tulip · workbench — the control runtime for agents

tulip The control runtime

Workbench

Connect a model provider

setup required

Click Provider settings to authenticate. The workbench runs against any model provider tulip supports — OpenAI or Anthropic. Keys are session-only — closing the tab discards them.

…

streamable

Prompt

Selection mode

Rule-based LLM picker

Rule-based: deterministic _rank_key ranker — auditable, reproducible. LLM picker: opt-in LLMProtocolPicker — the model picks among filtered candidates and emits a rationale.

…

Edit the source. Run executes a Python subprocess against your configured provider; events stream into the Output pane on the right.

Output

tulip · the control runtime for agents you can let act · Apache-2.0

Provider settings

Held in this tab's memory only. Never persisted to localStorage and never sent anywhere except the local BFF — closing the tab discards everything.

Provider

API key

Base URL optional, any OpenAI-compatible endpoint (vLLM, Together, LiteLLM)

Model A

Model B optional · falls back to A

Model C optional · falls back to A