What raia CX is

raia CX is the control room for your organization's AI agents. Everything you do as an admin happens inside one organization workspace — you build agents, train them on your content, give them skills, launch them into channels, and then monitor what they do.

Steps

1. Sign in at raia2.com

   You land on the Launch Pad dashboard for your default organization.

2. Check which org you're in

   The org switcher sits at the top-left of the header. If you administer more than one org, everything on screen is scoped to the one selected here.

3. Confirm you have admin rights

   Only Owner and Admin org roles see the full Launch Pad. If your sidebar has no Users, Conversations or Settings group, you're signed in as a regular user.

The Launch Pad layout

The left rail has two levels: a slim icon strip for the top-level areas — Launch Pad and Copilot — and an expanded panel for the section you're in. The gear at the bottom swaps the panel over to organization settings.

Steps

1. Open Agents from the Build group

   Cards show the agent's public name, description and type. The dot on the right of each card is its status.

2. Use the status pills to triage

   All, Public, Private, Inactive and Favorites filter the library instantly — useful once you're past a dozen agents.

3. Switch between card and list view

   The two icons to the right of the page-size selector toggle the layout.

4. Note the Import button

   Agents can be exported from one raia organization and imported into another via a token — that's how you promote an agent from a sandbox org to production.

Set your brand

Brand controls what your people see on the login screen, on invitation and system emails, and in the chat widget footer. Set it once, early — it's the difference between an internal rollout that looks official and one that looks like a side project.

Steps

1. Go to Settings › Brand
2. Upload a square logo

   It appears on sign-in, sign-up and system emails.

3. Set the Brand Name

   Used in email copy and page titles.

4. Set Terms and Services URL and Knowledge Base URL

   Both are required. The knowledge base link is what your agents' welcome screens point users to for self-service.

5. Choose a URL Slug

   This gives you branded auth URLs — raia2.com/auth/signin/<slug> renders your logo and name instead of raia's.

Authentication and default limits

Two decisions live here, and both are org-wide: how your people prove who they are, and how much any single agent or user is allowed to spend before raia stops them.

Steps

1. Go to Settings › Settings
2. Pick your Enabled Authentication Methods

   Authenticator (TOTP) and Phone Number can both be on. Whatever you enable here is what users can choose during their own two-factor setup.

3. Pick the Default Authentication Method

   The one new users are pushed toward. Phone Number is the lower-friction default; TOTP is the stronger one.

4. Set Agent defaults under Default Limits

   Monthly ceilings for tokens, SMS, emails and call minutes that apply to every agent that doesn't have its own custom limits.

5. Set User defaults

   The same four ceilings, applied per end user — your main defense against one runaway conversation burning the month's budget.

Invite your team

“Admins” is the org membership list — the humans who work in raia, as opposed to the end users who talk to your agents. Get the role model right here and the rest of your governance follows.

Steps

1. Go to Settings › Admins
2. Press Invite User

   Enter the email and pick an access mode (role). raia emails them an invitation.

3. Track the invite under Invited Users

   That tab shows email, role, who invited them and when — and lets you revoke a pending invite.

4. Edit a member with the pencil icon

   You can change their role and flip status between Active and Inactive. Deactivating is the reversible move; deleting removes them from the org.

Create an agent

Creation is deliberately thin — one screen, four decisions. Everything that makes the agent good happens afterward, on the Train and Skills tabs.

Steps

1. From the dashboard or agent library, press Create an Agent
2. Choose an Agent type

   Twelve types: Analyst, Support, Sales, Creative, Assistant, Knowledge, Task, Code, Research, Customer Success, Workflow and Marketing. The type seeds the agent's default posture and shows up as a filter across the product.

3. Enter a Public name

   What end users see in chat — e.g. “Amy”. Keep it human.

4. Enter an Internal name

   Must be unique across the org. This is what you and your team search for — e.g. “Amy Sales Agent”.

5. Write a Description

   One or two sentences on the job this agent does. Optimize with AI will tighten it for you.

6. Press Create agent

   Takes a few seconds. raia provisions the agent and attaches three skills automatically: Live Chat, Scoring and Copilot.

Identity and the Info tab

The Info tab is the agent's identity card, and the live preview on the right shows exactly what a user will see. It also tells you which API key the agent was created against.

Steps

1. Open an agent from the library, stay on Info
2. Check Used API keys for agent creation

   A green dot next to OpenAI API Key means the agent is bound to a working org key. If it's red or missing, the agent has no model credentials and won't answer.

3. Adjust Agent role, Public name, Internal name and Description

   All editable after creation.

4. Add Tags

   Free-text labels — Sales, Support, Outbound. They power search and filtering in the library.

5. Upload a background image or generate one with AI

   This is the card art in the library and the header of the chat widget.

Train: instructions and packs

This is the agent's system prompt, and it is versioned. Every save can become a numbered version you can roll back to — which means you can experiment on a live agent without fear.

Steps

1. Open the agent's Train tab, sub-tab Agent Instructions
2. Write the instructions in Current Instructions

   Markdown works. A structure that holds up: Identity & Role → What you know → How to answer → What you must never do → When to escalate.

3. Use Optimize with AI to tighten a draft

   It rewrites for clarity, not for content — review before saving.

4. Press Save for a quick edit, Save version to snapshot

   The Instructions Version dropdown at the top lists every version with its author and timestamp, so you can compare and revert.

5. Attach shared Packs from the right rail

   Add Packs pulls in a pack's content; Link Packs keeps it live so future edits to the pack flow through to every agent linked to it.

Train: documents and knowledge

Documents are the agent's private knowledge base. raia uploads them to a vector store and the agent retrieves from it before answering — this is what stops it inventing your policies.

Steps

1. Open Train › Documents
2. Press Upload Document

   Markdown, text, PDF, JSON and similar. The status column shows processing progress, and “Uploaded to OpenAI at …” once it's live in the vector store.

3. Use the per-row toggle to include or exclude a file

   Toggling off removes a document from retrieval without deleting it — the clean way to test whether a file is helping or confusing the agent.

4. Search and page through the list

   Large agents run to hundreds of documents; the search box matches on document name.

5. Download All to snapshot the knowledge base

   Useful before a big content refresh.

Advanced settings: model and retrieval

Where you pick the brain. Model choice, reasoning depth and retrieval behaviour all live on one screen, and they are the biggest levers on both quality and cost.

Steps

1. Open Train › Advanced Settings
2. Choose a GPT Model

   Models marked (Reasoning) think before answering — better for analysis, slower and dearer for chat.

3. Set Reasoning Effort and Reasoning Mode

   Low/Medium/High effort, Standard or Pro mode. Start Medium/Standard and only raise it if answers are shallow.

4. Leave Use File Search on for any agent with documents

   File Search Max Num Results controls how many chunks are retrieved per answer. More context, more tokens.

5. Enable Code Interpreter only if the agent must compute

   And JSON Schema Output only when a downstream system consumes the reply — it forces every response into your schema.

6. Write a Default Error Message

   What the user sees when a model call fails. Make it human.

7. Leave Vector Store alone unless you're sharing knowledge

   Overriding the vector store ID points this agent at another agent's knowledge base.

The Skills catalog

Skills are how an agent reaches the world. Attached skills sit on the left; the right rail, Skills to Add, shows what's still available. A skill that says “Click SET UP” is attached but not yet configured — and does nothing until you configure it.

Steps

1. Open the agent's Skills tab
2. Add a skill with the + button in Skills to Add
3. Configure it with the pencil icon, or the SET UP link in its description

   Unconfigured skills carry a SET UP badge.

4. Archive (box icon) to pause, delete (bin) to remove

   Archiving keeps the configuration; deleting discards it.

Reference

Configure Live Chat

The most-used skill, and the one your customers actually see. Five tabs; the live preview on the right updates as you type.

Steps

1. Open the agent's Skills tab and edit Live Chat
2. Write the two welcome lines and the Disclaimer

   The disclaimer supports Markdown and is where your “this is an AI” notice belongs.

3. Set the Main Button Label and Subtext

   “Ask a question” / “Do you need product support?” is the pattern — verb first, then the promise.

4. Add Links to your docs or knowledge base

   They appear on the welcome screen as escape hatches for people who'd rather read.

5. Add up to four Predefined options under Predefined Conversation Elements

   These set expectations about what the agent is good at. Use real top-intent questions.

6. On Security, restrict allowed origins to your own domains

   Otherwise anyone can embed your agent on their site and spend your tokens.

Functions and actions

Functions let an agent do things, not just say things. Each one is a named webhook with a JSON parameter schema — the agent decides when to call it and fills in the arguments.

Steps

1. Build it once in BUILD › Functions, or per-agent in the Function skill
2. Give it a Function Name and Internal Name

   The name is what the model reasons about — make it descriptive: “Get Customer Details”, not “fn3”.

3. Write Function Instructions

   Tell the model when to use this and when not to. Optimize with AI helps here.

4. Set the Webhook to Call

   Your endpoint. Add Header Name/Value pairs for authentication — e.g. Authorization: Bearer ….

5. Define Function Parameters as JSON

   The arguments the model must supply. This is the contract; be strict about required fields.

6. Attach it to an agent via the Function skill and test in the sandbox

Launch and deploy your agent

The Launch tab is the deployment console: every channel the agent is reachable on, its public status, a QR code and a share link, and the sandbox to test in before any of it goes live.

Steps

1. Open the agent's Launch tab
2. Test first — press Test Agent

   Opens a sandbox conversation. Run your top five real questions through it before touching status.

3. Set Agent Status

   Public — anyone with the link or QR can chat. Private — only people you've granted access. Inactive — off.

4. Press View Embed Code for Live Chat

   One script tag with the agent's API key. Paste it before </body> on your site and the floating widget appears.

5. Check the channel rows

   Email shows the agent's address, SMS and Voice show its number, Webhook and Function link to their configuration.

6. Share the QR or Copy public link

   Good for events, printed material and internal announcements.

7. Use Export Agent to promote between orgs

   Generates a token another raia organization can import.

Read the dashboard

Your daily standing check. Four KPI cards with 7-day deltas and sparklines, an activity chart, per-agent performance, system log health and Copilot usage.

Steps

1. Scan the four KPI cards and their delta badges
2. Switch the activity chart to Tokens once a week

   Token spikes that don't track message spikes mean an agent's context has grown — usually too many retrieved documents.

3. Look for high-volume, low-score agents in Agent performance

   That combination is where retraining pays back fastest.

4. Click into Errors if the count is non-zero

Conversations and scoring

The archive of everything your agents have said, scored 0–10 and summarized automatically by the Scoring skill. This is your quality function.

Steps

1. Open MANAGE › Conversations
2. Set the date range

   Defaults to the last month. The range picker is top-right.

3. Open Filters

   Source (Live Chat, Voice, Email, SMS, Slack, API…), Status, Agent, State, Favorite, Read, Assignee and a Score range.

4. Sort by Score ascending and read the worst ten

   This single habit, once a week, is the highest-value thing an admin does in raia.

5. Read the auto Summary column before opening anything

   It tells you what the user wanted and what the agent did, in two lines.

6. Click the message count to open the full transcript

   It opens in the Copilot view, where you can read every turn, see which model answered, and check the sources it cited.

7. Export with the download icon on a row

End users

Every person or phone number that has talked to any of your agents, with their conversation count. Distinct from Admins — these are your customers, not your staff.

Steps

1. Open MANAGE › Users
2. Search by name, email or ID
3. Sort by Created at to see new arrivals, or by conversations to find your heaviest users
4. Click a user ID to see their history and set per-user limits

Human in the loop with Copilot

Copilot is the shared inbox where your team works alongside the agents — reading live conversations, taking over when the agent is out of its depth, and rating answers so they improve.

Steps

1. Open Copilot from the icon rail

   You get an agent picker and the full conversation list for the selected agent.

2. Triage with the tabs and state chips

   All / Archived / Favorites / Unread / Assigned, plus New, Escalated, Open, Complete.

3. Open a conversation to read the full thread

   Each agent reply shows the model used, a Sources link for what it retrieved, and Good / Bad feedback buttons.

4. Take over by typing in the message box

   The footer tells you the mode — “AI is active. Human in the Loop Available.”

5. Rate answers with Good / Bad

   The Feedback skill collects these for retraining.

6. Assign a conversation to a teammate

   Assignment plus the Assignee filter on the Conversations page is how a small team divides an inbox.

Logs and troubleshooting

When something isn't working — a webhook silently failing, an SMS not sending, an API caller getting 401s — this is where the answer is.

Steps

1. Open Settings › Logs
2. Filter by Source

   User Action, SMS, Email, Slack, Webhook, External API, Auditor, Suspicious Behavior, System.

3. Filter by Status

   Success, Warning, Error.

4. Narrow the date range and search the message text
5. Click Debug on a row for the full payload

   Method, endpoint and response — enough to hand to whoever owns the failing system.

Usage, limits and governance

Three screens make up your cost and risk controls: org-wide Resources Usage, per-agent Usage limits, and per-agent Security.

Steps

1. Check Settings › Resources Usage weekly

   Tokens, Conversations, Calls, SMS and Emails against your monthly limits, in card or table view. Force Refresh recalculates on demand.

2. Set per-agent ceilings on the agent's Usage › Limits tab

   Press Set Limits for monthly tokens, SMS, emails and call minutes. When a limit is hit the agent stops performing the action that would exceed it.

3. Control access on the agent's Security tab

   Grant users per-agent roles — Owner, Admin, Editor, User — and invite people directly to a single agent.

4. Turn on Restrict Access to Training Documents to Owner Only where knowledge is sensitive

   Only the agent owner and org owner can then view or download its documents.

5. Attach Data Retention to any agent handling personal data

   Sets how long conversations are stored and whether they're anonymized.

6. Consider Auditor and Message quota on public agents

   Auditor watches for sensitive data and policy breaches; Message quota pauses a contact who exceeds a daily message count.

Slack and Microsoft Teams

Org-level connections that let agents live where your team already talks. Connect the workspace once here, then attach the matching skill to each agent that should respond there.

Steps

1. Open Settings › Integrations
2. Connect Slack

   Authorizes a Slack workspace so agents can answer bot @mentions in mapped channels and DMs. Requires an org admin on both sides.

3. Connect Microsoft

   Authorizes a Microsoft tenant for Teams channels and chats.

4. Attach the Slack or Microsoft Teams skill to the agents that should respond

   The org connection alone does nothing until an agent has the skill.

5. Map the channels

   Only mapped channels route to the agent — deliberately, so an agent can't be dragged into every conversation in the workspace.