OpenAI-compatible & local

Beyond the built-in providers, crab’d can talk to any OpenAI-compatible endpoint (a self-hosted model, an internal proxy, a niche vendor, or a local runtime) via providers.custom.

Custom provider

Declare the provider; its id becomes usable in model specifiers.

.crabd.yml

providers:
  allowlist: [my-llm]          # a custom id must be allowlisted to be used
  custom:
    - id: my-llm
      base_url: https://llm.internal/v1
      api: openai-completions   # optional; the default
      api_key_env: MY_LLM_KEY   # optional; env var holding the key

model: my-llm/some-model

workflow

- uses: louisescher/crabd@v0
  env:
    MY_LLM_KEY: ${{ secrets.MY_LLM_KEY }}

Field	Description
id	Provider ID used in specifiers (my-llm/model).
base_url	OpenAI-compatible endpoint root.
api	Wire protocol. Default openai-completions.
api_key_env	Env var whose value is the API key.

Caution

A custom provider ID must also be in providers.allowlist. An org can lock providers.allowlist / providers.custom so repos can’t route code to an arbitrary URL.

Local models (Ollama)

Ollama is a custom provider with a shortcut. Set its base URL via env and use ollama/*:

workflow env

CRABD_OLLAMA_BASE_URL: http://localhost:11434/v1

.crabd.yml

model: ollama/llama3.1:8b
providers:
  allowlist: [ollama]

Egress gateway

To route built-in providers through a gateway or proxy (central logging, caching, data-egress control) without changing model specifiers, set providers.gateway_url:

.crabd.yml

providers:
  allowlist: [anthropic, openai]
  gateway_url: https://gateway.example.com

Each allowlisted built-in provider is then reached at ${gateway_url}/<provider> (e.g. https://gateway.example.com/anthropic), keeping its normal credentials. Custom providers and ollama are untouched.