The Routing Layer for AI Agents

Run More Agents.
Spend the Same.

A drop-in router that picks the cheapest model that clears your quality bar.

Book a CallRead the Docs
agent.pyThe Only Change
from openai import OpenAIclient = OpenAI(    base_url="https://routing.dari.dev/{router_id}",    api_key=os.environ["DARI_API_KEY"],)reply = client.chat.completions.create(    model="dari/routing",  # Dari picks the model    messages=[{"role": "user", "content": "..."}],)
Router LogSpend $0.0000

Your Router. Fully Customizable.Describe Your Router, Use the Finetuned SLM Router, or Use the Heuristic Router.

name: Support Agentenabled_models:  - openai/gpt-5.5  - anthropic/claude-fable-5  - fireworks/deepseek-ai/DeepSeek-V4-Flashprovider_key_sources:  openai: managed  anthropic: managed  fireworks: managedrouting_strategy: customcustom_config:  rules:    - when: code review or refactoring      use: openai/gpt-5.5    - when: customer-facing reply, high stakes      use: anthropic/claude-fable-5  default: fireworks/deepseek-ai/DeepSeek-V4-Flash

Same Endpoint. Any Client.OpenAI SDKs, Your Agents, or Codex.

completions.pyPython
import osfrom openai import OpenAIclient = OpenAI(    base_url="https://routing.dari.dev/{router_id}",    api_key=os.environ["DARI_API_KEY"],)reply = client.chat.completions.create(    model="dari/routing",    messages=[{"role": "user", "content": "..."}],)print(reply.choices[0].message.content)

Frequently Asked Questions

What happens to prompt caching when you switch models?

Switching models mid-session normally wrecks cache accounting. Dari tracks which of the existing context is cache-eligible on the new model and counts everything after the switch as fresh input. When a provider reports cached-token usage we use its numbers; otherwise we estimate, so usage and cost stay accurate.

Do I get a custom router model?

Three options: deterministic weights over benchmarks and price, your own routing rules, or Dari's SLM router trained on coding agent traces. Enterprises can also train their own SLM for their traffic and use case. Book a call for custom models.

Can Dari handle full agent deployment?

Yes. We run hosted agent infrastructure behind the same control plane. Book a call to walk through your deployment.

Bring Real Traffic. Leave With a Policy.

A 30-minute router audit against your actual workload.

Book a CallRead the Docs