Models and roles

A role is a job the system needs done — hold a conversation, review code, speak a sentence. A provider is who does it. The two are wired together in config/roles.yaml, so swapping the model behind any role is a config edit, not a code change: nothing above the role notices.

A role either names its own model or points at a shared chain — a list written once and used by several roles, so moving all of them is one edit rather than many.

These are the shipped defaults. What you can change lives in a config file you own once TESSERACT is installed, so your copy may differ, and should, once you tune it.

Role	Ships pointing at	Falls back to	Via
audio_transcribe	local.whisper.local_whisper	—	chain_6
auditor	cli.codex.gpt6_astra	—	chain_4
channel_vision	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
chat_brain	api.openai.gpt6_luna	api.google.gemini_38_flash, api.openai.gpt61_sol	chain_1
claude_cli	cli.claude.opus_55	—	chain_3
coder	cli.claude.opus_55	—	chain_3
codex_cli	cli.codex.gpt6_astra	—	chain_4
consolidate	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
image_generator	api.google.gemini_31_flash_image	—	chain_5
observer_agent	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
omni	api.jev.jev_1_13_0	api.openai.gpt6_luna, api.google.gemini_38_flash	chain_7
outbound_writer	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
playbook_extract	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
provider_check	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
scheduled_task	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
subagents_default	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2
watchman	api.google.gemini_38_flash	api.openai.gpt6_luna	chain_2

Refs read <tier>.<provider>.<model>. The tiers are api (billed per token), cli (covered by a subscription, free at the point of use) and local (runs on your machine).