The console is where your team works: supervising the assistant, taking over conversations, and shaping what it knows. Here is every screen and what it is for.
Tephlo has two roles inside a workspace, and the split is deliberately simple. The console hides what a role cannot use; the backend refuses it regardless, so the hiding is a courtesy rather than the security boundary.
AgentThe escalation queue and the conversations in it. Everything an agent needs to take over a chat and reply, and nothing that could change how the assistant behaves.
Workspace adminEverything an agent sees, plus knowledge, channels and settings, the whole Assistant section — behaviour, goals and performance — analytics, the team, plan and usage, and the Privacy Center.
The daily work
Workspace overviewThe first screen after signing in. Message volume, escalations broken down by reason, the mix of conversations handled by the assistant versus your team, and a setup-health panel showing enabled channels and how many knowledge sources are ready, still processing, or failed. Switch between the last 24 hours, 7 days and 30 days.
ConversationsThe full transcript of a single conversation, with the reply composer your agents use to answer over the customer’s own channel.
Escalation queue
The centre of the working day: conversations the assistant handed off for a human. Each row carries the reason it escalated and the channel it came in on.
Filter by assignment — All, Unassigned, or Mine — and further by channel and by how the conversation was escalated.
The list refreshes on its own and shows when it last updated, so nobody works from a stale screen.
Taking a conversation assigns it to you, so two agents do not answer the same customer.
Reasons are explained in human handoff. Reading them as a group is the fastest way to find what your knowledge base is missing.
Setup and configuration
Knowledge baseUpload, watch processing, set validity windows, mark a source verified, and delete. See knowledge.
SettingsThe connective tissue: routing tag, welcome message, directory visibility, your own WhatsApp and Telegram channels, the web chat widget and its allowed origins, order lookups, your physical locations, customer access, and the escalation-alert recipients. See channels.
ShareSandbox share links and QR codes for WhatsApp and Telegram, with your routing command pre-filled. See the sandbox link.
Assistant
Everything about how your assistant behaves and how well it is doing, in three screens. The first two are yours to configure; the third is where you inspect and tune what the platform has learned.
BehaviorBrand voice and behaviour, escalation sensitivity, the hand-off message customers see, how voice notes and attachments are handled, and your business hours. Hours sit beside the hand-off message because they finish it: with hours set, a customer escalated outside them is told when your team is back rather than left waiting in silence.
GoalsWhat your organization is for, what a good outcome looks like, what your team handles off-chat, what it needs collected first, tone, never-say phrases, and when to bring in a person. The same screen lists what your assistant can actually do right now. See conversation goals.
PerformanceWhere you inspect and tune the assistant rather than configure it. Nothing here publishes knowledge automatically, and its tabs are described below. Day-to-day numbers — volume, escalations, resolution rate — are on analytics instead, so “how is it doing?” has one answer rather than two.
Inside Performance
Reuse & memoryThe consent and retention controls: whether de-identified conversations may be used for quality analysis, whether customer memory is on, and how long improvement data is kept.
Approved answers Rolling outReview and approve a canonical answer to a question you get constantly, so it can be served directly rather than regenerated.
Cache Rolling outReuse windows and similarity thresholds, inside platform-enforced safe bounds.
Escalations Rolling outWhy conversations escalate, and a place for your team to record what they actually found when they took over.
Improvements Rolling outRepeat questions mined from real conversations, with a suggested answer for one-click approval.
Evaluations Rolling outDeterministic test cases for the reuse layer, run as a suite before a change goes live. Generation quality is measured separately, in an offline harness — see how any of this is checked.
Most of Performance is the intelligence layer, and it is still rolling out. Each part needs both a platform switch and your workspace’s consent before it does anything, and every part is off by default. The consent and retention controls on the Reuse & memory tab work today. Business hours are not part of any of this — they live on Behavior and work whatever the intelligence layer is doing.
Analytics
A read of the last 7, 30 or 90 days. It is built to answer “what should I fix next?” rather than to look impressive:
Daily activity — messages in, assistant replies, and your team’s replies, per day, with escalations alongside.
When customers write — messages by hour of day across the window, so staffing can follow demand. Times are shown in UTC.
What the assistant handled — the topics customers asked about and how often the assistant closed them out.
Why conversations escalate — the single most useful chart here. A tall no supporting knowledge bar is a content backlog, not an AI problem.
What your team found — how escalated conversations turned out once a person looked at them.
Support tickets — the callbacks, complaints and refund requests the assistant wrote down for you.
Languages served — what your customers actually wrote in, which is often not what anyone assumed.
How answers were reused — where work was saved by reusing an answer already trusted.
One more block appears only once the reuse layer is running and has something to report: what reuse saved you — model calls avoided, the share of answers served from reuse, how many cache entries are currently live, and which reuse action served them. These are savings, not quality: they say how much work was skipped, never how well the assistant answered. When there is no such data the block is absent rather than showing zeroes, so a number on this page always means a measurement.
Plan, team and privacy
TeamInvite colleagues and set each one as an agent or a workspace admin.
Plan & usageWhat your workspace has consumed against its entitlement, daily usage, a breakdown by type, alert preferences, a CSV export, and a plan-change request.
Privacy CenterFulfil a verified customer access or deletion request, with an audit trail. See privacy tooling.
Account & securityYour own profile, password, multi-factor authentication, Telegram recovery, and a button to revoke every session you have anywhere.
The operator console
Tephlo also ships a separate operator console for the people running the platform itself — tenants, shared channels, routing, release state and system health. It is a different application on a different host, and nothing in it is part of your workspace. If you are a customer of Tephlo, the console described on this page is the only one you will use.