[aiapienthusiast]

Summary

• Add Cheaper Inference as an open_code gateway provider, next to Infron.
• It mirrors the Infron path (feat: add Infron as an open_code gateway provider #126) in the OpenCode config, the manifests, and the Python and Go resolution chains.
  Closes Add Cheaper Inference as an open_code gateway provider #164.

Cheaper Inference is an OpenAI-compatible LLM gateway.
One API key gives access to models from several labs.

Its model ids are bare (one level), so a role ref is the provider prefix plus a bare id:

export CHEAPERINFERENCE_API_KEY="ci_live_..."
SWE_DEFAULT_MODEL=cheaperinference/gpt-5.4-mini

With only a CHEAPERINFERENCE_API_KEY set (no Anthropic key, no other gateway key, no SWE_DEFAULT_RUNTIME), the main paths auto-select open_code and default to cheaperinference/gpt-5.4-mini.
A gateway key already configured keeps precedence.

Validation

• make check
• Relevant manual test performed (if needed)

Behavior Impact

• No behavior change
• Backward compatible behavior change
• Breaking change (explain below)

Notes

Every new path is gated on a Cheaper Inference key being the only gateway key present, so existing deployments do not change.

Details

Files:

• opencode.json: a cheaperinference (@ai-sdk/openai-compatible) provider with base URL https://api.cheaperinference.com/v1 and the chat model ids it declares.
• swe_af/execution/schemas.py and go/internal/config/resolve.go: _cheaperinference_only_env / cheaperInferenceOnlyEnv, the auto default constant, and the auto-select wiring.
• .env.example, README.md, agentfield-package.yaml, go/agentfield-package.yaml: document the key.
• docker-compose*.yml: pass CHEAPERINFERENCE_API_KEY to the services.
• tests/test_model_config.py, go/internal/config/config_test.go, tests/test_dockerfile.py: cover auto-select, precedence, overrides, and the provider entry.

Live check: the default model gpt-5.4-mini returned a reply through the cheaperinference config's base URL, and every declared model id is a live text model from GET /v1/models.