One base-URL change. Any harness.

Open-source harness users pay list price on their own provider keys. Point the same harness at Flow AI and every request routes to the cheapest model that completes the task — subscription, free-tier, and marketplace supply included — with per-agent cost analytics in Cortex. Exact configs below; all three agent protocols (OpenAI, Anthropic, Responses) are served with full tool-calling parity.

Buzz (Block) — agent teammates in shared rooms

Buzz runs its room agents through ACP harnesses — Goose, Codex, and Claude Code. Configure each agent's harness environment with Flow and one org key routes your whole deployment; tag each agent so Cortex reports cost-per-completed-task per teammate, extending Buzz's audit-everything model to inference spend.

export ANTHROPIC_BASE_URL="https://api.flowaiapi.com"
export ANTHROPIC_AUTH_TOKEN="fa-..."          # one org key for the whole hive
export ANTHROPIC_CUSTOM_HEADERS="x-flowai-app: buzz"   # per-agent Cortex analytics

model = "auto"
model_provider = "flowai"

[model_providers.flowai]
name = "Flow AI"
base_url = "https://api.flowaiapi.com/v1"
wire_api = "responses"
env_key = "FLOWAI_KEY"

deepseek-harness (dsh)

dsh is plugin-everything; any OpenAI-compatible provider works. Point it at Flow and you keep DeepSeek's models (we serve them) while auto-routing to whatever finishes each task cheapest — often a canary-verified free model.

llm-pi-ai:
  providers:
    flow:
      apiKeyEnv: FLOWAI_API_KEY
      api: openai-completions
      baseURL: https://api.flowaiapi.com/v1
      models:
        - id: flow-1            # balanced
        - id: flow-1-savings    # max savings
        - id: flow-1-quality    # quality first

Full guide with a recorded session (14 tool calls, source read, MCP protocol executed, $0.0042): Flow AI in DeepSeek Harness.

Aider

export OPENAI_API_BASE="https://api.flowaiapi.com/v1"
export OPENAI_API_KEY="fa-..."
aider --model openai/flow-1

OpenCode

{
  "provider": {
    "flowai": {
      "npm": "@ai-sdk/openai-compatible",
      "options": { "baseURL": "https://api.flowaiapi.com/v1", "apiKey": "{env:FLOWAI_KEY}" },
      "models": { "flow-1": { "name": "Flow AI auto-routing" } }
    }
  }
}

Cline / Goose / anything OpenAI-compatible

Choose the "OpenAI compatible" provider type, set base URL https://api.flowaiapi.com/v1, key fa-…, model flow-1. Goose can also run Anthropic-style using the Buzz block above. Every response carries the routing trace (_flowaiapi): model used, exact cost, latency, and the reason it was picked.