Frequently Asked Questions

Cromus is the AI department you don't have to hire. Three roles make up the department: the Specialist reviews every SKILL.md change in your CI pipeline before it merges, the Manager watches cost drift and Croms accumulation across your stack, and the CWO signs the governance report leadership needs to see. Builder ($49/month) starts you with the Specialist. Try the interactive demo to see it in action.

Cromus is built for engineering teams, founders, and operations leads who are shipping AI workflows and need governance, cost visibility, and accountability without hiring a dedicated AI department. If someone in your organization needs to answer 'what did this workflow cost, who reviewed it, and who's accountable when it misbehaves' — Cromus is the function that answers those questions.

A SKILL.md file is a structured, reusable AI skill definition. It contains metadata (name, category, version), a list of steps, model requirements, and prompt templates. Skills are the building blocks of Cromus workflows.

Skill Usage separates saved workflow references (Declared), graph-based estimates for recorded workflow runs (Estimated), and executions explicitly supplied by a runtime or trusted integration (Reported). Successful approved-skill fetches are counted separately as retrievals. Estimates are not runtime traces, and usage events store metadata rather than skill content or prompts. The dashboard preview summarizes workspaces you can access; detailed views use a selected workspace. The preview and 30-day details require sign-in and workspace access; the full page and 90-day history also require account eligibility. Read the Skill Usage guide for definitions and access details.

No. Cromus is a planning and design tool. All intelligence — scoring, optimization, and cost simulation — is deterministic and rule-based, with zero LLM inference calls. This means you can plan workflows without incurring any AI inference costs.

The cost simulator estimates token usage and model costs for each step in your workflow. It supports 96 verified active generative AI models across 13 text-generation providers and 5 simulation modes (Eco, Cost, Balanced, Quality, and Open Source). The registry also tracks TypeSafe's separate input-priced decision model, bringing the provider total to 14.

Eco mode is one of five simulation modes in the cost simulator. It prioritizes the most cost-efficient model choices for each workflow step, helping you find the cheapest way to run your AI workflows without sacrificing quality. For self-hosted alternatives, try Open Source mode. See our pricing page for details on which plans include full simulation modes.

Yes. Cromus lets you export portable skill packs as ZIP files. These contain your SKILL.md definitions, wrapper prompts for ChatGPT, Claude, and Gemini, an ops savings report, and an interoperability profile — ready to be integrated into your existing tooling, deployed to LangGraph, CrewAI, n8n, or local stacks, or shared with team members.

Yes. Cromus generates read-only share links for any workflow. Recipients can view the workflow graph, skill details, and cost estimates without needing a Cromus account or login.

Agent frameworks like LangChain or CrewAI focus on runtime orchestration — executing AI tasks in real time. Cromus focuses on the design and planning phase: converting SOPs into skills, building workflows visually, simulating costs, and optimizing before any execution happens. Cromus also works alongside these frameworks — you can target your compiled skills for specific ecosystems like LangGraph, CrewAI, or n8n so they're ready to deploy.

Yes. Since Cromus does not execute AI models or make LLM inference calls, your SOPs, skills, and workflow data stay within the platform. No data is sent to third-party LLM providers during planning and simulation.

Team collaboration features — including shared workspaces, skill libraries, and version history — are available on the Ops plan. Check our pricing page for full plan details.

Yes. Your Workspace Provider Scope controls which channels Cromus may suggest. There are two layers: (1) enable the channels you actually have set up (e.g., OpenAI API, Bedrock, OpenRouter), and (2) optionally restrict to specific ecosystem tiers — Mainstream (direct provider APIs, easy setup), Managed (cloud-vendor APIs like Bedrock/Vertex/Mistral, moderate setup), Aggregator (OpenRouter and friends, easy setup), or Infra (self-hosted open-weight models, advanced setup). Cromus only recommends savings paths through channels and tiers you've enabled, so you never see a model you can't actually run today. Each model name in the app is tagged with its tier chip and the channel it would reach through.

LLM pricing changes frequently as providers update their models. Cromus maintains a verified, admin-curated pricing registry so your cost simulations reflect current market rates. The platform covers 96 verified active generative AI models across 13 text-generation providers, plus TypeSafe's separate input-priced decision model (14 providers in the registry) — including the latest GPT-6, Claude Opus 5.5, Gemini 3.8 Live, and Grok 4.7 additions — with weekly automated checks and manual verification. For teams using self-hosted models, Open Source mode estimates costs based on compute tiers instead of API pricing.

No. Gate scoring (/api/ci/score) is a checking operation and is unlimited on every plan, forever. It is never metered and never counted against any quota. Fair-use limits only apply to the compile endpoint (/api/ci/compile), which creates new SKILL.md files from SOP documents. Builder and Founding plans include 20 compiles per day and 200 per month; Ops plans include 50 per day and 500 per month; Agency plans include 200 per day and 2,000 per month; Enterprise accounts have unlimited compiles. See the fair-use policy for full details.

The Open Workflow Ecosystem is Cromus's registry of 24 open-source AI tools across 7 categories — including agent frameworks (LangGraph, CrewAI), orchestration libraries (LangChain, LlamaIndex), local inference runners (Ollama, vLLM), workflow engines (n8n, Prefect), and more. When compiling skills or exporting workflows, you can target a specific ecosystem so your output is ready to deploy. Explore the ecosystem.