Configure Cloud Models

FlowDown supports any OpenAI-compatible HTTPS service using either chat completions or responses.

Create or import a model

1. Open Settings → Model.
2. Tap the ＋ in the top-right corner.
3. Under Cloud Model, choose Empty Model to start from scratch.
4. To reuse saved profiles, select Import Model → Import from File and load an exported .fdmodel or .plist.

Connect your provider

1. Create a blank profile or open an existing one.
2. Enter the full inference URL (for example, https://api.example.com/v1/chat/completions or /v1/responses). FlowDown auto-detects and sets Content Format; switch it manually if detection is wrong.
3. Set the model identifier. Tap the field to Select from Server, which calls the model list endpoint (defaults to $INFERENCE_ENDPOINT$/../../models; adjust if your provider uses a different path).
4. Enter the Authorization token, which is sent as Authorization: Bearer <token>, and add any required custom headers. Custom headers can override Authorization for other authentication schemes.
5. Add JSON in Body Fields. The quick menu inserts reasoning toggles (enable_thinking / reasoning with budgets), sampling parameters, input/output modalities, or provider flags.
6. Toggle capabilities (Tool, Vision, Audio, Developer role), set context length and nickname, then save.

Tip: In the editor, ⋯ lets you Verify model (connectivity), Duplicate, or Export model for version control.

Best practices

• Endpoint & format: keep the inference URL aligned with Content Format (chat completions vs responses) to avoid HTTP errors.
• Model list: configure the model list endpoint and use Select from Server instead of typing IDs.
• Body fields: add provider-specific keys (reasoning budgets, top_p / top_k, modalities, etc.) via Body Fields, ensuring valid JSON.
• Backups: model definitions sync with iCloud and database exports. Before major edits, run Settings → Data → Export Database.

Advanced: Custom / Enterprise Setup

For private deployments or bespoke gateways. Connect only trusted endpoints—misconfigurations can leak data or incur costs.

• Create: Settings → Model → ＋ → Cloud Model → Empty Model. Edit inline or export .fdmodel, tweak externally, then re-import.

• Key fields (unused fields can be empty strings/collections):

  Key	Purpose
  endpoint	Inference URL such as /v1/chat/completions or /v1/responses; must match response_format.
  response_format	chatCompletions or responses, aligned with the endpoint.
  model_identifier	Model name sent to the provider.
  model_list_endpoint	List endpoint (defaults to $INFERENCE_ENDPOINT$/../../models) for Select from Server.
  token / headers	Auth info; custom headers can override the default Authorization: Bearer ....
  body_fields	JSON string merged into the request body—use it for reasoning toggles, budgets, sampling keys, modalities, etc.
  capabilities / context / name / comment	Declare capabilities, context window, display name, and notes to drive UI toggles and trimming.

• Verify & audit: after saving, run ⋯ → Verify model; audit calls in Settings → Support → View Logs. Remove/disable unused configs to avoid accidental calls.