Channel · Telegram

Telegram for customers and agents

OrcaLinq runs Telegram as a customer channel and as an agent endpoint. Bring your bot. Reply from your phone. The Pod travels with the conversation.

Two roles, one channel

Telegram as a customer channel and an agent endpoint

Telegram fills two slots in the semantic PBX: it's where some customers reach you, and it's where many agents prefer to reply from. The platform treats both first-class.

Customer channel

Inbound from your Telegram Bot. AI-first answering, semantic routing via Sonar and Matriarch, Wake audit. Identical to other channels in the routing layer.

Agent endpoint

Each agent picks Telegram as their Den. A private thread becomes their workspace. Replies route back to the customer on whatever channel started the conversation.

Webhook or long-poll

Production runs on webhook with HMAC signature verification. Long-poll is supported for development and constrained networks.

BYOK bot tokens

Bot token, webhook secret, command grammar — all stored encrypted, scoped to tenant, never logged, never returned via API.

Setup

Three steps from a fresh bot to a live channel

  • 1. Create a bot with @BotFather. Capture the bot token and a memorable bot username.
  • 2. Add the bot token and webhook secret in the dashboard. Stored encrypted, never returned, used only at request-decryption time.
  • 3. Configure the webhook URL. The platform handles inbound delivery and verification; you confirm test messages route correctly via the Podium.
  • 4. (Optional) Configure the command grammar. Slash commands like /start, /help, /agent can map to specific routing policies.
  • 5. (Optional) Add an agent-facing bot if your team prefers that to a shared customer-facing bot.
Limitations and tradeoffs

What Telegram does well, what it doesn't

Strengths: excellent media handling, large file support (up to 2 GB), inline keyboards for clean quick-replies, no template approval bureaucracy, free at any volume, no per-message cost.

Tradeoffs: users must initiate the conversation (you cannot DM a stranger), no rich formatting beyond Markdown / HTML basics, no native typing-indicator API for outbound, smaller user base in Western markets compared to WhatsApp.

Best fit: markets with strong Telegram adoption (CIS, parts of EU, India), tech-savvy customer bases, communities and education, agent-side use (a Telegram-as-Den agent endpoint works well almost anywhere).

Not ideal for: mass-market US-only consumer support (where SMS or iMessage dominate), regulated industries where Telegram's encryption posture is a procurement issue.

Security and credentials

How tokens, webhooks, and media are handled

  • Bot tokens encrypted at rest. Per-tenant key wrap; plaintext never leaves the request handler.
  • Webhook secret verified on every inbound. Telegram's X-Telegram-Bot-Api-Secret-Token header is mandatory.
  • Tenant scoping at the data plane. Tenant-scoped state boundaries ensure no cross-tenant query path exists.
  • Media stored in tenant-scoped buckets. Signed URLs with short TTLs. No public file links.
  • Rotation without downtime. Update the token in the dashboard; old token is revoked at the same instant the new one becomes active.
  • No token logging. Audit logs reference token IDs, never plaintext tokens.
Example conversation

What a typical Telegram-first interaction looks like

Customer (Telegram bot, 14:23): "/start" → "Hi, looking for the EU shipping address for returns?"

AI: "Returns within the EU go to: [warehouse address]. Please include your order number on the parcel. Want me to email you a return label?"

Customer: "Yes please, order #SF-90211."

AI (live tool call to OMS): "Got it — order found, return label sent to the email on file. You'll see it in a minute. Want me to track the return once you ship it?"

Customer: "Yes, and can I get a refund on shipping too? It arrived damaged."

AI: "Damaged-on-arrival is something our team handles directly. I'm paging an agent now — could you send a photo of the damage while we wait?"

Agent (replies from her Telegram Den, 14:25): "Hi, Tanya here from support. Sorry about the damage — I'll process the shipping refund as soon as I see the photo."

Telegram FAQ

Frequently asked questions

Can Telegram be used as an agent endpoint?

Yes. Telegram is one of the most popular agent endpoints because agents already have it on their phones. A private thread becomes the agent's workspace; replies route back to the customer on the channel they originally used. The Pod travels with the conversation.

Webhook or long-poll?

Both are supported. Webhook is the production default with HMAC signature verification on every inbound. Long-poll is useful for local development, constrained environments, or networks where inbound HTTP is restricted.

Can I use multiple bots?

Yes. Each bot is BYOK with its own token and webhook secret. Tenants can run separate customer-facing and agent-facing bots, or split bots by language, region, or product line. Routing policy maps each bot to a different department.

What about inline keyboards and quick replies?

The platform supports inline keyboards for quick-reply suggestions, file handovers (image, document, audio, video), and command routing for agent workflows. The bot's command grammar is configurable per tenant.

What about media — images, documents, voice notes?

Full media support. Media is fetched via the Telegram Bot API, stored in tenant-scoped storage with signed URLs, and surfaced to the agent endpoint. Voice notes can be transcribed (when the speech-to-text tool is enabled).

Is the bot token ever exposed?

Never. Tokens are stored encrypted at rest, decrypted only in-memory at request time, and never logged or returned via API. BYOK rotation is supported without downtime.

What happens if Telegram rate-limits the bot?

The platform respects Telegram's per-chat and global rate limits, queues messages when needed, and surfaces backpressure to the agent UI so no one tries to send a 30th message in 60 seconds.

Wire a Telegram bot during the demo.

Bring a sandbox bot or create one with @BotFather on the call. We'll connect it live and route a message end-to-end through Sonar, Matriarch, and Wake.