Problem
When using CloudCLI with Claude Code configured for a custom Anthropic-compatible gateway, the Claude provider model list in CloudCLI still returns only the hardcoded fallback Claude models.
This means custom gateway models cannot be selected from /models, even though Claude Code itself can be configured to use the gateway via environment variables.
Environment
- CloudCLI: 1.34.0
- Claude Code: 2.1.176
- Self-hosted CloudCLI
- Claude provider authenticated via API token / custom gateway
- Gateway exposes
/v1/models and /v1/messages
Claude Code environment is configured with variables like:
ANTHROPIC_BASE_URL=http://127.0.0.1:8080
ANTHROPIC_AUTH_TOKEN=...
CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1
The gateway /v1/models returns custom model IDs such as:
gpt-5.3-codex
gpt-5.4-mini
codex-auto-review
Current Behavior
CloudCLI's endpoint:
GET /api/providers/claude/models
returns only the static Claude fallback models, for example:
default
fable
sonnet
sonnet[1m]
opus[1m]
haiku
As a result, /models in the CloudCLI chat UI cannot select gateway-provided custom model IDs.
Expected Behavior
CloudCLI should provide a supported way to include gateway/custom model IDs in the Claude provider model list.
Possible approaches:
- Re-enable safe dynamic model discovery when Claude Code gateway model discovery is enabled, without creating unwanted side-session/jsonl artifacts.
- Add explicit environment-variable configuration, for example:
CLOUDCLI_CLAUDE_MODELS=gpt-5.3-codex,gpt-5.4-mini,codex-auto-review
CLOUDCLI_CLAUDE_DEFAULT_MODEL=gpt-5.3-codex
- Add a settings UI field for custom Claude provider model IDs.
Related Context
This appears related to PR #806, which intentionally disabled runtime supportedModels() discovery and always serves a static fallback list.
It is also adjacent to #234 and #132, but those discuss proxy/custom endpoint integration more broadly. This issue is specifically about the Claude provider /models list not exposing gateway/custom model IDs.
Workaround
I was able to work around this locally by patching dist-server/server/modules/providers/list/claude/claude-models.provider.js to append the custom gateway model IDs to CLAUDE_FALLBACK_MODELS.OPTIONS, but that is fragile and will be overwritten on upgrade.
Problem
When using CloudCLI with Claude Code configured for a custom Anthropic-compatible gateway, the Claude provider model list in CloudCLI still returns only the hardcoded fallback Claude models.
This means custom gateway models cannot be selected from
/models, even though Claude Code itself can be configured to use the gateway via environment variables.Environment
/v1/modelsand/v1/messagesClaude Code environment is configured with variables like:
The gateway
/v1/modelsreturns custom model IDs such as:Current Behavior
CloudCLI's endpoint:
returns only the static Claude fallback models, for example:
As a result,
/modelsin the CloudCLI chat UI cannot select gateway-provided custom model IDs.Expected Behavior
CloudCLI should provide a supported way to include gateway/custom model IDs in the Claude provider model list.
Possible approaches:
Related Context
This appears related to PR #806, which intentionally disabled runtime
supportedModels()discovery and always serves a static fallback list.It is also adjacent to #234 and #132, but those discuss proxy/custom endpoint integration more broadly. This issue is specifically about the Claude provider
/modelslist not exposing gateway/custom model IDs.Workaround
I was able to work around this locally by patching
dist-server/server/modules/providers/list/claude/claude-models.provider.jsto append the custom gateway model IDs toCLAUDE_FALLBACK_MODELS.OPTIONS, but that is fragile and will be overwritten on upgrade.