ClearFox logoClearFox
Product tour · real screenshots, unedited

From a blank server to an
AI that briefs the CEO

A walkthrough of a fresh ClearFox install: setup, the first integration, the first useful answer, and an autonomous agent delivering a morning briefing to Telegram.

Chapter 01

From license key to a live portal

ClearFox is deployed on your infrastructure during onboarding. Once the Docker stack is up, the whole setup is three screens.

01

Paste your license key and one LLM key

The portal starts with two fields. You bring your own model — a cloud key from Anthropic or OpenAI, or a local endpoint via Ollama, vLLM, or llama.cpp for full air-gap.

  • No vendor lock-in: any provider, switchable later
  • Keys are stored encrypted in your own database
ClearFox setup screen asking for a license key and an LLM provider API key
02

Name the company, brand the assistant

Company name, assistant name, your logo, allowed email domains, and the first admin account. This is everything ClearFox needs to become an internal tool your team recognizes.

  • White-label: your logo and assistant name everywhere
  • Registration restricted to your email domains
Configure Portal screen with company details and admin account
03

Sign in to your own AI portal

The login page carries your branding, not ours. Email + password out of the box; Google or Microsoft SSO can be switched on later from the admin panel.

Branded ClearFox login screen for Acme Corp
04

First chat — and an honest assistant

Nothing is connected yet, and the assistant says exactly that instead of pretending. It still helps with everything a general assistant can — drafts, translations, brainstorming — and suggests connecting your tools when you’re ready.

  • No hallucinated “insights” — the AI is told to say what it can’t do
  • Useful out of the box: drafting, translating, brainstorming
Empty first chat where the assistant explains no integrations are connected yet
05

Pick the brain

Every chat can switch models. The picker shows what each model is good at, the approximate cost per message, and compliance warnings — like a provider that can’t offer zero data retention.

  • Per-message cost estimates before you spend
  • Data-retention warnings surfaced right in the picker
Model selection dropdown with pricing and data retention warnings
Chapter 02

Connect your first tool, ask a real question

Personal integrations are self-serve: each user connects their own accounts and the assistant acts with their permissions. We pick Google Calendar here, but the flow is identical for any of the 46 connectors.

06

Pick any connector from the catalog

The sidebar lists 46 connectors — from Jira and HubSpot to MySQL and Zapier. Google Calendar is just the example of the day. The managed ClearFox proxy gets Google OAuth working in one click for a quick start; for production we recommend registering your own OAuth app — setup guides are built in.

  • 46 connectors, one per tool, each in its own container
  • Managed proxy for a quick start, your own OAuth app recommended for production
New Google Calendar connection screen with the full integrations catalog in the sidebar
07

Let the AI describe the connection

“Fill with AI” inspects the live connection — real calendars, real data — and writes the system prompt itself: what this source is, what it contains, what it should be used for. That context is what makes answers precise later.

  • AI samples real data and documents the source for itself
  • Editable Markdown — you stay in control of the prompt
Fill with AI writing a system prompt describing the Google Calendar connection
08

Ask the first useful question

“Are there any overlapping meetings on my calendar next week?” The assistant calls the calendar tools, finds two conflicts with exact times, and notes the timezone. The same works with any connected tool — deals in the CRM, rows in a database, tickets in Jira.

  • Expandable thought process — see every tool call
  • Real API data, not summaries from training
Chat answer listing two overlapping meetings found in Google Calendar
Chapter 03

Shared tools, connected once

Some tools don’t belong to any single person — a company Telegram bot, the knowledge base, web analytics. The admin connects them once and shares them with the roles that need them; everything personal stays personal.

09

Workspace-level integrations

The admin panel mirrors the personal catalog, but connections made here are shared across the team. It’s the right place for tools where an individual account doesn’t matter: a Telegram bot, Confluence or Notion knowledge bases, Google Analytics, shared databases. For Google, the proxy mode is the quick start — we recommend switching to your own OAuth app as you roll out.

  • Configure once, share per role
  • Proxy, own app, custom proxy, or disabled — per vendor
Workspace-level integrations screen with shared Google OAuth configuration
10

Connect the company Telegram bot

Paste a bot token from @BotFather, hit “Generate description”, choose which roles may use it. Now any chat — and any agent — can message the team on Telegram. Prefer email? Connect Gmail or Microsoft Mail and the assistant sends emails the same way.

  • Role access is part of the configuration
  • Telegram, Slack, or plain email — same pattern for 40+ tools
Telegram bot configuration with AI-generated system prompt and role access
Chapter 04

An agent that works while you sleep

Agents are standing orders that run on a schedule, on demand, or via webhook. They don’t require admin-level integrations: an agent runs with its creator’s own connections — shared, personal, or none at all.

11

Create the “Morning briefing for CEO” agent

An agent is standing orders in plain Markdown: what to gather (calendar, HubSpot deals, Odoo finance, Analytics), how to format it, where to send it. Schedule it for weekday mornings — done. “Improve with AI” rewrites the task for you if prompting isn’t your thing.

  • Cron scheduling with timezone, or trigger via webhook from n8n/Zapier
  • Runs with the creator’s integrations and permissions
  • Keeps its own memory: notes it writes are loaded on the next run
  • Can pause and ask a human when it needs input
New agent form with the Morning briefing task, schedule and model settings
12

08:00 — the briefing lands in Telegram

Next morning the CEO reads one message: urgent items on top, today’s meetings with conflicts flagged, yesterday’s sales vs. last week, cash position, web traffic — and a one-line bottom line. Four systems, zero questions asked.

  • Calendar + HubSpot + Odoo + Analytics in a single message
  • Decision-ready in under two minutes, every workday
Telegram message with the Acme morning briefing: attention items, calendar, sales, finance and bottom line
Chapter 05

The guardrails behind it all

Everything an admin needs to roll ClearFox out to a whole company — access, identity, memory, and models — lives in one panel.

13

Roles decide who sees what

Access is granted per connection, not per tool: marketing can read the Telegram bot and use HubSpot, and nothing else. PII and code visibility are separate switches — the output filter masks customer emails, phones and IBANs for roles that shouldn’t see them.

  • Per-configuration access with read/write levels
  • PII and code/SQL filters per role
  • Custom system prompt appended per role
Role permissions screen with per-integration access and PII filters
14

Sign-in your way

Share an invite link — anyone with an allowed email domain can register. Or plug in Google / Microsoft SSO; the same OAuth app that powers integrations powers sign-in.

Sign-in settings with invite link and SSO provider status
15

AI Memory — supervised

The assistant remembers clarifications and corrections across conversations, scoped per user or globally. Admins see every memory, can edit or delete any of them, and search semantically.

AI Memory admin screen listing stored memories with scopes and topics
16

Models and company context

Provider keys are stored encrypted; default and query-understanding models are set here. The company context — injected into every conversation — can be generated straight from your website.

LLM providers admin screen with API keys, default model and company context

That's ClearFox: an AI workspace for the whole company

Everything you just saw took one morning on a fresh install. No AI expertise, no data leaving your infrastructure — just people asking questions and agents doing the routine.

Secure by design

Self-hosted on your servers, role-based access to every connection, PII filters, audit log — and your own choice of model, cloud or local.

Easy for everyone

No prompt engineering required: the AI documents connections, improves agent tasks, and remembers your corrections. Anyone who can chat can use it.

Practical value, fast

Real answers from your calendar, CRM, finance and analytics on day one — and agents that turn them into routine deliverables for the business.

Stop discovering problems in meetings

Let ClearFox surface risks, explain context, and create actions — every morning, before your first coffee. Any question answered in seconds. Deployed on your servers.