# Rote

> Rote is an Apache-2.0 open-source CLI that compiles proven AI skills into typed, durable workflows. It moves fixed logic and tool orchestration into code and retains models only where judgment is required.

## Canonical resources

- [Product overview](https://roteskills.com/)
- [Token cost](https://roteskills.com/reduce-ai-agent-token-costs)
- [Speed](https://roteskills.com/agent-workflow-latency)
- [Determinism](https://roteskills.com/deterministic-ai-agent-workflows)
- [Skills to code](https://roteskills.com/skills-to-code)
- [MCP token cost](https://roteskills.com/mcp-token-cost)
- [Benchmarks](https://roteskills.com/benchmarks)
- [Skill economics](https://roteskills.com/claude-skills-token-cost)
- [Cost per employee](https://roteskills.com/ai-agent-cost-per-employee)
- [Agent or workflow](https://roteskills.com/when-to-replace-an-agent-with-a-workflow)
- [Durable execution](https://roteskills.com/durable-execution-for-llm-pipelines)
- [Compare tools](https://roteskills.com/ai-workflow-tools-comparison)
- [Open-source repository](https://github.com/trevhud/rote)

## Product facts

- License: Apache-2.0
- Package: rote-cli on PyPI; executable: rote
- Input: Anthropic-style SKILL.md plus references
- Outputs: validated pipeline.yaml, extracted deterministic modules, typed LLM-judge boundaries, and durable-runtime code
- Runtimes: DBOS, Temporal, Cloudflare Workflows, plain Python, DBOS TypeScript, and Inngest
- Node kinds: pure_function, external_call, llm_judge, agent_loop, hitl_gate
- MCP: compiled workflows can retain MCP-backed nodes and can be exposed to MCP clients as typed tools

## Important qualification

Rote makes the deterministic parts of a workflow explicit. LLM judges, agent loops, external systems, and human decisions can still vary. Savings depend on the workflow. Static scorecards are estimates; rote eval --run performs empirical evaluation with real tools.