AxonHub Tutorial: Deploy an AI Gateway with Docker in 5 Minutes

AxonHub promises "any SDK, any model, zero code changes" — but does it deliver? I deployed it with Docker in under five minutes and pointed an existing OpenAI-SDK app at it. Here's exactly how it went, step by step.

Step 1: Deploy with Docker Compose

The official image is looplj/axonhub:latest. The compose file is refreshingly small:

services:
  axonhub:
    image: looplj/axonhub:latest
    restart: unless-stopped
    ports:
      - "8080:8080"
    volumes:
      - ./data/axonhub:/data

Run docker compose up -d, then open http://localhost:8080. The first-run wizard asks you to create an admin account — that's it. The whole process took about three minutes on a 2 vCPU / 4GB VM.

Step 2: Add Your First Channel

Head to Channels in the admin UI and add a provider — OpenAI, Anthropic, DeepSeek, OpenRouter, whatever you use. Paste your API key. AxonHub automatically discovers the models available on that channel, so there's no model list to maintain by hand.

Step 3: Point Your Existing SDK at AxonHub

Here's the magic. Your existing OpenAI-SDK code needs one change: the base URL.

client = OpenAI(
    base_url="http://localhost:8080/v1",  # AxonHub
    api_key="sk-axonhub-key"
)

That's it. The same code now calls any model you configured — Claude, Gemini, GLM, Kimi — without touching a single function call. AxonHub handles the protocol translation between OpenAI, Anthropic, and Gemini formats.

Step 4: Watch Failover and Tracing Work

Add a second channel for the same model and AxonHub load-balances between them. Kill one provider key and requests fail over in under 100ms — my test app didn't skip a beat. The Trace view shows a full timeline for every request: latency, token counts, cost, and which channel served it.

Step 5: Control Costs

The dashboard breaks down spend per user, per channel, and per model — including separate input, output, and cache token pricing. Set per-user quotas via the RBAC system and let AxonHub enforce them.

Verdict

If your team juggles multiple LLM providers — or just wants an escape hatch from vendor lock-in — AxonHub is one of the most impressive self-hosted gateways right now, with 192K+ Docker pulls to back it up.