How it works

Three wizard steps. You're live in 30 minutes.

We provision a VPS for you. You do the wizard. We never see your Telegram credentials, your messages, or your AI key.

Step 1 · ~5 minutes

Connect your Telegram

The wizard shows a QR code. You scan with your phone's Telegram app. Session established. Done.

Prefer phone + SMS? The wizard falls back gracefully. Either way:

  • Session saved to /var/lib/crgram/session/user.session on your VPS, chmod 0600, owner crgram.
  • Encrypted at rest via LUKS full-disk encryption (VPS-level) plus per-column envelope encryption (session-file level).
  • We never see it. We can't read it. If we tried, we'd see ciphertext.

Each customer registers their own api_id at my.telegram.org during onboarding. We never pool credentials across customers — that's a TOS violation and we won't do it.

app.crgram.com/wizard/telegram

Scan to log in

Open Telegram on your phone → Settings → Devices → Scan QR

— or —

Use phone number instead

Step 2 · ~2 minutes

Paste your AI key

Bring your own Z.AI API key OR Anthropic API key. We validate it against the vendor's "list models" endpoint, then store it encrypted on your VPS.

You pay the vendor directly. We don't proxy AI calls. Your usage is your relationship with the vendor.

  • Default: Z.AI API. GLM-4.7 workhorse ($0.60/$2.20 per M tokens) + GLM-5.2 deep ($1.40/$4.40). Cheapest path.
  • Alternative: Anthropic API. Claude Haiku / Sonnet. 2-3x more expensive but some customers prefer Western vendor.
  • NOT allowed: Z.AI Coding Plan, Claude Max. Both are personal-developer subscriptions. Using them in a commercial product gets your account banned. Read why →
app.crgram.com/wizard/llm

Connect your AI

Get one at z.ai/model-api. Stored encrypted on your VPS, never sent to CRgraM Cloud.


Step 3 · optional, ~30-90 minutes

Backfill from export

Without this step, CRgraM Cloud starts fresh — only new messages from this moment forward. With it, you get the magic: your entire Telegram history as a structured CRM on day one.

  1. Open Telegram Desktop → Settings → Advanced → Export Telegram data
  2. Tick "Personal chats" only. Turn media on (for vision OCR). Format: Machine-readable JSON.
  3. Export. You'll get a result.json + media folder.
  4. Upload via HTTPS in the wizard. File goes to your VPS only.
  5. We ingest 1M+ messages in ~1 hour. AI Tier-1 enrichment runs over 24-48h.

Don't want to backfill? Skip this step. You'll have a clean-slate CRM going forward. Re-run any time.

app.crgram.com/wizard/backfill

Upload your export

Drag in your result.json file from Telegram Desktop export.

Drop result.json here

or click to browse

result.json✓ uploaded

Ingesting 412,341 messages…

After the wizard: your dashboard

From minute one you can browse chats and search. AI enrichment runs in the background and the dashboard populates progressively over 24-48 hours.

Today view

Ranked list of partners needing attention: bring-back-to-life targets, broken promises (yours and theirs), stuck conversations, payments owed. The morning standup you didn't know you needed.

Pipeline board

Partners (not chats) as cards in states: Hot, Active, Stuck, Cold, Bring Back to Life, Dead, Never Started, Fraud. Drag-and-drop. Linked chats and contacts. Open obligations.

Per-chat intelligence

Click any chat → AI summary, detected opportunities with citations to specific message IDs, extracted money signals, suggested next message in chat's tone and language.

Search

Full-text AND semantic over all your messages, AI summaries, vision OCR output, and notes. Find "that conversation about Canada cap with Acme from August" in milliseconds.

Send / organize

AI drafts the next message — you click send (round-trips through your own Telegram session). AI proposes folder layouts — you click apply (round-trips through Telegram folders). Always your click.

Audit log

Every write action your instance takes is logged: timestamp, target, originating UI event. Customer-visible in settings. Nothing happens on your account that you can't see.

Want a walkthrough?

We onboard beta customers personally — a 30-minute call where we provision your VPS, walk through the wizard with you, and answer questions. No automation, no support queue.