Get your team ready.

• ✓Windows 10 or 11 machine — Always-on so your team can reach the AI whenever
• ✓~4 GB free disk + ~8 GB RAM — Enough for the AI engine and your conversation history
• ✓Docker Desktop installed — We'll prompt you if it's missing
• ✓A Telegram bot token — 5 min via @BotFather — see Step 1
• ✓Every team member's Telegram numeric ID — 2 min per person via @userinfobot — see Step 2
• ✓Your AI engine choice — See Step 3 — most pick Anthropic Claude for v1

Create your Telegram bot

One bot for your whole team. You only do this once — anyone you allowlist (Step 2) can use the same bot.

1. Open Telegram on your phone or desktop.
2. Search for @BotFather, open the chat, send /start.
3. Send /newbot.
4. Reply with a bot name — display name, can be anything (e.g. "Acme Internal AI").
5. Reply with a bot username — must end in bot (e.g. acme_internal_ai_bot). This is how your team will find it.
6. BotFather replies with a token that looks like 8123456789:AAH-XXXxxxx.
7. Copy that token somewhere safe — you'll paste it during setup.

Get YOUR Telegram user ID (just the admin)

The installer only needs ONE ID at install time — yours, as the install admin. Your team gets added afterwards via the admin UI (web form, bulk-paste, labels) — much faster than terminal prompts.

On your own phone or desktop:

1. Open Telegram.
2. Search for @userinfobot, open the chat, send /start.
3. The bot auto-replies with your numeric ID (looks like 1234567890).
4. Keep that number handy — you'll paste it into the installer.

Pick your AI engine

Two paths — both keep your conversations on your machine. Pick by hardware and privacy posture, not by "which is better" (they're both good for different jobs):

• →Local Ollama (privacy-first default — v1.1): the AI runs entirely on your machine. Nothing leaves your office. The more powerful the host machine (more RAM, faster CPU, ideally a GPU), the larger the local model you can run and the better the answers. A modern office desktop runs a small model perfectly well for drafting and summarising.
• →Cloud (Anthropic Claude — v1.0): the fallback when your hardware can't run a large enough local model and you need frontier-grade reasoning. Anthropic contractually does not train on API traffic. Your conversation history still lives on your machine; only the per-message inference call goes out.

Onboard your team via the admin UI

The install drops a desktop shortcut named AI In Your Pocket Admin. Double-click it — or open http://localhost:5680 in any browser on the office machine.

1. In the Allowlist panel, paste each team member's numeric Telegram ID into the textarea (one per line, or comma-separated). Optional shared label like “Sales team” or “Partner” saves time. Hit Add user(s).
2. Tell each new user the bot's username (e.g. @acme_internal_ai_bot). They open Telegram, search for it, send /start once. From then on they can DM the bot freely.
3. Want the AI to know your client / project files? Use the Shared client folders panel to attach a folder by path (e.g. D:\Matters\Conveyancing). The bot can read its files when team members ask — “where are we with Mr Khan's house sale?” gets a real answer citing source files. Read-only — nothing on disk is modified.
4. Need to remove a leaver, swap AI models, or wipe a user's memory for a GDPR request? All from the same web page. Changes take effect within ~1 second — no terminal, no restart.

Each person's conversation memory is private to them. The bot remembers your conversations separately from your colleague's. Full walkthrough at /admin-ui.