Agent Configuration

Everything that shapes how a project's agent behaves lives here.

LLM Settings

Defines which language-model provider and model power the agent, and which capabilities are switched on:

• Provider & model — the AI provider and model, plus behaviour settings like creativity (temperature) and retry behaviour.
• Capabilities — turn features on or off: tool calling, knowledge-base retrieval (RAG), and voice.
• Embeddings — the provider used for knowledge-base search (with a "same as chat" option).

Changes are staged and saved together, with a warning before leaving with unsaved edits. Saving always writes the complete configuration, so edits across tabs are preserved.

Provider Keys (BYOK)

A per-project vault of "bring your own key" AI provider credentials. Add, replace, delete, and test keys; you can also reuse a key you've already connected to another project. Keys are stored securely — only masked metadata is ever shown, never the secret.

System Prompts

The instructions that shape how the agent responds. There's one always-present System Prompt, plus optional advanced overrides for specific situations — when the agent uses the knowledge base, when it calls tools, and when it does both. A preview shows how the pieces assemble (the situation's template, the agent's personality, and any structured-output requirement). A prompt must include the required context placeholder before it can be saved.

Your Agent (personality)

Manage the AI agents (personalities) available to the project and choose which one is live. An agent's identity, character traits, and behaviour determine how it responds. You can create, edit, duplicate, and delete agents, set one active (exactly one is live per project), and clone from a saved template or the curated catalogue. With no active agent, the project uses a default personality.

Tool Settings

Project-wide settings that apply across all of the project's tools (as opposed to any individual tool): the default base URL prepended to tools defined by path only, and the authentication callback used to validate incoming /ask requests.