Configuration overview

Outcome: Know where settings live and how they merge.

Layers (later wins)

1. Built-in defaults.yaml
2. ~/.daari/config.yaml
3. Profile overlay (DAARI_PROFILE)
4. Project .daari.yaml (via X-Daari-Project)
5. Request headers
6. Environment: DAARI_<SECTION>__<KEY>

Edit safely

• YAML: ~/.daari/config.yaml
• Config editor API (when observability.config_editor: true): GET/PATCH /v1/daari/config with persist: true
• Web UI config card (daari web-ui serve) — send Bearer if API key set

Major sections

Section	Purpose
server	host/port, api_key, virtual_keys
models / ollama / mlx	Local backends
asr	Local speech-to-text; frontier_fallback defaults off (guide)
tts	Local text-to-speech (POST /v1/audio/speech); empty base_url → 501 (guide)
routing.local_pool	Multi-host pool; kind: openai for vLLM / llama.cpp / LM Studio (guide)
cache	L0/L1, disk|redis
routing	prefer, confidence, caps
frontier	L6 providers, budgets, PII
guardrails / boundaries	Safety + product scope
observability	prometheus, postgres, config_editor
enterprise	org cache, SSO, policy sync

Full table: Config reference (generated).

Next

→ Project profiles · Auth and keys · vLLM / llama.cpp local tier