Let an agent ask before it acts
Publish the question with inline buttons. A tap comes back as an app.command event, with who pressed it, on your app's topic. The agent reads it over SSE or a webhook and carries on.
Publish once. Reach every chat, webhook, and phone.
hooka is a self-hosted, Telegram-first notification bus for people and agents. Publish to a topic, and its routes deliver to Telegram chats and forum threads, webhooks, and SMS. Every message is stored, along with the outcome of each delivery.
Open source under Apache-2.0. The repository goes public soon.
POST /publish/deploys
{"title":"Release",
deploys
stored as message 4711
-
Telegram forum thread sentOps / deploysRelease api@a1b2c3 is live on prod 14:02
-
Webhook ci.example.com/hooks/deploy sent
-
SMS +1 ••• ••• 0142 sent
"delivered": 3, "failed": 0
Publish in one line
A publish is one HTTP request with a bearer token. The response tells you where the message went, and the message stays in history with each route's outcome.
- Add an
Idempotency-Keyheader and a retried request returns the original message instead of sending twice. - Mint a scoped
hk_token that can only publish todeploys*, and keep the master key to yourself. - Failed deliveries retry with backoff. What still fails lands in dead letters you can replay.
Request
curl https://api.tryhooka.xyz/publish/deploys \
-H "Authorization: Bearer hk_your_token" \
-H "Content-Type: application/json" \
-d '{"title":"Release","text":"api@a1b2c3 is live on prod"}'
Response
{
"ok": true,
"message": "published",
"message_id": 4711,
"topic": "deploys",
"delivered": 3,
"failed": 0,
"idempotency_hit": false,
"destinations": [
{
"provider": "telegram",
"chat_id": "-1001234567890",
"message_thread_id": "42",
"status": "sent"
},
{ "provider": "webhook", "status": "sent" },
{ "provider": "sms", "status": "sent" }
],
"store_only": false
}Placeholder token. Use your own instance URL when you self-host.
The whole delivery path in one binary
The server is a single Go binary backed by Postgres. The MCP server, agent bridge, and dashboard are optional companions that talk to it over HTTP.
Deliver
- Topics and routes
- Publish once to a topic. Its routes deliver to Telegram chats or forum threads, webhooks with custom headers, and SMS through Twilio or Telnyx.
- Safe retries
- Idempotency keys make retries safe. Failed deliveries back off and try again, and terminal failures become dead letters you can replay.
- Telegram, fully
- HTML and Markdown, rich messages, inline keyboards, documents, photos, albums, and voice notes, posted into the forum thread you choose.
- Push to your own iOS app
- Optional APNs support lets an iOS app you build and sign register devices and receive pushes.
Read it back
- History and search
- Every message is kept with its per-route outcome. Filter a topic's history, or search across all topics with full-text search.
- Live streams
- Subscribe to any topic over Server-Sent Events, filtered by priority or tag, and resume from where you left off.
- Chats that listen
- Send
/capture deploysin a Telegram chat or thread, and what people type there is stored on that topic too.
Build on it
- Commands become events
- Register an app and its bot commands. Each command, or tap on one of its buttons, arrives as an
app.commandevent on a topic, and the app answers by publishing. - Capability URLs
- Give a third party a publish-only
/in/{token}URL: one topic, rate-limited, revocable. GitHub webhooks sent to it arrive rendered, and Alertmanager has its own endpoint. - Reminders and forms
- One-shot and cron reminders go through the same delivery path, retries included. Native forms give you a public link and stored responses.
- MCP server and agent bridge
hooka-mcpgives MCP-capable agents publish, search, reminders, and forms. The agent bridge connects a topic to OpenCode or Pi.
Run it
- Access control
- A master key for you, scoped
hk_tokens per topic pattern for everything else, and anaudittopic that reports token, route, and login changes. - Operations built in
- Prometheus metrics, health and readiness checks, and retention cleanup you configure per kind of record.
- Optional dashboard
- A Next.js control panel for topics, routes, tokens, reminders, forms, and the audit log.
What to wire it to
Four common jobs, all handled by the server and its own clients.
Post CI and deploy results
Publish from any job with a scoped token. The idempotency key makes a re-run return the original message instead of posting twice. For GitHub itself, register a capability URL as the webhook, and pushes, PRs, reviews, releases, and CI failures arrive rendered.
Route alerts by severity
Point Alertmanager at /alertmanager/{topic}. Severity sets the priority, watchdog alerts arrive silently, and a topic for critical alerts can post to its own forum thread.
Collect answers with a form
An agent can create a form through MCP or the CLI and share one public link. Respondents need no account, and you read responses in the dashboard, as CSV, or over the API. The waitlist on this page is a hooka form.
Use the hosted beta, or run your own
Join the waitlist for the hosted beta, or run the open-source server yourself.
Waitlist open
Hosted beta
We run hooka for you, so there is no server or database to look after. The hosted tier isn't open yet. Join the waitlist and we'll tell you when it is.
Join the waitlist
The waitlist is itself a hooka form. It asks for your email and, if you like, what you'd use hooka for.
Apache-2.0
Self-host
One Go binary and Postgres on your own infrastructure. Telegram updates arrive by long-polling, so you don't need a public URL to start.
One click on Railway Soon
A Railway template that starts hooka, Postgres, and the dashboard from your bot token. It builds from the source, so it opens when the repository does.
Docker Compose Soon
A compose file for hooka and Postgres, with the dashboard behind a profile. The clone-and-run steps appear here when the source goes public.