Dari

Route AI agent requests across language models by cost, latency, and quality

Pricing model

Paid

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use Dari to automatically route prompts and agent requests across different large language models based on cost, latency, quality, and your own policies. It acts as an OpenAI-compatible proxy layer that evaluates task context, benchmark data, and cache warmth to choose the best model for each turn. Dari runs speculative evaluations in parallel to switch between smaller and larger models mid-loop without creating latency bottlenecks. This helps reduce API spending while maintaining task accuracy for coding agents and automated workflows.

What Dari does

Route API calls based on latency, cost, and accuracy thresholdsSwitch models mid-agent run based on reasoning difficultyConnect proxy routing to CLI coding tools with a single commandCalculate projected token costs including prompt cache statusTrack API savings across autonomous agent loopsDeploy an open-weight router model on local hardwareSpeculative routing that runs in parallel with models to avoid latency bottlenecksDrop-in OpenAI-compatible proxy endpoint for existing clientsMid-loop model switching between small and large language modelsContext-aware routing decisions based on task history and benchmarksCache-aware cost projections accounting for token cache warmthOpen-weight router model available for self-hosted deploymentCLI installation for instant use with terminal coding agents