Two models think.
One answers.
The Crowe Logic CLI is the operator surface for AI orchestration with live streaming text — it runs flagship models in parallel, synthesizes the best answer, then executes across 79 tools with a multi-provider chain that falls back when a model fails. One agent, every provider.
then run crowe-logic · or as an MCP server: uvx crowe-logic-mcp
$ crowe-logic run "design a retry policy for the webhook consumer" › model A exponential backoff + jitter, cap at 5, dead-letter after › model B idempotency keys first, then bounded retry with circuit breaker ◆ synthesis › idempotency key + backoff-with-jitter (max 5) + breaker + DLQ. ✓ wrote consumer.py · 12.4k tok · cache 38% · $0.03
what you get
A universal agent, not a wrapper
AI orchestration
Run two flagship models in parallel, keep the route visible, and synthesize the best response through a third instead of trusting one model's guess.
Streaming text
Answer tokens, tool feedback, and synthesis all stream live so the operator can watch the response form instead of waiting on a hidden batch.
79 tools
Filesystem, shell, git, browser automation, iTerm2/OS control, quantum, music, vision, training pipelines — in a sandbox.
Every provider, with fallback
Azure OpenAI, Anthropic, NVIDIA NIM, self-hosted, Ollama — routed intelligently, falling back automatically on failure.
Server mode
Run as an MCP server (uvx crowe-logic-mcp) for Cursor, Claude, or any MCP client; reach 5,800+ MCP servers on demand.
Per-turn transparency
Every turn shows tokens, cache hits, USD, and credits — no surprise bills.
Specialized roles
Code, cultivation, quantum, music, research — each with its own system prompt and tool subset.
pricing
Bring your keys, or let us run it.
Launch promo: 50% off the first month with code FOUNDRY50.