Open Intelligence Router for LLM Agents

README

Optimize your agent runtime to perform cheaper, safer, and better. Meet Runtime Routing Intelligence

npx auth init
DASHBOARD
1.2M
Requests routed
3,400+
Agents connected
$2.4M
Saved this month

Model distribution

Sonnet55%
Haiku20%
Opus15%
GPT-4o10%

7-day cost trend

Recent routing decisions

2s agogpt-4sonnet
cost$0.04
5s agoopushaiku
low complexity$0.02
12s agogpt-4opus
escalation
18s agosonnethaiku
tool output$0.01
31s agogpt-4sonnet
code gen$0.03
CONFIGURATION
yaml
providers:
  - name: anthropic
    api_key: ${ANTHROPIC_API_KEY}
  - name: openai
    api_key: ${OPENAI_API_KEY}

routing:
  strategy: cost-optimized
  rules:
    - match: { complexity: high }
      model: claude-opus-4
    - match: { task: code-gen }
      model: claude-sonnet-4
    - fallback: claude-haiku

cache:
  enabled: true
  ttl: 3600

Two files. Full routing intelligence.

FEATURES

01 MULTI-PROVIDER

Multi-Provider

Route across OpenAI, Anthropic, Google, Mistral, and more through a single endpoint. Add providers in config — no code changes.

02 SMART ROUTING

Smart Routing

Configurable routing tables map model names to providers. Cost and performance optimization with automatic fallbacks and escalation. Sub-10ms overhead.

03 COST OPTIMIZATION

Cost Optimization

Automatically route to the cheapest capable model. Track savings per session, per agent, per team.

04 FALLBACK & RETRY

Fallback & Retry

Automatic retries with exponential backoff. Cross-provider failover when a provider goes down.

05 USAGE ANALYTICS

Usage Analytics

Per-agent, per-model usage dashboards. Token counts, latency percentiles, cost breakdowns.

06 AGENT SDK

Agent SDK

First-class TypeScript and Python SDKs. KYA identity, payment delegation, and skill discovery built in.

07 RATE LIMITING

Rate Limiting

Per-agent and per-model rate limits. Quota management across teams with automatic queuing.

08 CACHING

Semantic Caching

Cache identical and semantically similar requests. Reduce costs and latency for repeated patterns.

09 SECURITY

Security & Guardrails

Content filtering, PII redaction, and audit logging. Policy-as-code for agent behavior boundaries.

ECOSYSTEM

LLM Providers

OpenAIOpenAI
AnthropicAnthropic
GoogleGoogle
MistralMistral
DeepSeekDeepSeek
MetaMeta
GroqGroq
CohereCohere

Agent Frameworks

AntigravityClaude Code
OpenClawOpenClaw
opencodeOpenCode
CursorCursor

Infrastructure

AWSAWS
AzureAzure
GoogleCloudGoogle Cloud
CloudflareCloudflare
OllamaOllama