Smart routing configuration

Once you have two or more providers connected, you can tell distro which provider to use for which operation. This saves money and latency without sacrificing quality.

Operations

• Voice analysis — cheap model, runs once per sample add. Default: Haiku / gpt-4o-mini / Gemini Flash.
• Idea evaluation — fast model, runs on every evaluation. Default: Haiku / gpt-4o-mini.
• Hook workshop — creative, 6 variants in parallel. Default: Sonnet / gpt-4o / Gemini Pro.
• Draft / repurpose — best quality. Default: Sonnet / gpt-4o / Claude Opus.
• Post-mortem — analytical, lots of context. Default: Sonnet / gpt-4o.
• Marketing plan generation — highest quality, runs once per app. Default: Opus / gpt-4o.

Configuring

Settings → Providers → Routing. For each operation, pick a provider from your connected list and a model. Fallbacks can be configured: if the primary fails, try the secondary.

Recommended setup

Voice analysis       → OpenAI gpt-4o-mini
Idea evaluation      → OpenAI gpt-4o-mini
Hook workshop        → Anthropic Claude Sonnet
Draft / repurpose    → Anthropic Claude Sonnet
Post-mortem          → Anthropic Claude Sonnet
Marketing plan       → Anthropic Claude Opus (or gpt-4o)

Rationale: fast/cheap for operations that run frequently, best-quality for the ones users read carefully. At a typical indie founder's usage, this costs $3–$8/month in BYOK inference.

Caching

We cache responses on content hash + voice hash with a 30-day TTL. Hit rate is 30–40% in practice, which cuts inference cost further. Cache is per-user, not cross-user.

Model updates

When providers release new models (e.g. Claude Opus 4.7 → 4.8), distro's Settings page shows the new models in dropdowns automatically. You opt in — we don't silently switch models on you.