Models

Models

agent-notes uses a role-based model system. Each role maps to a model tier, and agents inherit their model from their assigned role.

Roles

Role	Description	Default Model	Color	Volume
Orchestrator	Plans, delegates, reviews multi-step tasks	Opus	Purple	Low
Reasoner	Deep debugging, architecture analysis	Opus	Red	Low
Worker	Implements code, writes tests, reviews	Sonnet	Green	Medium
Scout	Fast file discovery, pattern search, docs	Haiku	Cyan	High

Available Models

Opus tier (highest capability, highest cost)

Model ID	Notes
claude-opus-4-7	Latest, most capable
claude-opus-4-6	Current default for reasoner/orchestrator
claude-opus-4-1	Legacy

Sonnet tier (balanced capability and cost)

Model ID	Notes
claude-sonnet-4-6	Current default for worker
claude-sonnet-4	Previous generation

Haiku tier (fastest, lowest cost)

Model ID	Notes
claude-haiku-4-5	Default for scout

Configuring Models

Via CLI

agent-notes set role reasoner claude-opus-4-7
agent-notes set role worker claude-sonnet-4-6
agent-notes set role scout claude-haiku-4-5
agent-notes config show    # verify current configuration

Via wizard

agent-notes config         # interactive wizard

Per-CLI configuration

agent-notes set role worker claude-sonnet-4-6 --cli claude
agent-notes set role worker claude-sonnet-4-6 --cli opencode

Cost Implications

The role system is designed to minimize cost while maintaining quality:

• Opus ($15/M input, $75/M output) — reserved for orchestration and deep reasoning. Low volume.
• Sonnet ($3/M input, $15/M output) — the workhorse. Most code writing and analysis.
• Haiku ($0.25/M input, $1.25/M output) — high-volume exploration. Reads many files cheaply.

See Cost-Report for per-session cost tracking.

Model Capabilities

All models support:

• Tool use (function calling)
• Vision (image analysis)
• Long context

Model selection should be based on reasoning requirements, not context length.