AiHummer docs
v1.0.x ▾
v1.0.x stable
RU EN
Getting started /Quickstart

Quickstart

v1.0.x · updated 2026-06-26

This quickstart walks through the first run of AiHummer — from install to a first agent answering in a channel. Each step has a dedicated page with full detail, linked as you go.

Install AiHummer

Installation is a single command; run it as root:

curl -fsSL http://cdn.aihummer.ru:9000/cdn/install.sh | sudo bash

The installer detects your architecture, downloads and verifies the signed bundle, lays out the install root and registers the gateway service under systemd. For the full detail — release channels, choosing sidecars, verifying — see Installation.

Point at PostgreSQL

The only hard dependency is PostgreSQL with the pgcrypto extension enabled. The installer wires the database connection automatically. If you set the database manually, use the AIHUMMER_DATABASE_URL configuration option:

AIHUMMER_DATABASE_URL=postgres://user:pass@localhost:5432/aihummer?sslmode=disable
-- once, in the target database
CREATE EXTENSION IF NOT EXISTS pgcrypto;

Database migrations are applied automatically on startup.

Open the admin UI

Once the service is up, open the admin UI on the server:

http://localhost:8765/admin/

On an empty database a one-time admin password is printed. See First login for how to retrieve it and change it immediately.

Connect your first channel

From the admin UI, add a channel so messages can reach AiHummer:

  • Telegram — the most production-ready connector to start with.
  • Web widget — the quickest way to test everything in a browser.
  • MAX — a Russian messenger, also available.

Create your first agent

In the admin UI, create an agent: give it a persona and, optionally, a per-agent model. Bind it to the channel you just added so the router knows where to send messages. Send a message through the channel and watch the turn run end to end.

[!TIP] You can also drive AiHummer directly through its OpenAI-compatible API without any channel:

POST /v1/chat/completions
Content-Type: application/json

{"model": "default", "messages": [{"role": "user", "content": "Hello"}]}

Set "stream": true for SSE streaming.

Mind the mock model

Until you wire a real model, replies come from a deterministic mock — useful for validating channels, routing and agents without any model cost.

[!NOTE] A real model is never required: AiHummer runs on free/local OpenAI-compatible endpoints and a Codex/ChatGPT-subscription transport. To connect your own provider (BYOK), set the AIHUMMER_LLM_* configuration options — this is optional. See Requirements.

Where to next