission to customize the sender. Use the account mentioned in the opening guide to give instructions. Channels are public within your Slack workspace.
Handoff alerts include a request summary and a link to the conversation.
Pause agent replies and send your own messages as written. Return control when finished.
Give instructions for a customer reply. This is different from sending your own message during takeover.
Email Integration
Automatic email notifications and reply handling.
Email types
A short note from your agent with one request summary, contact details, and a conversation link. Reply with instructions when agent follow-up is enabled, or contact the customer directly when it is disabled.
Notification when someone leaves a message while the agent is offline.
Preview activity, trends, and unanswered questions from the dashboard. Automatic weekly delivery is not currently enabled.
Replying to emails
Each alert email has a special reply-to address that routes your response to the correct conversation:
From: Julie <[email protected]>
Reply-To: [email protected]
Sarah wants to reschedule her highlights appointment and asked whether Friday afternoon is available.
What should I tell her?
Reply naturally and I will take care of the response. If I do not need to respond, reply "no response needed."
- JulieChoose how your agent helps
Start with what is included
Your agent answers from business knowledge, gathers useful details, and asks staff for help. The built-in scheduler collects a request, not a calendar reservation. Live order lookups, return labels, and bookings require a connected business service that supports those actions.
Decide who handles staff follow-up
In your agent configuration, review the owner contact, contact fields, and agent email-reply setting. With email relay enabled, staff can reply through the agent. With it disabled, staff handles customer email follow-up directly. Connected tools can work in either arrangement. An alert means the request was sent, not that someone has read it or joined the chat.
Email conversations versus a monitored inbox
Continuing a website conversation by email uses a conversation-specific reply address. It does not automatically connect your existing support mailbox. General business-inbox monitoring is not a self-service launch feature. Do not forward an entire mailbox without a supported routing setup. In configured Cloudflare inboxes, messages with attachments go to the designated person with the available files. The agent does not automatically interpret those attachments. Other channels can have different attachment support.
When something does not work
- Widget missing: confirm the snippet is on the published page, check blocked script requests, and verify installation status.
- Wrong answer: review the source, add a correction, and repeat the question in a fresh private test.
- Order not found: check the order and email, then confirm the connected service has that order. Not found is different from outside the return window.
- Pending action or timeout: check the existing operation before retrying. Do not create another booking, payment, or return to test whether the first worked.
- Broken document link: keep the original message and report the link failure. Do not request a second transaction just to obtain another link.
- Missing email: check the recipient, spam folder, and thread. Ask support to distinguish processing time from send time and delivery time.
- Agency account: select the correct client workspace before changing settings or billing. Test each client separately and confirm who receives alerts.
For help, email [email protected] with the agent, approximate time, and what you expected. Do not send passwords or API keys.
Model Context Protocol (MCP)
Connect Your Fellow Agent directly to Claude, Codex, Cursor, and other AI tools.
What is the MCP integration?
Claude and ChatGPT directory listings are being prepared, not published. Use the direct MCP connection below for now. Tools describe whether they read settings or can change data and run external actions. These descriptions do not replace your permission approval. Preview conversations can run enabled actions or notify your team; use a sample agent without external connections for safe testing.
The Model Context Protocol (MCP) allows AI coding assistants, chat apps, and custom agents to interact directly with Your Fellow Agent. You can build and test an agent for your website right from chat, or manage your existing business knowledge without opening the web dashboard.
When connecting an app, choose its permissions separately for agents, knowledge, connected services, business actions, channels, and billing. Reading and changing each area are separate permissions. Your coding assistant only sees commands it can use. Changes to owner email or phone, email follow-up, Slack, or SMS require channel write permission as well as agent write permission. Existing connections need fresh approval for the new permissions; they do not gain access automatically.
Ask your coding assistant to read get_yfa_guide first. The guide covers setup, business tools, examples, testing, installation, billing, and troubleshooting. Clients that support MCP resources can also read yfa://guide/overview. Your own services run on your hosting; Your Fellow Agent connects them to customer conversations. Use test services before enabling actions that send emails, buy labels, or change orders.
Report action completion and customer-email sending separately in your tool results. Creating a return label or notifying staff does not confirm a customer email was sent. Include an explicit customer-send result or provider acceptance receipt, and return the label link so the customer can use it in chat. Acceptance by an email provider does not prove inbox delivery.
Connect without an upfront account or API key. Tell your AI assistant to set up customer support for your website, and it will analyze your site, begin training, and return an interactive live chat preview and claim link in ~90 seconds.
Sign in with your owner account to inspect agent configurations, update business knowledge and Q&A, run test conversations via the preview channel, and check installation or billing status.
Zero-Dashboard Setup: Build, Calibrate, and Embed in Chat
Agencies, developers, and power users can configure complete customer-facing agents without ever opening the web dashboard. Your coding assistant (Claude Code, Codex, Cursor, or Cline) handles every step through MCP tools:
Prompt: "Set up customer support for https://example.com". The assistant calls setup_agent to scrape public pages, synthesize starter knowledge, and draft Q&As.
Prompt: "Ask the agent about pricing and return policies". The assistant calls test_agent to evaluate responses right in your terminal or chat window.
Claim the agent in the browser and authorize the connected session first. Then ask: "Get the embed script and install it in our layout". With your permission and website access, the assistant can insert the script before </body>. Verify installation detection afterward.
Use list_teams and switch_team to switch between client accounts and configure distinct brand settings without repeated sign-ins.
/mcp/pair?code=...) to authorize your device.Connecting your AI client
Your Fellow Agent exposes a standard Streamable HTTP MCP endpoint at:
https://yourfellowagent.com/mcpDesktop Setup (Claude Desktop)
In your claude_desktop_config.json, add the server using stdio bridging:
{
"mcpServers": {
"yourfellowagent": {
"command": "npx",
"args": ["-y", "mcp-proxy", "https://yourfellowagent.com/mcp"]
}
}
}Codex CLI
codex mcp add yourfellowagent --url https://yourfellowagent.com/mcpAvailable Public Tools
Analyzes any business website and begins training an agent in ~90 seconds. If the site is already registered, returns a 1-click pairing link to authorize your IDE; if new, returns a live chat preview and claim link.
Checks whether a domain is already registered on Your Fellow Agent or available for new agent setup. For claimed domains, generates a 1-click pairing link for browser authorization.
Checks the crawl status, training phase, and chat readiness for an agent.
Test an agent right in your chat/IDE session with zero login required for new draft agents. For claimed agents, requires an authorized session.
Retrieves the one-line embed snippet. Draft snippets can be inspected without login, but the agent must be claimed before production activation. Claimed agents require an authorized session; website changes still require your permission.
Completes browser approval using the pairing code and the private device secret returned to your MCP client. Your client keeps the secret; you only open the approval link. Start a new connection if approval takes longer than 15 minutes.
Authenticated Workspace & Management Tools
Once paired, AI assistants can configure, customize, and operate agents across all your businesses:
Enables multi-brand owners and agencies to view all accessible client workspaces and switch between them instantly without re-authenticating.
Inspect and update business details, owner handoff contacts, and widget appearance, including positioning (left or right side), brand colors (primaryColor), launcher styles (bubble or edge-tab), and desktop modes (widget or sidebar). Use get_agent_channels to inspect channel settings and connect_agent_channel for the Slack authorization link, then return to your conversation to finish setup.
Add additional URLs or raw text from policies, price lists, and business notes directly into the agent's synthesized knowledge base.
List agents in the active workspace, continue a private test by passing its conversation ID, and retrieve the one-line installation code. Private tests work after trial expiration. Installation is complete only when the production widget is detected.
Custom Actions & External MCP Tool Engine
Connect external MCP servers or custom HTTP APIs so your agent can take action on behalf of customers (look up live orders, generate return shipping labels, check appointment availability, or issue bookings):
Ask Claude or Codex, connected to Your Fellow Agent, to add your business tool. It can connect the server, show you the actions it finds, and enable the ones customers should use. Refreshing the connection preserves those choices. Your developer or coding assistant hosts the business function; Your Fellow Agent handles the customer conversation. The regular dashboard shows connected tools without requiring you to enter endpoint URLs or credentials there.
Connection credentials are encrypted at rest and kept outside the agent's model inputs. Some services ask you to authorize access in a browser. You can then return to Claude or Codex to finish configuration.
Use manage_tool_connection to create, refresh, update, or disconnect a service. Use manage_custom_tool to configure an action, remove it, or register an HTTP action. Each call includes a change object with the chosen operation. inspect_tool_connections and list_tool_executions show operational status without revealing credentials or customer arguments.
Widget Embed Codes
Installation snippets for your website.
Primary agent widget
Add this before your closing </body> tag. Paste it once, then change colors, launcher style, side, tab text, and uploads from Your Fellow Agent:
<script
src="https://yourfellowagent.com/widget.js"
data-bot-id="YOUR_BOT_ID"
defer>
</script>data-config-source="inline".Embedded widgets
Embed a widget inline on your page (great for landing pages, contact pages, or documentation):
<div id="yfa-YOUR_WIDGET_ID"></div>
<script
src="https://yourfellowagent.com/embed.js"
data-widget-id="YOUR_WIDGET_ID"
data-container-id="yfa-YOUR_WIDGET_ID"
></script><div style="max-width: 400px; margin: 0 auto;">
<div id="yfa-YOUR_WIDGET_ID"></div>
</div>Widget styling tips
The widget inherits styles from its container. Use CSS to control width, max-width, and positioning. For centered widgets, use margin: 0 auto.
Widgets are responsive by default. On mobile, they'll adapt to the available width. Consider using max-width: 100% for fluid layouts.
Use one embedded widget per page. Set data-container-id to the ID of the element that should contain it. Multiple inline instances on the same page are not supported.
Primary widget configuration
| Setting | Values | Default |
|---|---|---|
| data-bot-id | Your agent ID | (required) |
| Desktop | Bubble or side tab | Bubble |
| Mobile | Side tab or bubble | Side tab |
| Side | Left or right | Right |
| Tab text | Short label | "Ask us" |
| Side tab position | 20 to 80 percent down the page | 52% |
| Image uploads | On or off | On |
CMS Platform Guides
Step-by-step instructions for adding the agent to your website platform. During setup, our scanner automatically detects your platform and pre-selects the matching guide.
- In WP Admin, go to Plugins → Add New and install the free WPCode plugin.
- Navigate to Code Snippets → Header & Footer.
- Paste your script into the Footer box and click Save Changes.
Alternatively, paste before </body> in your child theme's footer.php.
- In Shopify Admin, go to Online Store → Themes.
- Click … (Actions) → Edit code on your active theme.
- Open
layout/theme.liquidfrom the sidebar. - Scroll to the bottom, paste the snippet right before
</body>, and click Save.
- In Squarespace, go to Settings → Advanced → Code Injection (or Developer Tools).
- Scroll down to the Footer text box.
- Paste the snippet and click Save at top left.
- In Wix Dashboard, go to Settings → Custom Code.
- Click + Add Custom Code and paste your script.
- Set "Add Code to Pages" to All Pages (Load code once).
- Set "Place Code in" to Body - end and click Apply.
- Open Project Settings → Custom Code.
- Scroll down to Footer Code (before </body> tag).
- Paste your snippet, click Save Changes, and publish your site.
- Open Site Settings → General.
- Scroll down to the Custom Code section.
- Paste into End of <body> tag and click Save.
- Publish your site updates.
- In your GTM Workspace, click Tags → New.
- Choose Custom HTML as the Tag Configuration and paste the script.
- Set Triggering to Initialization - All Pages (or All Pages).
- Save the tag and click Submit → Publish.
- In Square Online Dashboard, go to Settings → Tracking Tools.
- Click Add new code / Custom Code.
- Paste the snippet into the Footer code area and save.
Website Performance & Compatibility
How the widget loads and what to check after installation.
The install script uses defer. It downloads without blocking HTML parsing and executes after parsing completes; it does not wait for every image to load.
Test representative pages before and after installation. The widget uses network and browser resources; performance depends on your site, device, and connection. Search rankings and unchanged performance scores are not guaranteed.
Namespaced selectors reduce style conflicts. Check launcher placement, fonts, and chat controls against your theme, especially on mobile.
The snippet loads the widget separately from your page. Check the browser network panel for its current transfer size rather than relying on a fixed size estimate.
Loading time depends on caching, network conditions, and visitor location. Keep the generated install snippet; static code delivery can change without replacing it. Customer conversations and live configuration are not shared-cacheable. If the widget does not appear, check the script request and your site's content security policy.
Paste the snippet once. Any visual changes (colors, tab text, side, avatars) update dynamically from your dashboard with zero code changes.
Analytics & Reports
Track performance and identify opportunities.
A bounded MCP business-context reader is implemented for stored knowledge, recent customer conversations and website activity. Once deployed and authorized, your connected AI can use that data for its own analysis without starting another agent conversation. Conversation and analytics access have separate read permissions. Existing subscription/trial access and request limits apply; these reads do not count as customer conversations. The reader does not execute inventory tools, research competitors or send messages. Deployment and real-client testing remain unverified; directory publication is separate.
Updated widgets collect public-page activity without requiring chat: page views, active foreground time, observed exits, link-click categories, referring origins and supplied campaign tags. A random session cookie groups activity for a visit, without identifying people or classifying returning visitors. No new banner or approval step is added. Private routes and sensitive URL details are excluded. A private reporting backend summarizes activity by day, hour, page and source, but does not add an Analytics tab, new charts, automatic purchase detection or a reply-to-report conversation. Deployment and event receipt must be verified before relying on it.
A limited, administrator-run pilot can compare no visible agent, a quiet launcher, and delayed proactive toasts with adjustable traffic weights. It is off by default, not self-service or a published conversion result. Help can wait for a quiet moment and stay suppressed after active use. Quiet mode preserves chat and team handoff without hints or pulses. Verified purchases and leads require a server-side outcome integration; a thank-you page or phone click alone is not a confirmed sale or lead. Optional gradual optimization uses mature verified conversions, not chat volume or visit duration. Adaptive results do not establish a fixed-split causal lift, and no automatic performance refund is offered.
The agent can use the original landing page, supplied UTM campaign tags, and the last five recent public pages as context for a chat. These are clues to a visitor's task, not an exact search query or evidence of a completed conversion. Keeping entry tags across page loads requires attribution-storage permission. Private-page journeys are excluded; see the visitor journey integration guide for consent and custom private-page exclusions.
Dashboard metrics
Total chats this week with trend indicator
Estimated time saved by automation
Percentage handled without human help
Conversations awaiting your response
Peak days and hours for traffic
Unread messages from offline mode
Weekly digest preview
Preview a weekly summary from the dashboard including:
- Customer conversations active in the past seven days and pending team requests
- Questions and concerns supported by customer conversations
- Practical suggestions and pages associated with conversations, with sampled results identified
FAQ
Common questions about Your Fellow Agent.
Setup & Installation
How quickly can I try my agent?
Enter your website URL and your first agent is typically ready to test in about 90 seconds. It analyzes your highest-value public pages first, while deeper knowledge synthesis can continue in the background. You can review its starter knowledge and try private conversations before installing it on your site.
Is installation difficult?
No. Installation requires pasting one line of JavaScript before the closing </body> tag. Our scanner automatically detects your CMS platform (WordPress, Shopify, Squarespace, Wix, Webflow, Framer, and others) and provides tailored, pre-filled instructions.
Does the widget affect page speed or SEO?
The install script uses defer, downloading without blocking DOM parsing and executing after HTML parsing completes. Minified code, CDN caching, and namespaced styles reduce overhead and conflicts. Test performance and compatibility on your site; search rankings are not guaranteed.
Can I customize the launcher after installation?
Yes. The site script dynamically fetches its settings from Your Fellow Agent. You can change colors, desktop launcher styles (bubble or side tab), position (left or right), tab text, and Chat Toast greetings directly from the dashboard or via MCP without replacing the installed code. On mobile viewports up to 640px wide, the launcher becomes a compact translucent round chat icon and starts closed on page load without losing the conversation history. When opened, the panel adjusts to the visible screen above the keyboard. Use the collapse button in the header to shrink it back to the icon; your conversation is retained. Test keyboard behavior on the devices your customers use.
Lane Assist & Proactive Assistance
What is Lane Assist?
Lane Assist provides proactive hesitation assistance powered by adaptive behavioral intelligence. It acts as an ambient sensor observing reading pace, section dwell, and visible headings on key surfaces (such as pricing, checkout, or service listings) and surfaces an unblocking microhint in the Chat Toast before visitors leave.
How does the Email Form Optimizer (Klaviyo) work?
Keep your Klaviyo script installed, publish your form with "Only show on custom trigger", and enter its ID in Installation settings. YFA can suggest an invitation, but browser-side protection prevents it while visitors are chatting, typing, using controls, or checking out. It waits at least 15 quiet seconds and does not reopen the form after dismissal or submission in the same browser tab session. Custom choosers can supply an explicit activity signal; arbitrary interfaces are not guaranteed to be detected. This is interruption protection, not a proven signup or sales optimizer.
Can Lane Assist be turned off?
Yes. Lane Assist is on by default to maximize customer engagement and prevent visitor drop-off, but it can be toggled off at any time in agent configuration.
What is the Chat Toast greeting?
The Chat Toast is an elevated floating card positioned directly above the closed launcher button. It displays a welcoming greeting note of up to 120 characters without stretching or distorting the launcher button.
Model Context Protocol (MCP) & AI Assistants
Can I configure an agent entirely via Claude or Codex without opening the dashboard?
Yes. Connect your AI coding assistant to https://yourfellowagent.com/mcp. It can call setup_agent to analyze your site and test_agent to try answers. Claim the agent and approve access in the browser before going live. With your permission and website access, your assistant can then retrieve and install the one-line snippet.
Do I need an account or API key to start with MCP?
No. Public tools allow unauthenticated site analysis, draft agent training, and private testing. When you are ready to claim the agent or activate billing, the server returns a secure browser pairing link (/mcp/pair?code=...) to authorize your device.
Can an agency manage multiple client workspaces through MCP?
Yes. Authenticated AI assistants can use list_teams and switch_team to switch between client accounts, adjust brand configurations, update knowledge, and inspect tool executions across multiple client domains.
Team Coordination & Remote Replies
What happens when the agent cannot answer a question?
It exercises judgment instead of hallucinating or guessing. When a custom quote, appointment, availability check, or business exception is needed, the agent asks useful follow-up questions, captures contact details, and notifies your team via email, Slack, or dashboard.
How do owner replies work?
If agent follow-up is enabled, reply to a notification email with your instructions and the agent writes back to the customer. In Slack, mention your agent to give instructions or choose Take over to send your own replies as written. You can return control when finished or contact the customer directly.
What is "No response needed"?
If a customer message does not require a reply, the owner can select "No response needed" in the dashboard or reply with that phrase by email to archive the notification.
Trial, Billing & Offline Fallback
When does the 14-day free trial start?
The 14-day trial begins only when the agent is installed on your approved website and has its first real customer conversation. Creating, configuring, or privately testing the agent in the dashboard does not start or consume trial time. No credit card is required to begin.
What happens after the trial if billing is not activated?
The installed widget stays on your website but switches to contact-form mode. It stops generating full agent answers and owner-reply workflows, but it continues collecting visitor messages and contact details so you never miss a lead. Private testing in the dashboard remains fully functional.
How does pricing work?
Starter is $29/month for 100 customer conversations, Business is $79/month for 500, and Growth is $199/month for 2,000 conversations across up to three agents. Annual billing charges for ten months (two months free). Modest per-conversation overages ($0.40, $0.25, and $0.12) keep your agent online during busy spikes instead of cutting off visitors.
Do you support SMS or WhatsApp? SMS is in limited beta.
Your Fellow Agent works through your website widget, web dashboard, email, and Slack. SMS follow-up is available in a limited beta for selected agents; contact our team about access. Setup and provider approval are required, and texting is not automatically included in every plan. Before texting from a website conversation, the agent asks permission and explains that texts concern the customer's request, not marketing. It saves the explanation and agreement; the consent flow must also meet the sending number's registered campaign requirements. Customers can continue in chat instead. A separate, disabled-by-default integrations pilot adds MCP dedicated-number setup and server-triggered transactional notifications after fee approval, provider activation, customer permission, and server integration; it is not self-service. Incoming photos by text, WhatsApp, and voice are not supported in this beta.
Ready to get started?
Start with your website and build an agent your team can test.
Create your agent