- What it does
- How much can agents safely do for you today? Five questions, an honest level and a first job to hand over now.
- Who it is for
- People.
- Start in one step
- Use it on this page, or ask your agent to call
autonomy_questionsat llpro.ai/mcp. - What to expect
- Live: open today. Free: campus limits per caller. No sign-up.
New to this? Start here.
What it does
Autonomy Check asks five questions about what you, or your business, actually let AI tools do today: what you hand off, whether tools have limits on what they may spend, send or sign, how you check what a tool did, where the records are kept, and which jobs take your time. You get a stated level on the LLPRO autonomy ladder, the one thing that stands between you and the next level, and always a first job: the first job you picked that a live campus tool can take now, with the exact call for your agent. The scoring rule, word for word: Autonomy Check rule v1 (15 Nov 2026). Five questions: what you let AI tools do today (use: 0 = nothing yet, 1-4 = the four public levels of the LLPRO autonomy ladder, in its own words), whether tools have limits on what they may spend, send or sign (limits 0-3), how you check what a tool did (records 0-3), where the records the work needs are kept (data 0-3), and which jobs take your time. Stated level = the lower of 'use' and what your guardrails can hold. Guardrails always hold level 1. Level 2 needs limits written down for each tool (limits 2 or more). Level 3 also needs a log or record for every action (records 2 or more). Level 4 needs limits the tool cannot go past (limits 3), records someone else can check (records 3) and the records in one place with an export or a connection (data 3). A stated level comes from answers only and is never proof. A proven level is counted only from signed records of finished jobs; that count is not open yet, so the proven level shown is 0. The first job is the first job you picked that a live campus tool can take now without changing anything (a read-only tool); if none of your picks can, a starter job. Tools that can change things are not handed over until the campus consent check is on. Nothing runs until you, or your agent with your yes, send the call.
- Who it is for
- Anyone who uses AI tools, or wants to, and wants to know how much they can safely hand over today; and agents that run the check for the person or business they work for.
- Price
- Free, no sign-up, nothing to pay or refund. Campus limits apply: 40 calls a minute and 1,500 a day per caller. Run by LLPRO LLC; help at llpro.ai/support.
- Limits
- It scores answers, not behaviour: a stated level is what your answers support and is never proof. It hands over only jobs a live campus tool can take without changing anything; tools that can change things wait for the campus consent check. It never runs a job itself: you, or your agent with your yes, send the call. It knows 10 kinds of personal jobs and 13 kinds of business jobs; a job with no tool yet is noted as demand. Nothing you answer is kept: the check scores your answers and forgets them. Like every call to the campus, the call itself is logged (time, tool, a hashed caller key, how it arrived). The check also adds one to a daily count of finished checks, of which next-step card they showed and of which tool took the first job; that count holds no answers and no caller. Your stated level is never shown as proven and is never published.
Take the check
For developers: try it
POST https://llpro.ai/mcp?via=autonomy-check
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"autonomy_check","arguments":{"kind":"business","answers":{"use":1,"limits":0,"receipts":1,"data":2,"tasks":["underpay","books"]}}}}
-> stated (name, says, guardrails_allow, is_proof false), proven (0 until counting opens), tasks, first_job (tool, page, exact call tagged via=autonomy-check, receipt line only if the tool signs), next_card, privacy. Send only kind first to get the questions.For agents
Tool autonomy_check on https://llpro.ai/mcp and in the tools list. It signs nothing: a stated level is not a record of work. Call it with only kind to get the five questions, then again with the answers. Every answer carries the scoring rule's version and sha256 (rule_sha256); the rule itself comes back with the questions.
For agents: this page as Markdown · JSON · llms.txt · MCP endpoint llpro.ai/mcp · agent card