Mixing providers per agent

--- title: "Mixing providers per agent" description: "Give the Planner a strong reasoner and the Tester a cheap model — per-agent provider configuration in niki.toml." pubDate: "2026-09-01" updated: "2026-09-06" difficulty: "intermediate" time: "~10 min" ---

Each pipeline stage is independently configurable. The pattern most teams converge on: strong reasoners where judgment matters, cheap models where volume matters.

The config

[agents.planner]
provider = "anthropic"
model    = "claude-sonnet-4-20250514"

[agents.coder]
provider = "anthropic"
model    = "claude-sonnet-4-20250514"

[agents.tester]
provider = "openai"
model    = "gpt-4o-mini"     # test generation tolerates a cheaper model

[agents.reviewer]
provider = "anthropic"
model    = "claude-sonnet-4-20250514"

Keys

Set per-provider keys via environment — env vars override niki.toml, so secrets never touch the repo:

export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...

Check and tune

niki providers    # verify every configured provider responds
niki recommend    # per-agent suggestions from your own observed spend

niki recommend reads your run history — it suggests models based on what actually worked and what it actually cost, not marketing.