Build

The agent

The agent runs a tool loop over the live workspace: list_nodes, read_node, write_node, delete_node, list_files / read_file / write_file / delete_file (shared helpers under lib/), list_views / read_view / write_view / write_view_file / delete_view_file / delete_view (see "Views"), run_node, define_param, and query_data (DuckDB over every node's current output). It looks at real column names before writing, runs what it writes, reads the traceback when something fails, and fixes it. Every tool call shows up in the chat, and nodes it touches flash on the canvas.

Pick the model with environment variables; the provider is auto-detected from whichever key is set (Claude wins if both are):

ProviderKeyDefault model
anthropicANTHROPIC_API_KEYclaude-opus-5-5 (adaptive thinking, server-side fallback)
openrouterOPENROUTER_API_KEY (or OPEN_ROUTER_KEY)deepseek/deepseek-v4-pro
openaiOPENAI_API_KEY + EXUVIA_BASE_URL + EXUVIA_MODELnone (must be set)
# DeepSeek through OpenRouter
OPENROUTER_API_KEY=sk-or-... exuvia serve my-project
# any other OpenRouter model
EXUVIA_PROVIDER=openrouter EXUVIA_MODEL=qwen/qwen3-coder exuvia serve my-project
# a local model
EXUVIA_PROVIDER=openai EXUVIA_BASE_URL=http://localhost:11434/v1 EXUVIA_MODEL=qwen3:32b OPENAI_API_KEY=x exuvia serve .

The model has to support function calling. The UI's agent header shows which model is active.

The same model from a node. A Python node can call the configured model once, for text, with exuvia.llm.complete. It uses the provider the agent uses and raises LLMError when none is set, or when the reply is empty or cut off, so a node never stores half an answer.

from exuvia import llm

@exuvia.node
def note(figures):
    return {"summary": llm.complete(figures.to_csv(index=False), system="Summarise these figures in two sentences.")}

Project notes. Put an AGENTS.md next to exuvia.json (or an instructions string in it) to tell the agent about this project's data sources, helpers and conventions. It is appended to the system prompt on every turn, which is how a host app teaches the agent its own data without changing exuvia.

Sessions. A conversation is stored with the project (GET /api/agent/{session_id} returns its transcript and whether a turn is running). A turn keeps running if the client disconnects, and its events are also published on /api/events, so a UI can reload and re-attach.

Things to ask it:

  • "Pull in daily prices and add a 60-day rolling correlation to SPY"
  • "Why did returns fail?" (with the node selected)
  • "Schedule prices every 30 minutes during market hours"
  • "Chart the drawdown for each ticker"