Admin Console › AI Providers

Path: Avatar dropdown → Administration → AI Providers

Tab bar: Provider · Model · Embedding

---

Tab: Provider

Form for creating or editing an AI provider.

Field	Type	Condition	Description
Provider ID	Text	always	Unique internal identifier (e.g. openai). Cannot be changed after creation.
Display Name	Text	always	Human-readable name shown in dropdowns.
Provider Type	Dropdown	always	HTTP API — connection via REST. CLI — uses a local command-line program.
Base URL	Text	HTTP only	Base URL of the API (e.g. https://api.openai.com).
Test Path	Text	HTTP only	Path for the connection test (e.g. /v1/models).
Test Method	Dropdown	HTTP only	GET or POST.
Authentication	Dropdown	HTTP only	Bearer Token, Custom Header, Query Parameter, or None.
API Key / Secret	Text	HTTP only	API key or reference to a server-side secret (secret:KEY_NAME).
Header Name	Text	Auth = Custom Header	Name of the HTTP header (e.g. X-API-Key).
Parameter Name	Text	Auth = Query Parameter	Name of the URL query parameter (e.g. api_key).
Test Model ID	Text	HTTP only	Model ID for POST connection tests (e.g. gpt-5.6-terra).
OpenAI-compatible API	Checkbox	HTTP only	Marks the provider as OpenAI-compatible for model discovery.
CLI Command	Text	CLI only	Program to execute (e.g. gemini).
CLI Format	Dropdown	CLI only	Gemini, Claude, or Generic — determines output interpretation.

Buttons: [Test Connection] · [Save Provider] · [Reset]

Registered Providers (Accordion):

Each provider appears as a collapsible entry with: connection status, base URL, last test timestamp, and registered models with capabilities.

Actions: Edit · Retest · Toggle Composer Visibility · Delete.

---

Tab: Model

Form for manually adding a model to an existing provider.

Field	Type	Description
Select Provider	Dropdown	The provider to which the model is assigned.
Model ID	Text	Exact model identifier as used by the API (e.g. gpt-5.6-terra).
Model Label	Text	Human-readable name displayed in the interface.
Capability	Dropdown	Chat, Embedding, Text-to-Speech, Speech-to-Text, or Image.
Chat API	Dropdown	Only shown for capability Chat and an OpenAI-compatible provider. Default (Chat Completions) or Responses API.
Reasoning Effort	Dropdown	Shown for capability Chat on OpenAI-compatible or Anthropic providers. (not set), None, Low, Medium, High, X-High, Max.
Metadata (JSON)	Textarea	Optional JSON for model-specific settings (e.g. {"dimension": 1536} for embedding models). chat_api and reasoning_effort are managed by the dropdowns above and do not appear here twice.

Button: [Save Model]

Chat API / Reasoning Effort: These dropdowns are a guided interface for the metadata keys of the same name (see Manage Models) — they write to the same field as a manual JSON entry, just structured. Chat API is only shown for OpenAI-compatible providers (detected automatically via known provider ID, matching hostname, explicit metadata hint, or a local/private host). Reasoning Effort is additionally shown for Anthropic providers, where it drives the adaptive thinking depth (Extended Thinking) — tool use included.

---

Tab: Embedding

Configures which providers and models are used for vector embeddings (memory search).

Section: Primary Embedding Provider — required for all memory writes and searches.

Field	Type	Description
Provider	Dropdown	Provider with at least one model of capability Embedding.
Model	Dropdown	Specific embedding model within the chosen provider.

Section: Secondary Embedding Provider — optional fallback (e.g. a local model).

Field	Type	Description
Provider	Dropdown	Fallback provider.
Model	Dropdown	Fallback model.

Section: Mode & Fallback

Field	Type	Options	Description
Mode	Dropdown	Cloud (primary only), Local (secondary only), Hybrid (primary + fallback)	Defines which providers are active.
On Rate Limit (429)	Dropdown	Retry, Use Local	Action when the primary provider returns HTTP 429.
On Server Error (5xx)	Dropdown	Retry, Use Local	Action when the primary provider returns HTTP 5xx.

Has its own [Save Embedding Configuration] button. Changes take effect after the next server restart.