Agent Chain Generation is the compiler step that turns a natural-language strategy description into an executable agent DAG. Instead of writing code, users describe what they want — "sell weekly iron condors on SPY when IV rank is above 40" — and the generator produces a directed acyclic graph of tool calls the Meta-Agent Trading Stack can execute.

    Platform

    Agent Chain Generation

    Agent Chain Generation is the compiler step that turns a natural-language strategy description into an executable agent DAG. Instead of writing code, users describe what they want — "sell weekly iron condors on SPY when IV rank is above 40" — and the generator produces a directed acyclic graph of tool calls the Meta-Agent Trading Stack can execute.

    Quick definition

    Treeova's prompt-based strategy builder that lets users create AI trading agents using natural language descriptions instead of code.

    From prompt to DAG

    The generator parses intent, picks an archetype (alert-only vs trading, single-leg vs multi-leg), assembles a chain of compatible tools, and validates the result against modality rules. A trading-capable chain that accidentally requests an alert-only tool is repaired or rejected before it ever runs.

    Healing and contracts

    Malformed model output is auto-healed against a strict contract: missing nodes are added, orphan tools are pruned, symbols are pinned per node so the executor can enforce them at runtime. Every healed chain is logged to chain_generation_logs so users can inspect what changed between their prompt and the final compiled DAG.

    Modality safety

    Alert-only agents physically cannot call execution tools. The generator strips trade tools from any chain whose archetype is alert-only, and the executor double-checks at dispatch. This means a misbehaving prompt cannot escalate into unwanted live orders — the safety boundary lives in code, not in the model's good behavior.

    Related terms

    ← Back to the full glossary