Generate¶

The Generate node runs language-model inference. It is the workhorse of every Tryll workflow: it takes the current dialog, passes it through the agent's projection to build a prompt, and emits the model's response — either as a single chunk or streamed token-by-token.

NodeType: Generate.

Parameters¶

All parameters are optional. Each is a string key/value pair on NodeDescription.params.

Generation behaviour¶

The Mutable column indicates whether the parameter can be changed at runtime via ChangeParam.

Key	Default	Mutable	Description
`model_name`	(agent default)	No	Catalog name of the language model to run. Falls back to the graph's `default_model_name`. At least one of the two must be set.
`system_prompt`	`""`	Yes	Text injected at the head of the prompt by the default projection strategy. The KV cache is rewound lazily on the next `SendMessage` call.
`stream`	`"false"`	Yes	`"true"` to emit tokens one-by-one as `AnswerText` chunks. `"false"` to emit a single `AnswerText` with the full response and `is_final = true`.

Sampling overrides¶

Each field overrides the corresponding value from the model's default_sampling (see model management). All fields are optional strings; parse as the type shown. All sampling parameters are mutable at runtime.

Key	Type	Description
`temperature`	float	Sampling temperature.
`top_p`	float	Nucleus sampling threshold.
`top_k`	int	Top-K cut-off.
`min_p`	float	Min-P filter.
`max_tokens`	int	Generated-token cap for this node.
`seed`	uint32	RNG seed (`0` = random per turn).
`repeat_penalty`	float	Penalty for repeating tokens.
`presence_penalty`	float	OpenAI-style presence penalty.
`frequency_penalty`	float	OpenAI-style frequency penalty.

Exit routes¶

Route	Fires when
`default`	Always — Generate never branches. Wire this to the next node, or to `"END"` to terminate the turn.

Side effects¶

Appends the generated answer to the current turn as the assistant's reply.
Emits one or more AnswerText frames to the client.
Updates the per-model KV cache.

Diagnostics¶

When the agent has enable_diagnostics = true, the node's contribution to TurnComplete.debug_info includes the projected prompt and the model name that was actually run (after the model_name / default_model_name fallback).

Minimum working example¶

PythonC++Unreal

from tryll_client import GraphDescription, NodeType

graph = (
    GraphDescription()
    .add_node("answer", NodeType.Generate, {
        "stream":        "true",
        "system_prompt": "You are a terse assistant.",
    })
    .wire("answer", "default", "END")
    .set_start_node("answer")
    .set_default_model_name("My Local Model")
)

agent = client.create_agent(graph)

namespace TC = Tryll::Client;

TC::GraphDescription graph;
graph.AddNode("answer", TC::NodeType::Generate, {
        {"stream",        "true"},
        {"system_prompt", "You are a terse assistant."},
     })
     .Wire("answer", "default", "END")
     .SetStartNode("answer")
     .SetDefaultModelName("My Local Model");

auto agent = client.CreateAgent(graph);

FTryllGraphDescription Graph = FTryllGraphBuilder()
    .AddNode(TEXT("answer"), ETryllNodeType::Generate, {
        {TEXT("stream"),        TEXT("true")},
        {TEXT("system_prompt"), TEXT("You are a terse assistant.")},
    })
    .Wire(TEXT("answer"), TEXT("default"), TEXT("END"))
    .SetStartNode(TEXT("answer"))
    .SetDefaultModelName(TEXT("My Local Model"))
    .Build();

Or author the same nodes inside a UTryllWorkflowAsset in the Content Browser and assign it to UTryllAgentComponent.

Client bindings¶

C++: Tryll::Client::GraphDescription::AddNode(name, NodeType::Generate, params) — GraphDescription.h
Python: tryll_client.GraphDescription.add_node(name, NodeType.Generate, params) — graph.py
Unreal: add an FTryllNodeDesc with Type = Generate to FTryllGraphDescription.Nodes — TryllGraphDescription.h