AI

The platform includes a flexible, provider-agnostic AI layer that works out of the box. Instead of locking you into a single vendor, the platform is built around a few core concepts:

• LLM Providers - connections to language model backends. Every workspace already includes a managed default provider with over 450 models on prepaid credits - or connect your own backends (OpenAI, Anthropic, Gemini, DeepSeek, AWS Bedrock, OpenRouter, or any OpenAI-compatible endpoint).
• Models - the browsable catalog of every model your providers expose, with pricing, context windows, and capability badges.
• Agents - named, reusable configurations built on top of a provider. An agent pairs a specific model with custom instructions, capability flags, and an optional knowledge scope, so you can shape how it behaves without touching code.
• Knowledge Collections - named RAG scopes that group files, folders, resources, and items. Attach them to agents as defaults or to individual chats as extra context.
• Chats - streaming conversations with your agents. You can attach images and text files, dictate with your voice, pin extra knowledge collections per chat, and chat is context-aware: sessions opened from a Resource Item know about that item, and every Resource Item's data is indexed for natural-language retrieval.
• Tools & Connectors - give agents superpowers beyond text generation. Built-in tools let agents search your data, generate charts, create resource items, run workflows, and schedule recurring tasks. Connectors (MCP servers) bring in external tools from anywhere.

Once you've got the default provider (already present) and one agent, you're ready to start chatting.

How it fits togetherCopy Anchor Link

1. Pick a model - the default provider is ready immediately; browse it under Settings → AI → Models. Only add your own provider if you need a backend we don't already cover. See LLM Providers.
2. Create an agent - Pick a provider, choose one of its available models, and give the agent a name, description, and instructions. Optionally scope it with knowledge collections. See Creating an Agent.
3. Start chatting - Every chat session begins with a chosen agent, and you can switch agents mid-conversation without losing context. Attach extra collections per chat when needed. See Chatting with Agents.
4. (Optional) Extend capabilities - Enable built-in tools or add external tool servers via connectors to let agents act on your behalf.

Why providers and agents are separateCopy Anchor Link

Separating providers from agents keeps your setup flexible:

• You can register multiple providers for the same vendor - for example, two separate OpenAI providers, each with its own API key, to isolate billing or usage tracking on OpenAI's own dashboard.
• You can create multiple agents on the same model with different instructions - for example, a "Support Assistant" and an "Internal Docs Assistant" both running on the same model, but tuned for very different tasks.
• You can scope multiple agents to different knowledge - for example, one agent searching everything and another restricted to a single project's collection.

Where AI shows up in the productCopy Anchor Link

Beyond the dedicated chat experience, agents and models power several features across the platform:

• Semantic search - Resource Items and their attached documents are automatically embedded into a vector index, so both the global search page and chat agents can find records by meaning rather than exact keywords. Knowledge collections scope that retrieval per agent and per chat. See Resources.
• Workflow automation - The Agent Call node lets any workflow reason over unstructured data, extract structured output, and make dynamic routing decisions. See the Agent Call node.
• Form generation - Workflows can ask an agent to generate dynamic form schemas for human-in-the-loop tasks. See User Task nodes.