Install the agent with the one-liner for your platform, then check it is on your PATH.

# macOS / Linux
curl -fsSL https://omp.sh/install | sh

# Windows (PowerShell)
irm https://omp.sh/install.ps1 | iex

Also available via Homebrew (brew install can1357/tap/omp) or npm (the package is @oh-my-pi/pi-coding-agent, which requires bun).

Create ~/.omp/agent/models.yml (or extend it if it already exists) with one provider block: the gateway endpoint, the OpenAI wire, your key, and the models you want.

providers:
  selora:
    baseUrl: https://api.selora.lol/v1
    api: openai-completions
    apiKey: sk-gw-...
    models:
      - id: claude-sonnet-5
        name: Claude Sonnet 5
        contextWindow: 200000
        maxTokens: 16384
      - id: claude-fable-5-1
        name: Claude Fable 5.1
        contextWindow: 1000000
        maxTokens: 128000
      - id: kimi-k3
        name: Kimi K3
        contextWindow: 200000
        maxTokens: 16384
      - id: gpt-6-astra
        name: GPT-6 Astra
        contextWindow: 200000
        maxTokens: 16384

The contextWindow and maxTokens figures match the gateway defaults for these models; copy the real numbers for any other model off the model catalog. omp also accepts api: openai-responses and anthropic-messages, but openai-completions is the recommended wire here: it carries full tool-calling, streaming, and the thinking output (reasoning_content) these models produce. The key sits in this file — keep its permissions tight (chmod 600 on macOS/Linux).

Check that omp can see the Selora models, set your default model role, then run it from your project.

omp models selora

# set the default role (or use /model inside a session)
omp setup
cd /path/to/project
omp

omp models selora lists every model from the block above — that is the discovery check. If a model is missing, its id in models.yml does not match an id the gateway serves; check GET /v1/models.