LLM Configuration¶

Section: [llm]

This is the one section almost every agent overrides: it tells OpenSage-ADK which model to use, what provider it lives behind, and the sampling knobs. Every agent needs at least a main model; additional profiles let you point individual subsystems (summarization, claim-flagging, and similar) at different models without changing code.

Defining a Model¶

Each profile lives under [llm.model_configs.<profile>]. The main profile is the one every agent uses for reasoning; summarize is optional and falls back to main if absent.

[llm.model_configs.main]
model_name = "anthropic/claude-opus-4-7"
temperature = 0.7
max_tokens = 4096

Multiple Profiles¶

Use a cheaper or smaller model for summarization than for reasoning:

[llm.model_configs.main]
model_name = "openai/gpt-5.4"
temperature = 0.7
max_tokens = 4096
rpm = 60
tpm = 60000

[llm.model_configs.summarize]
model_name = "openai/gpt-4o"
temperature = 0.3
max_tokens = 2048
rpm = 30
tpm = 30000

See the [llm] field reference for the full list of per-profile fields and accepted profile names.