--- title: Spirit Room FAQ · Pilots, pricing, MCP and agent testing description: Answers about Spirit Room: early-access pilots, pricing, zero-code wrapping of APIs, CLIs and MCP servers, model × harness testing, scoring, docs interception and data handling. canonical: https://spiritroom.app/faq/ source: https://spiritroom.app/faq/ status: early-access (not generally available) generated: 2026-10-14 --- > Markdown version of https://spiritroom.app/faq/ for agents. Product screens are illustrative (Example data). Agent instructions: /for-agents.md · Index: /llms.txt · Contact API: POST /api/contact [Home](/) / FAQ # Questions, *answered.* Early access. No guarantees, just evidence. Can't find your question? [Ask us](/#contact). ## The basics ### What is Spirit Room? An early-access service that trains AI agents on your real API, CLI or MCP server. It runs agents across models and harnesses, scores every run against written requirements, rewrites your tool descriptions, docs and error messages, and re-scores until the winners are clear. ### Is Spirit Room available today? It's in early access. We run a limited number of hands-on pilots. You can book a pilot or join the early-access waitlist. ### Who is it for? Teams whose users are increasingly AI agents: MCP server authors, public API providers, developer-tool companies and teams running internal agent harnesses. ### What does "one day outside, one year inside" mean? Spirit Room compresses what would take months of real-world agent feedback into hundreds of automated rounds that run in about a day. ### Can this let us use cheaper models? Often, yes, and it is a key use case. Frontier models can work around imperfect tool surfaces; cheaper and smaller models need everything to be very clear. Spirit Room tunes your API, MCP server, CLI and harness (tool names, descriptions, schemas, errors, docs, prompts), then measures whether a cheaper model tier reaches your target pass rate and what each successful task costs. Results vary per surface; we report what we measure, with no guarantees. ## The pilot and pricing ### What do I get from a Train pilot? An Agent Readiness Audit across 3 models × 2 harnesses (up to 25 tasks), up to 10 rewrite/re-score rounds, an annotated agent-journey report, the winning variants as a reviewed patch, and a before/after evidence pack per model × harness. Early-access bonuses: an error-message rewrite pack, an agent-ready docs checklist, and one re-score after your next release (within 60 days). ### How much does it cost? The Train pilot is USD 2,500 one-time for two weeks, compute included up to a cap. Wrap (Team) starts at USD 990/month plus compute. Enterprise is custom. Early-access pricing may change. ### Is there a guarantee? No. A pilot delivers evidence: measured before/after results per model × harness, plus the journeys behind them. If your surface is already agent-ready, you get proof of that. ### What if my surface is already good? Then you get proof of that: measured results per model × harness and the journeys to back it up. ### How many pilot slots are there? Limited. Pilots are hands-on, so we take a small number at a time. ### What's the difference between Train and Wrap? Train is a pilot that exports winning changes into your own code and MCP definitions. Wrap is an always-on proxy that serves the optimized surface with zero code changes, adds runtime monitoring and journey analytics, and falls back to your original surface on errors. ## Technical ### Do I need to change my code or MCP server? No. Spirit Room wraps your surface as a proxy and overrides surface text during experiments. Winners come back as a patch you choose to merge, or are served live through Wrap. ### Which models and harnesses do you test? GPT, Claude, Gemini and open-weight models, across Claude Code, Codex, Cursor, ChatGPT and your own internal harness. You choose the matrix. ### How are runs scored? Each task has written requirements and pass/fail criteria. Every run is checked against them (scoring engine: Proof), so changes are measurable and traceable. ### What is an agent journey? A step-by-step record of what the agent read, called and guessed during a run, annotated with where it cut corners. For example, it read only the first 10 lines of --help when the key subcommand was on line 11. ### What is docs interception? An optional, opt-in feature, off by default. Spirit Room intercepts agent HTTP requests to your docs sites and serves tuned variants, so you can A/B test wording per model and harness. It only works on domains you own and allowlist. ### What are gateway mode and the local agent? Early-access ways to learn from real traffic. Gateway mode runs inside LLM apps and gateways, for example Tyk AI Studio or OpenRouter-style gateways. The local agent is an opt-in tool on developer machines that reads Codex, Claude Code or Cursor session history to find friction. ### Can Wrap serve a different surface to each model? Yes, if you turn it on. Wrap detects the model behind a session and serves the variant that scored best for it: tool names, descriptions, schemas, help, errors and docs. For specific tasks it can also merge several tool calls into one (only where you allow it) and expose scripted virtual tools composed from your existing endpoints. It is all opt-in, customer-controlled and versioned, with pass-through fallback to your original surface. ## Data and trust ### What about my data and API keys? Use sandbox credentials or bring your own keys. The local agent is opt-in and local-first: nothing leaves a machine without explicit consent. ### Are the numbers on the website real? No. Every product screen on the site is an illustrative mock labeled "Example data". ### Who are your design partners? Our early design partners are Proof and Probe (ProbeLabs). ### Can an AI agent contact you on my behalf? Yes, if you ask it to. See /for-agents.md for the JSON contact endpoint and the required fields. [Book a pilot ](/#contact)[Join early access ](/#contact)[For agents ](/for-agents.md)