Key Differences from Langflow¶

Agentlab is built on Langflow and retains full compatibility with its core features. This page summarizes what Agentlab adds or changes.

Additional Components¶

Component	Description
Context Input	Provides a configurable source selector for agents that perform context-specific lookups. The user's selection is delivered into the flow from the Karli Studio chat.
Identity Input	Returns identity information for the user running the flow, decoded from the session JWT injected by the KARLI proxy.
Memory Cell	Persistent key-value storage with three scopes (Session, User, Agent) and three operations (Read, Write, Delete). Survives across flow executions; User scope shares data across sessions for the same user.
Web Search Confirmation	Pauses the flow and asks the user, in the chat, whether the agent may perform a web search; routes execution down a confirm or deny branch.

Component	Extension
Read File	Adds an extraction backend selector (`docling` or `karli`); when `karli` is chosen, an extraction model picker appears and uploads are delegated to the KARLI data-extraction service. Also gains an "Upload Target" mode that exposes the component as a user-facing upload slot in the Karli Studio chat.
Chat Input	Adds a configurable list of chips that render as interactive controls in the Karli Studio chat. Each chip becomes a dedicated output on the component, enabling per-chip routing of the flow.

Agentlab registers two additional model providers alongside the standard set (OpenAI, Anthropic, etc.):

KARLI — the default provider, offering LLMs, embeddings, and data-extraction models.
tokeneurope — a European-hosted provider offering LLMs.

Models are surfaced across three categories:

LLMs — chat models from KARLI and tokeneurope, selectable from any model-aware component (Language Model, Agent, etc.).
Embeddings — KARLI-hosted, selectable from the standard Embedding Model component.
Data Extraction — used by the Read File component's karli extraction backend. KARLI is currently the only provider offering models in this category, which means document- and audio-extraction via a selectable model is specific to Agentlab.

See the Model Providers section for credentials, available models, and configuration details.

Agentlab integrates with the broader Karli Studio platform for:

Chat interactions — context selections, chip selections, and file uploads flow between the Karli Studio frontend and Agentlab.
Document extraction — the Read File component can delegate extraction to the KARLI extraction service.
Identity — components such as Identity Input and the KARLI extraction call rely on the JWT injected by the KARLI proxy.
Observability — flow execution is traced via Langfuse.