Register models and choose what each feature uses
Register each model from the superadmin dashboard after you install Kindo, then choose which models each feature uses. AI models covers preparing the endpoints and the facts each registration needs.
Decide which models to register
Section titled “Decide which models to register”Absolute minimum set
Section titled “Absolute minimum set”| Role | Purpose | Recommendations |
|---|---|---|
| Embedding | Semantic search, retrieval, and indexing | Open source: BAAI/bge-m3. Hosted: Gemini embedding |
| Chat | Complex reasoning, agent runs, long-form generation | Open source: gpt-oss 120B. Hosted: Gemini 2.5 Pro, Claude 4.5, GPT 5.2 |
| Transcription | Audio file processing and voice notes | Open source: Faster Whisper |
Recommended additional models
Section titled “Recommended additional models”- Small model (title generation, quick summaries, low-latency steps): Gemma 3 or Llama 3.2 7B
- Security-focused: DeepHat V3
- Multimodal: Gemini 2.5 Pro (image + text)
Register a model
Section titled “Register a model”- Open the superadmin dashboard’s Models page and select the Models tab.
- Use the Required models panel to add an embedding or transcription model with its preset. To register another model, select Add model.
- Choose the provider.
- Enter its credentials.
- Review the model settings, then save.
- Open the Feature flags tab on the Models page to choose which models each feature uses.
Register a model for each required role. Also register at least one chat model with tool-calling support, for users to select in chat and for agents to call tools. If any of these are missing, chat, agents, ingestion, or indexing fail.
| Role | Serving name | Used for |
|---|---|---|
| Embedding | embedding-generator | Vector search and retrieval |
| Transcription | transcription-generator | Audio and voice features |
The Required models panel checks these roles, and its presets fill in the serving names. Neither role needs a feature assignment.
Choose the models each feature uses
Section titled “Choose the models each feature uses”Each feature flag below lists one or more global model IDs.
Required features
Section titled “Required features”| Feature flag | Description |
|---|---|
CRON_EXPRESSION_GENERATION | Generates cron expressions from natural language schedule descriptions |
DEFAULT_WORKFLOW_STEP_MODEL | Default model for agent steps |
INGESTION_WORKERS | Models for data ingestion and extraction |
INTERNAL_AUTO_GENERATION | Model for automatic content generation and background tasks |
LOCAL_STORAGE_CHAT_ACTIONS_MODELS | Fallback default chat model for the model dropdown and chat action selection. Applies when an organization admin has not chosen a default chat model in Settings → Models, or the chosen model is unavailable |
TOOL_CALLING_MODELS | Models that support tool calling, used to filter the models offered in chat |
Optional features
Section titled “Optional features”| Feature flag | Description |
|---|---|
API_STEP_GENERATION | Models for generating API steps in agents |
DYNAMIC_API_REQUEST_PARSER | Parses dynamic API requests |
Replace a model
Section titled “Replace a model”- Register the new model.
- On the Feature flags tab of the Models page, point each feature that uses the old model at the new one.
- Delete the old model.
Models with multiple providers
Section titled “Models with multiple providers”Every model starts with one provider. You can attach more provider endpoints to it. For example, the same model can be served from two regions of your cloud provider, or a self-hosted model can sit behind two vLLM endpoints. In the Add model sheet, pick an existing model in the Model field instead of typing a new name. The endpoint you are configuring then becomes another provider for it. On instances where models with multiple providers are enabled, organization admins on the Enterprise tier can attach and reorder providers.
People pick the model, not a provider. The model’s fallback chain and session affinity decide which provider serves each request. So each limit on a model is the most restrictive across its providers:
| Control | How the model resolves it | Effect |
|---|---|---|
| Context window | Smallest across the model’s enabled providers | A 32k provider under a 200k model makes the whole model 32k. |
| Tool limit | Smallest across the model’s enabled providers | An OpenAI or Azure OpenAI endpoint accepts at most 128 tools per request, so one of them makes 128 tools the model’s effective limit. |
| DLP filters | Strictest across the model’s providers | A filter that redacts for any provider redacts for the whole model. |
| Provider enablement | Strictest across the model’s providers | A provider your organization has disabled in its security controls makes the whole model unavailable, even if its endpoint is turned off in the model’s provider list. Turning off an endpoint is separate; see Fallback chain. |
Group providers with like-for-like capacity
Section titled “Group providers with like-for-like capacity”Pick providers that are close to each other in capability:
- Match context windows. A 32k endpoint cuts a 200k model to 32k. Give a much smaller provider its own model, so people can choose it deliberately.
- Match tool support. If tool-heavy agents use the model, avoid endpoints with a lower tool limit. A request with more tools than an endpoint accepts can fail whenever that endpoint serves it.
- Match governance. A disabled provider or a stricter DLP filter applies to the whole model, not just to the requests that provider serves.
When you pick a model to attach to, the Add model sheet shows the context window the new endpoint should clear. It also warns you when the endpoint would lower the model’s tool limit.
Fallback chain
Section titled “Fallback chain”The providers attached to a model form a fallback chain, ordered by priority number, lowest first, as the provider list shows in Settings → Models. Move a provider up or down to change the order. If a provider fails, Kindo tries the next one in the chain, except when the model rejected the request itself. To take a provider’s endpoint out of the chain, turn the provider off. The model’s other endpoints keep serving, and the provider’s configuration stays so you can turn it back on.
An endpoint you attach starts at the top priority, tied with the model’s existing provider, and providers with the same priority share traffic. To set a primary with a backup behind it, move a row up or down once.
Session affinity
Section titled “Session affinity”Kindo keeps each conversation on the provider that served its earlier turns where it can, so the provider’s cached prompts speed up replies and lower cost. This applies to chat and agent runs. If that provider is unhealthy, the conversation moves to another one.
To change how long a conversation stays on its provider, set this on the LiteLLM service:
| Environment variable | Default |
|---|---|
LITELLM_DEPLOYMENT_AFFINITY_TTL_SECONDS | 3600 (one hour) |
