RPCS-1 Documentation
RPCS-1 measures a configured agent with five primitives — TI, SG, FT, UE, and AR — then tells you what runtime settings to change. The recommendations follow receiver laws derived in the IMM research program (see the framework: three blocks — estimate, detect, commit — of which the five primitives are the measurement coordinates). If you want a fast start, use the interactive tuner or connect the public MCP server.
Start here
Every recommendation flows through the same three steps:
- Describe the workload - task, entropy, predictability, stakes, context horizon, and commitment style.
- Receive the diagnosis - the output leads with the five-primitive profile, failure-risk score, posture, and next test.
- Ship the change - apply the runtime settings, then rerun one harder edge case.
All outputs are deterministic and research-grade until the assay battery has been validated on fresh, procedurally generated items.
Quick links
- Free tuner - run a sample workflow in under a minute
- Second brain connector - your own notes inside Claude Desktop, Cursor, VS Code, and other local MCP apps; one-click add-on
- Paid diagnostic - the written memo and sample preview
- Getting started - install the Python SDK
- MCP integration - connect the public read-only server
- Examples - support, coding, and research calls
- Translation layer - face-preserving posture
- The framework - derived receiver laws and the three blocks (IMM Paper 18)
- Matching laws - how the R-1 family becomes TI recommendations
- Five primitives - TI, SG, FT, UE, AR as measurement coordinates
- Four regimes - stable, near oscillation, near overload, near freeze
AI agent integrations
RPCS-1 is available as a public, anonymous, read-only MCP server. It requires no API key or OAuth authentication. See the MCP integration guide for the endpoint, first call, and tool details.
- OpenAPI schema - REST tool contract
- llms.txt - machine-readable product overview
Try it now
The interactive tuner requires no installation and no account. Start from support, coding, or research and get recommendations in under a minute.
Questions
What does Explicit Formula do?
It shows the different ways a piece of text can be read before you send it, so you can pick the reading you meant and send the version that lands. The check on the homepage runs in your browser and sends nothing. Try it on the homepage →
What is RPCS-1?
The measurement underneath the product. RPCS-1 describes a reader with five primitives — Temporal Integration (TI), Signal Gain (SG), Filter Threshold (FT), Update Elasticity (UE) and Ambiguity Resolution (AR) — and applies derived receiver laws to turn a profile into concrete runtime settings for an AI agent. The five primitives →
Does my text leave my computer?
The homepage check sends nothing. The browser extension sends the text being checked when you pause typing; the Loop plugin sends the panel text and, only if you opt in, small snippets from your notes. No text is stored on the server. The privacy page lists exactly what is sent, where it goes and what is kept. Privacy →
What does it cost?
Free for people. Licensed for organizations — site licenses, written privacy commitments, integration help and a written agent diagnostic. The pricing page has the details. Pricing →
How do I connect it to my AI tools?
Through the public, anonymous, read-only MCP server, which needs no API key or account, or through the Python SDK. The MCP integration guide covers the endpoint and the first call; Getting started covers the SDK install. MCP integration guide →
What are the usage limits?
The web tuner allows 10 recommendations per hour. The Python SDK allows 5 free calls per day. Run the tuner →
What evidence is behind the recommendations?
Registered tests with the pass criteria frozen before any data was collected, published together with what failed and what was withdrawn. The tests measure how AI models follow instructions; they are not studies of human readers. The research page carries each result at its grade. Research and results →